Abstract
A law firm sells a service that a client cannot judge in advance, under conduct rules that limit what the firm may say and how it may be recommended. Clients reach lawyers by referral, by search and now through the answer an AI assistant writes.
This paper sets out what the evidence says about each step. Asked for a family law attorney in 16 US cities, three AI assistants named 240 businesses, and 5.4% of them were named by all three while 75.8% were named by only one1. Each question was asked once, through each assistant's API1. ABA Model Rule 7.1 bars a false or misleading communication about a lawyer or the lawyer's services2, and the paper reduces that rule and its neighbors to requirements for a site, a profile and a review program.
Six chapters cover how clients find and choose a lawyer, law firm SEO and AI answers, the advertising rules, intake and the answering service, legal AI, and measurement with a first 90 days. The paper describes rules and research in general terms. It is not legal advice, and a firm's own counsel decides how any rule applies.
What this paper argues
- A client chooses a lawyer on trust, so a firm's marketing is the upkeep of facts that a client, or an assistant acting for one, can check.
- AI assistants seldom agree on which lawyer to name. For family law attorneys in 16 cities, 5.4% of the businesses named were named by all three assistants1.
- The model advertising rules are short enough to design to: no false or misleading communication2, and no payment for a recommendation outside the listed exceptions3. Each state adopts its own version.
- Intake is where a firm that was found and chosen can still lose the client, and it is where duties to a prospective client begin4.
- In our judgment the first use of AI likely to pay in a small firm is answering and intake, and any use of generative AI in legal work needs the independent verification that ABA Formal Opinion 512 describes5.
How to read this paper
- ~
- Our estimate, built from the nearest cited figures. Not a measurement.
- †
- A vendor’s own figure, from its customers or its own survey. Not an independent sample.
- NF
- Not found: we looked for a published measure and found none.
- ( )
- A figure followed by its working in brackets, or by “calc.”, is our arithmetic on cited figures.
- Rules
- We report what each rule says as read on the date shown. This is not legal advice, and counsel decides how a rule applies to you.
- Sources
- A small number points to the source in References, which says where we hold a source only through its abstract or a secondary account.
Chapter 1
Law firm marketing: how clients find and choose a lawyer
What AI assistants named when we asked for a family law attorney
In our study, the names an AI assistant gave for a family law attorney depended on which assistant answered. We put 199 local questions to ChatGPT, Gemini and Perplexity across 16 US cities and 23 kinds of business on 2026-11-10, and collected 597 answers through each assistant's API with web search on1. Each question was asked once, and answers through an API are a baseline, not what every signed-in user sees1. For family law attorneys, asked about in 16 cities, the assistants named 240 businesses (15 per question). Of those, 13 (5.4%) were named by all three assistants and 75.8% by only one1.
Two more lawyer rows are smaller, each from 3 cities. For personal injury lawyers the assistants named 46 businesses (15.3 per question); 3 (6.5%) were named by all three and 76.1% by only one1. For estate planning attorneys they named 47 businesses (15.7 per question); 1 (2.1%) was named by all three and 83% by only one1. Three cities make these small samples, not findings about personal injury or estate planning practice.
These rows describe disagreement among the assistants, not a verdict on any firm. Names were matched across assistants by spelling, and up to eight names were read per answer1. A firm written two ways can appear as two names, which would lower the overlap; we cannot say how often that happened. The study does not show that any assistant is wrong, that a named firm is good, or that a cited page caused a name to appear. The long paper's chapter on how the study was run sets out the method, and the study itself holds the data.
Family law is one of ten kinds of business asked about in all 16 cities1. For those ten kinds (160 questions) the assistants named 2,414 businesses, and 84 (3.5%) were named by all three1. The 5.4% for family law attorneys is higher, but it rests on 13 businesses, and we do not read the gap as a finding about lawyers1. A client who asks two assistants may be given different names, and the firm cannot see which conversation it is absent from.
The legal profession: how many lawyers a client chooses among
The ABA's 2025 Profile of the Legal Profession reports 1.37 million lawyers in the United States in 2025, up from 1.35 million the year before, and calls the rise the first significant increase since 20206. The ABA's news release on the same survey gives the exact count: 1,374,720 lawyers in 2025, against 1,355,963 the year before, a rise of 1.38%7. Over a longer span, the ABA's 2024 Profile found growth of roughly 1% a year since 2000, from 1,022,462 lawyers in 2000 to 1,322,649 in 2024, an increase of 29%8. The two editions give different figures for 2024: the 2024 edition counted 1,322,649 active lawyers as of January 1, 2024, while the 2025 edition gives 1.35 million for the previous year6,8.
The Bureau of Labor Statistics counts jobs in an occupation, not licensed lawyers, so its figures are not comparable with the ABA's: 863,700 jobs for lawyers in 2025, with 5 percent growth projected from 2025 to 20359. It also reports a median annual wage for lawyers of $159,670 in May 20259, which is not a fee. For scale, the Census Bureau's advance estimate of 2025 revenue of employer firms in legal services is $446,060 million, against $405,824 million in 2024, a rise of 9.9%10. It covers legal services as a whole, a wider group than offices of lawyers, and the Bureau revises advance estimates later.
The Quarterly Census of Employment and Wages counts offices, which is closer to the size of a firm. In 2024 it counted 188,520 private-sector establishments in offices of lawyers, with an annual average employment of 1,098,91811. In the first quarter of 2024 there were 187,335, of which 141,518 had fewer than 5 employees and 23,989 had 5 to 912,13,14. That is 75.5% of offices with fewer than 5 employees (141,518 / 187,335), and 88.3% with fewer than 10 (adding the 23,989 offices of 5 to 9).
Three limits apply. The count is of establishments, so a firm with five offices is five. The size is employees in the office, not lawyers in the firm. And an owner with no employees is not counted at all. We estimate that about three in four law offices have fewer than 5 employees, the scale of firm this paper has in mind (from the size classes above; likely low, because an owner with no employees is not counted). A firm-level size distribution in the ABA's profile pages or the Census Bureau's firm-size tables is NF in a form we could read; the nearest evidence is these office counts, which do not say how many lawyers practice alone.
Fees differ by kind of work. The Washington State Office of Public Defense's spring 2020 survey of private criminal defense attorneys found an average attorney cost of $3,500 for a DUI, excluding investigators and experts15. For personal injury, the Insurance Research Council reported that insurers' average bodily-injury claim payments rose from about $14,000 in 2017 to more than $20,000 in 202216. A State Bar of California advisory says a personal injury contingency fee contract will often provide for one-third, which it calls routine and commonly accepted17. One-third of $20,000 is $6,667 (calc.). That illustrates one kind of case in one state and is not a firm's revenue: the payments are insurers', and the one-third is guidance for fee arbitrators.
A survey of Michigan lawyers found a median hourly billing rate of $300 in 202318, and Clio† (legal practice-management software) reports in its 2025 data that 59% of firms billed flat fees, alone or beside an hourly rate, in 202419. An hourly rate is therefore a benchmark, not what a client pays for a matter.
Room for another client is a separate question. In Clio†'s 2025 Legal Trends Report (usage data), the average law firm utilization rate, the share of an eight-hour day put toward billable work, was 38%, so five hours of a lawyer's day go unbilled19. That is about 3.0 billed hours (38% x eight hours). Clio†'s 2023 report for solo lawyers put the 2022 average at 25%, about two hours of billable work a day, against 36% for lawyers at non-solo firms20. The two figures come from different reports and years, and neither is a sample of all US lawyers. Unbilled time is not idle time: it includes the work of running a firm, which these data do not break down. We think the average lawyer in Clio†'s data is still not short of hours, so a new client is more often a question of fit and fee than of room.
This paper does not count clients or matters, so the growth in lawyer numbers cannot say whether a client now has more firms to choose among for each matter. It shows only that a client is choosing within a large occupation, most of whose offices are small.
Why choosing a lawyer is hard: credence, licensing and trust
Darby and Karni (secondary) distinguished three types of purchase qualities: those known before purchase, those known costlessly only after purchase, and credence qualities, which are "expensive to judge even after purchase"21. Dulleck and Kerschbamer's survey of credence goods asks under which conditions experts have an incentive to exploit the informational problems of diagnosis and treatment22.
We think legal advice belongs in this category. A client who loses cannot tell whether the facts, the law, the other side or the lawyer decided it, and a client who wins cannot tell whether another lawyer would have won more. A 1995 consumer-evaluation model by Bendixen and Gault (secondary) places legal advice among the typical products of a segment with high credence qualities, in which word-of-mouth recommendation is key to decision-making23. It is a model, not a measured share. Day and Barksdale make a related point about professional services in general: because they are typically intangible and customized, "evaluation prior to purchase is difficult"24.
Licensing is one answer to that difficulty. Akerlof listed "the licensing of doctors, lawyers, and barbers" among the practices that reduce quality uncertainty by certifying a level of proficiency25. We think a license is a floor, not a ranking: it does not tell the client which of the lawyers in a city suits the matter. Stigler wrote that reputation commands a price because it "economizes on search"26. We infer that a referral, a star rating and a prompt reply are reputations in this sense: cheap signals that stand in for a judgment the client cannot make. A firm's marketing is then the work of supplying accurate signals that are easy to find. The long paper's chapter on the economics of choosing a professional argues the theory at length.
How clients find a lawyer: referral, search and reviews
The most recent figures come from FindLaw† (Thomson Reuters), which describes its 2024 U.S. Consumer Legal Needs Survey as an annual poll of 2,000 US adults who had a legal need in the past 12 months27. In it, 56% of respondents took action toward their legal need within a week or sooner28. Among respondents who learned about the attorney they contacted, 48% said they used referrals from family members, friends or trusted coworkers, the most popular offline resource27. Among those who contacted an attorney they found online, 82% used online reviews as part of the decision, and nearly 40% said reviews were their primary source of information27. These are stated answers from a poll whose sampling the summary we read does not describe. The review share is of the subgroup who found a lawyer online, not of all respondents, and a respondent who says a referral was one resource has not said it produced the hire.
Clio† ran a consumer survey of 2,000 US adults for its 2019 Legal Trends Report. Of those who had ever shopped for a lawyer, 57% said they contacted more than one law firm, and 44% of clients believed they needed to talk to more than one lawyer to find the right one29. In the same survey 42% said that if they like the first lawyer they speak with, they will not need to speak with any others29. That is a stated intention to stop looking, not a count of people who hire the first lawyer who answers.
Clio† also asked how quickly respondents expect a response to a phone message or email: 10% said within an hour, 24% within a few hours and 45% within 24 hours, so a reply after 24 hours misses the expectation of 79% on Clio†'s own sum29. These are expectations from 2019, not behavior. We think the Clio† results, read together, describe a prospective client who may contact more than one firm and may stop at the first lawyer who seems right, which is why Chapter 4 treats first contact as a stage where a firm can still lose the client.
In a consumer expectations survey summarized in 2019 by the Illinois Supreme Court Commission on Professionalism, Clio† reports that 59% of clients sought a referral to an attorney while 57% searched for one on their own30. Both shares appear in one sentence of the summary, and we do not know whether they describe the same clients, so we do not add them or infer how many did both. Clio†'s Legal Trends survey data also report that 46% of Millennials were likely to value referrals, against 60% of Boomers30. In Clio†'s 2025 consumer survey of 1,000 US adults, 70% of those who had hired a lawyer recently said they had an easy time finding and hiring one, largely thanks to referrals and online searches19. The size of that subgroup is not printed.
A September 2010 Harris poll for the ABA Standing Committee on the Delivery of Legal Services, as the ABA Journal reported it, points the same way: 46% of respondents would ask a friend, family member or colleague for a lawyer referral and 34% would contact a lawyer they knew or had used before31. These are stated intentions from 2010, and we cannot assume the shares hold today.
BrightLocal†'s Local Consumer Review Survey 2023 reports that 81% of consumers feel reviews are "important" or "very important" in the financial and legal sector, up from 66% in 202232. The sector pools financial and legal services, and the figure records how important respondents say reviews are, not what they did. BrightLocal†'s 2024 survey reports that 75% of consumers always or regularly read online reviews when researching local businesses, and 3% never do; it covers local businesses generally, not law in particular33.
These figures come from different instruments, years and populations, so they do not make a trend, and none measures how many signed matters began with a referral, a search or an AI answer. We estimate that a referral takes part in the choice of roughly half of clients or more (from the two surveys closest to the question; likely high as a share of signed matters). The inputs are the 48% of FindLaw† respondents, among those who learned about the attorney they contacted, who used referrals27, and the 59% of clients in Clio†'s survey who sought one30. The estimate runs high because using or seeking a referral is not the same as hiring through one, and the two surveys differ in year, wording and base. A share of new clients by source, taken from firms' own records in law and in the other professions, is NF; the nearest evidence is surveys of opinion or stated behavior, such as these two.
What the review studies can and cannot say about lawyers
The review and sales studies cited here concern restaurants and books, not lawyers. Luca's 2011 study, revised in 2016, used the way Yelp rounds displayed ratings and Washington State Department of Revenue data to find that a one-star increase in a Yelp rating led to a 5 to 9 percent increase in restaurant revenue34. The effect came from independent restaurants: ratings did not affect restaurants with a chain affiliation, and chain restaurants lost market share as Yelp penetration grew34. Luca concluded that online consumer reviews "substitute for more traditional forms of reputation"34.
Anderson and Magruder studied Yelp ratings and restaurant reservations and found that an extra half-star rating caused restaurants to sell out 19 percentage points more frequently, "with larger impacts when alternate information is more scarce"35. Chevalier and Mayzlin found for online book reviews that an improvement in a book's reviews led to an increase in relative sales at that retailer's site, and that for most samples the impact of one-star reviews was greater than the impact of five-star reviews36. They also found reviews overwhelmingly positive at both retailers they studied36.
We think two findings may transfer to law. Where a buyer has little other information, a rating counts for more35, and a prospective legal client with a short list and no way to judge the advice is in that position. And where most reviews are positive, a low one may count for more36.
Other findings may not transfer. We think a meal is judged within hours by the person who ate it, while legal skill is a credence quality that a client may never be able to judge. We expect reviews of a lawyer to speak mostly to what the reviewer could see, such as courtesy, speed of reply and clarity about fees, and much less to the quality of the legal work. A solo or small firm is closer to Luca's independents than to his chains in the one respect his study turns on: how few other signals of quality a stranger has. The FindLaw† answers describe respondents' own use of reviews, not what a change in a rating does to a firm's inquiries.
Why lawyers may advertise: Bates in one paragraph
In Bates v. State Bar of Arizona, decided June 27, 1977, the Supreme Court held that the State could not prevent publication of a lawyer's truthful advertisement of the availability and terms of routine legal services, because applying the disciplinary rule against it violated the First Amendment37. The Court framed the issue narrowly, as whether the State may prevent publication of such an advertisement in a newspaper37. It also said that "Advertising that is false, deceptive, or misleading of course is subject to restraint"37, and that there may be reasonable restrictions on the time, place and manner of advertising37.
Two further observations bear on a credence good. The Court noted that "the public lacks sophistication concerning legal services", so misstatements that might pass in other advertising may be quite inappropriate in legal advertising37. And on a matter it said it did not address, it observed that advertising claims about the quality of legal services are "not susceptible of measurement or verification" and may be so likely to mislead as to warrant restriction37.
Stigler called advertising "the obvious modern method of identifying buyers and sellers" and said that identifying them "reduces drastically the cost of search"26. We take Bates as the starting point for asking what a law firm may say to a stranger. This paper does not review later cases, so Bates is a beginning, not a statement of the rule in force today; Chapter 3 takes the story on to the Model Rules.
Law firm marketing, attorney marketing and legal marketing: what firm owners search for
Table 1 lists phrases that a law firm owner or a vendor might type about marketing, search optimization, the website, intake and AI, with US monthly search volume and average cost per click from Google Ads keyword data for November 2026. Volume is the average number of Google searches for a phrase in a month; cost per click is the average price an advertiser pays when someone clicks an ad for it.
| Phrase | Group | Searches a month | Average cost per click | Source |
|---|---|---|---|---|
| law firm marketing | Marketing | 1,300 | $75.68 | 38 |
| attorney marketing | Marketing | 720 | $134.66 | 38 |
| lawyer marketing | Marketing | 720 | $134.66 | 38 |
| law firm marketing agency | Marketing | 880 | $135.97 | 38 |
| legal marketing | Marketing | 480 | $81.14 | 38 |
| legal marketing agency | Marketing | 320 | $150.98 | 38 |
| law firm digital marketing | Marketing | 320 | $58.21 | 38 |
| law firm marketing strategies | Marketing | 170 | $73.62 | 38 |
| personal injury marketing | Marketing | 170 | $105.57 | 38 |
| law firm seo | Search optimization | 3,600 | $84.53 | 38 |
| attorney seo | Search optimization | 2,900 | $108.78 | 38 |
| lawyer seo | Search optimization | 2,900 | $108.78 | 38 |
| legal seo | Search optimization | 880 | $99.40 | 38 |
| law firm website design | Website and intake | 1,600 | $96.06 | 38 |
| legal answering service | Website and intake | 720 | $214.51 | 38 |
| law firm crm | Website and intake | 390 | $108.04 | 38 |
| legal intake software | Website and intake | 140 | $106.44 | 38 |
| law firm intake | Website and intake | 70 | $92.41 | 38 |
| law firm lead generation | Leads and rules | 90 | $137.20 | 38 |
| legal lead generation | Leads and rules | 70 | $67.60 | 38 |
| attorney advertising rules | Leads and rules | 70 | none reported | 38 |
| legal ai | AI | 18,100 | $31.62 | 38 |
| ai for lawyers | AI | 2,400 | $58.21 | 38 |
| ai for law firms | AI | 1,900 | $33.87 | 38 |
Table 1. US monthly searches and average cost per click for 24 phrases a law firm owner or vendor might type, Google Ads keyword data, November 2026 (our data).
The phrases are ones a firm owner or a vendor would type, not ones a client would, so Table 1 leaves out phrases such as a kind of lawyer plus a city, and this paper does not measure how many people search for a lawyer. Volume shows interest, not purchase. Cost per click is what advertisers pay, not what a client is worth. The data are for November 2026 only and say nothing about whether interest is rising.
Three patterns stand out. First, legal ai is searched about 18,100 times a month, which is about 5 times the 3,600 for law firm seo and about 14 times the 1,300 for law firm marketing (18,100 / 3,600 and 18,100 / 1,300)38. It also has the lowest average cost per click in Table 1, at $31.6238. Second, owners search for search optimization by name more than for marketing in general: law firm seo is about 2.8 times law firm marketing, and attorney seo and lawyer seo draw 2,900 each38. The data give identical figures for attorney seo and lawyer seo, and for attorney marketing and lawyer marketing; we treat each pair as one interest.
Third, the phrases about intake, lead generation and the rules are the least searched in Table 1, at 70 to 140 a month38. Legal answering service does not fit that pattern: it is searched about 720 times a month, and at $214.51 a click it has the highest cost per click in Table 138. A low volume does not make a subject unimportant; owners may use other words, and we draw no inference from it.
Our interest and the map of this paper
We sell some of what this paper discusses, and the reader should weigh the paper with that in mind. Technical Marketing's plans cost $999, $1,999 and $2,999 a month per location, or $9,990, $19,990 and $29,990 for a 12-month term paid in full, two months free. The $2,999 plan carries a $3,000 setup fee, $1,400 on a 12-month term, and the two lower plans a one-time onboarding fee of $499 and $999, waived on a 12-month term39. Its AI voice agent costs $299 a month with 250 call minutes included, then $0.39 a minute39. The current terms are on our pricing page, and our page for law firms describes the work. The argument does not depend on buying either. Chapter 4 sets these prices against the hours and matters a firm would have to recover, and the full law firm ledger is in the long paper's chapter on AI profit in dental, medical, legal and accounting practices. Where a point favors something we sell, or the evidence is thin, we say so.
We think a law firm sells a service a client cannot judge in advance, and it markets that service under conduct texts such as the ABA Model Rules, which address what a firm may say and how it may be recommended2,3. The Model Rules are a model, and each state adopts its own version. Respondents in the surveys above said they used referrals and online reviews27. An assistant will now answer the same question, naming firms1, though in our study the three assistants rarely named the same ones1. A firm that treats being found, being chosen, being reached and being measured as one system, built inside the conduct rules, should lose fewer of the clients who were already looking for it. That is an argument, and each chapter says where its evidence is thin.
A website that only displays the firm is a billboard. A website that also answers a visitor's first question and passes the inquiry to the intake team is part of a working system, inside limits that Chapter 4 sets out.
Chapters 2 to 6 follow the client's path. Chapter 2 covers how a firm is found in local search and in AI answers, using our study. Chapter 3 describes the attorney advertising rules, reads four states' texts against the model and reduces them to design requirements for a page, a profile and a review program. Chapter 4 treats intake, website design and the legal answering service. Chapter 5 weighs the evidence on legal AI. Chapter 6 shows what a small firm can count and gives an order of work for a first 90 days.
Chapter 2
Law firm SEO and attorney SEO: how a firm is found in search and in AI answers
Law firm SEO and local ranking: what Google says it weighs
Google's Business Profile help page says that local results are mainly based on relevance, distance and popularity40. It adds that prominence is based on information such as how many websites link to a business and how many reviews it has, and that more reviews and positive ratings can help a business's local ranking40. It also says there is no way to request or pay for a better local ranking on Google40. This is Google's own description, not an independent measurement of how the factors are weighted, and "mainly" means the three are not the only ones.
For a law firm, relevance is whether the profile and the pages match the matter a client types, such as a divorce or a car accident. Distance is where the office sits relative to the client, which a firm changes only by choosing where it has offices. We think popularity is the factor the page calls prominence, and it is the one a firm can influence over time, through the reviews it earns and the sites that link to it. A vendor's promise to buy a better local ranking would be at odds with what Google's page says. We do not know what any particular vendor offers; a firm can ask in writing.
Distance and position may matter more on a phone. Ghose, Goldfarb and Han reported that ranking effects are higher on mobile phones, with links at the top of the screen especially likely to be clicked, and that stores located in close proximity to a user's home are much more likely to be clicked on mobile phones41. Their subject was stores in a 2013 study, and a lawyer's office is not a store. The mechanism they suggest, higher search costs on a phone, so that the first entries on the screen carry more of the choice, may transfer. The size of the effect may not, since a client looking for a lawyer may take more care than a shopper.
Attorney SEO and Google's AI features: what Google says
Google says that a page must be indexed and eligible to be shown in Google Search with a snippet to appear as a supporting link in AI Overviews or AI Mode, and that "There are no additional technical requirements."42 A second Google Search Central page says that the best practices for SEO "remain relevant for AI features in Google Search" and that no special optimizations are necessary43. Google adds that existing SEO fundamentals "continue to be worthwhile"42.
Taken at face value, these statements say that attorney SEO is also the work of making pages eligible for Google's AI features. Three limits apply. These are Google's statements about Google's products, and eligibility is not a promise of appearing. Our study did not test Google's AI Overviews or AI Mode; it asked ChatGPT, Gemini and Perplexity through their APIs, which give a baseline and not what every signed-in user sees1, so nothing below shows what Google's summaries name. And this paper does not test whether ordinary SEO moves the answers of the other assistants.
The makers of the other assistants say little about how they choose local businesses. OpenAI's Help Center, read on November 10, 2026 and edited often, says ChatGPT ranks search results using multiple factors intended to help users find relevant, reliable information, does not list them, and does not guarantee placement44. It also says ChatGPT collects general location from the user's IP address and may rewrite a "near me" prompt into a search query that names the area, sharing the general location, not the IP address, with third-party search providers44. We infer that the area a user appears to be in can shape the query, so a profile or page that names the cities the firm serves supplies relevance. Statements by the makers of Gemini, Claude and Perplexity on how they choose local businesses are NF; the nearest evidence is OpenAI's statement above and the citation counts from our own study below, which show which sites each assistant cited, not why.
What happens to clicks when an AI summary appears
SISTRIX† (SEO software) and Backlinko† report steep differences in clicks by position in the ordinary results. Unless the text says otherwise, the studies in this section are not limited to local or legal searches. A SISTRIX† keyword study from 2020 found an average click-through rate of 28.5% for the first organic position in Google and 2.5% for the tenth45. That puts the first position at about 11 times the click rate of the tenth (28.5 / 2.5 = 11.4). Backlinko† reports an average click-through rate of 27.6% for the first result, which it puts at 10 times more likely to be clicked than a page in the tenth spot46.
Pew Research Center's analysis of US adults' browsing data found that users who encountered a Google AI summary clicked a traditional search result link in 8% of visits, against 15% of visits without one47. Users ended their browsing session entirely after 26% of search pages with an AI summary, against 16% of pages with only traditional results47. Pew also found that 18% of the Google searches in its study generated an AI summary, and that 88% of the summaries cited three or more sources while 1% cited a single source47. The search-page dataset held 68,879 unique Google searches, 12,593 of which produced an AI summary47.
A current measurement of local-pack clicks by position for lawyers is NF; the nearest evidence is a 2018 test by BrightLocal† (local-SEO software) in which 5,500 online testers chose among mock results for service-area businesses in San Francisco: the first local pack position received 16.1% of the clicks when Local Services ads were present and 17.3% when they were not48. We estimate that the first local pack entry for a lawyer draws roughly one choice in six (from those two shares; likely high). The test was a stated choice among mock results in which every tester picked one, the searches were for home services and not lawyers, and Google's layout has changed since 2018.
How often an AI summary appears depends on the kind of query. SE Ranking† (SEO software) checked 1,200 keywords, 300 per niche, on September 15, 2024, and found AI Overviews on 233 of 300 legal keywords (77.67%)49. The authors chose those keywords, they are not shown to be searches for a lawyer, and the data are two years old.
Whitespark† (local SEO) ran 540 queries by hand across Houston, Phoenix and Denver in six industries, one of them personal injury lawyers, and found AI Overviews on 15% of local-intent queries against 92% of informational-intent queries50. Local packs appeared on 93% of its local-intent queries50. That is a May 2025 snapshot on the authors' own query set, and this paper does not use its per-industry split. The two studies used different queries, dates and methods, and Google changes when AI Overviews appear, so we do not set them side by side as one trend. We think they are consistent with an AI summary sitting above a question about the law and a local pack above a search for a lawyer nearby, but neither shows that for lawyers alone.
A SparkToro† analysis of Datos clickstream panel data reports that 58.5% of American Google searches in 2024 resulted in zero clicks51. A search with zero clicks is not necessarily a lost client, because the searcher may have found what they needed on the results page, but the firm's website received no visit from that search. These sources count different things, from visits in one panel in 2025 to searches in another panel in 2024 to click rates by position, and should not be added or read as one trend.
How many people use assistants at all is a separate question. Chatterji and co-authors document that ChatGPT's consumer product had been adopted by around 10% of the world's adult population by July 202552. In a Pew Research Center survey of 5,119 US adults in February 2026, 49% said they ever use AI chatbots, and 42% said they ever use them to search for information53.
BrightLocal† reported in 2025 that 6% of consumers used ChatGPT or other AI tools to find local reviews, down from 9% in 202454. Its 2026 survey of 1,002 US adults reports that use of ChatGPT and other generative AI tools for local recommendations rose from 6% in 2025 to 45%55. The two BrightLocal† panels are separate and the question wording is not printed, so a jump of more than seven times may partly reflect a change of wording. The sources count different things, and each is stated use. None says how many clients ask an assistant for a lawyer, or contact a firm afterward.
The nearest measurements of AI referrals are of website traffic in general. Conductor† (SEO platform) found that AI referral traffic was 1.08% of all website traffic across its enterprise customers56, and Microsoft Clarity† (site analytics) reported that AI referrals were still under 1% of overall traffic on more than 1,200 publisher and news sites57. Neither dataset is a sample of law firms. Clarity† also measured conversion on those sites: sign-ups from LLM traffic converted at 1.66%, against 0.15% for search57, about 11 times (1.66% / 0.15%).
We estimate that AI referrals bring a law firm's site about a tenth of its conversions through clicks (from Conductor†'s share of visits and Clarity†'s conversion ratio; likely high). Weight each visit by how often it converts: with AI referrals at 1.08% of visits and converting 11 times as often as other visits, AI visits count 0.0108 x 11 = 0.119 and other visits count 0.9892, so AI visits supply 0.119 / (0.119 + 0.9892), or about 11%, of conversions.
The estimate runs high for a law firm because it joins two datasets and counts sign-ups on publisher sites, not signed matters. It runs low for recommendations that lead to a call without a click, which neither dataset counts. A measurement of AI referrals or AI recommendations turning into inquiries or clients at a law firm is NF in the vendor reports and research papers we could open; the nearest evidence is the two datasets above.
A count of referrals also misses answers nobody clicks. ChatGPT's own business dashboard counts a mention for a local business when ChatGPT shows a result for that location, even if the customer does not open the listing or visit the website58, and none of the figures cited here measures a recommendation that produces no click. Where an AI summary appears, a smaller share of visits end in a click on a result47, so we think being named in the answer carries more weight. This paper does not measure the size of that effect for law firms.
What AI assistants named: our study applied to lawyers
Table 2 sets the three lawyer rows from our study beside the overall figures. Counts are of businesses named, counting a business once per question1. The two three-city rows are small samples. Each question was asked once, names were matched by spelling, and up to eight names were read per answer1.
| Kind of business | Cities asked | Businesses named | Per question | Named by all three | Named by one assistant only | Source |
|---|---|---|---|---|---|---|
| Family law attorneys | 16 | 240 | 15 | 13 (5.4%) | 75.8% | 1 |
| Personal injury lawyers (small sample) | 3 | 46 | 15.3 | 3 (6.5%) | 76.1% | 1 |
| Estate planning attorneys (small sample) | 3 | 47 | 15.7 | 1 (2.1%) | 83% | 1 |
| All 23 kinds of business, 199 questions | up to 16 | 2,975 | not given | 122 (4.1%) | 79.7% | 1 |
| Ten kinds asked in all 16 cities, 160 questions | 16 | 2,414 | not given | 84 (3.5%) | not given | 1 |
Table 2. Businesses named by ChatGPT, Gemini and Perplexity in our study, for lawyers and for all kinds of business (our data, 2026-11-10; each question asked once).
In every lawyer row most of the businesses named were named by only one assistant. The 6.5% for personal injury lawyers and the 2.1% for estate planning attorneys are the gap between 3 businesses and 1, which is too little to tell apart1. Across the ten kinds asked in every city, the share of named businesses that all three assistants named ranged from 1.9% in Charlotte to 7.9% in Seattle1. That range covers the ten kinds together, and this chapter does not break the lawyer rows out by city, so it cannot say how far they move from city to city. Each question had one fixed form, with the city and the kind of business swapped in, and told the assistant to name specific businesses1.
The overall figures, which cover all 23 kinds of business and not lawyers alone, show how rarely the assistants agree. All three assistants named the same business first in 9 of the 199 questions, and at least two agreed on the first name in 751. In 108 of the 199 questions (54%), no business was named by all three1. Any two assistants overlapped on between 11.9% and 14.3% of the businesses they named1. A firm therefore cannot check one assistant and know how it appears to the others, and a firm named first by one may be absent from another's list. The study does not say why, and it does not show that a firm named by all three is better than one named by none, because a name in an answer is not an endorsement1.
Because our study asked each question once, a repeat-asking study on other subjects is relevant. SparkToro† and Gumshoe.ai†, in a 2026 report, had 600 volunteers run 12 prompts through each of ChatGPT, Claude and Google's AI Overview, 2,961 runs in all59. They report a chance of less than 1 in 100 that ChatGPT or Google's AI, asked a prompt 100 times, would give the same list of brands in any two responses59, and of less than 1 in 1,000 for the same list in the same order59. The study is not peer reviewed, its author discloses a possible conflict of interest because his research partner works at Gumshoe.ai†, volunteers pasted the answers in by hand, and its prompts were not about dentists, doctors, lawyers or accountants. We think a single answer is one draw from a spread, and a firm's absence from one run is weak evidence of absence from the next.
Which sites the assistants cited, and how the assistants differ
The 597 answers cited 1,455 different websites, counting a site once per answer, for 3,530 citations in all1. A site was usually cited by one assistant: 88% of the websites cited were cited by only one of the three1. The 25 most-cited websites drew 29% of all citations, and each was cited in at least 9 of the 16 cities; they are directories, review platforms, list sites, publishers and a forum, not any one local business's site1. Websites cited in only one city, where a local business's own site would fall, drew 39% of citations1. The other 61% went to websites cited in two or more cities (calc., from the 39%). These site figures cover all 199 questions and 23 kinds of business, not lawyers alone.
Table 3 gives per-answer averages over all kinds of business, not lawyers alone. The assistants differ in how much they say and how they source it.
| Assistant | Model | Businesses named per answer | Websites cited per answer | Answers citing Reddit | Source |
|---|---|---|---|---|---|
| ChatGPT | gpt-6-luna | 4.8 | 3.5 | 2% | 1 |
| Gemini | gemini-3.8-flash | 6 | 6.5 | 42% | 1 |
| Perplexity | sonar | 7.8 | 7.7 | 0% | 1 |
Table 3. Average businesses named and websites cited per answer, and the share of answers citing Reddit, by assistant, over all kinds of business (our data, 2026-11-10).
The sources differ as well. Gemini's most-cited website was reddit.com, in 83 of 199 answers1. Perplexity's was bbb.org, in 58 of 199 answers1. ChatGPT's was a different site again, in 32 of 199 answers1. We think these are different citation habits, so a firm's profile on one platform may matter to one assistant more than to another. The study cannot show that a profile caused a name to appear.
Generative engine optimization: what the research found and its limits
Generative engine optimization (GEO) is the research literature's name for methods that raise a source's visibility in the answers of generative engines. The paper that introduced it reports that its methods can raise that visibility by up to 40%60, and that its evaluations across benchmark queries show improvements up to 37%61. An "up to" figure is a ceiling, not a typical effect. The same paper reports that its strategies' efficacy varies across domains, "underscoring the need for domain-specific optimization methods"60. It also describes generative engines as black boxes and says "content creators have little to no control over" when and how their content is displayed60.
The cited claims do not say that the research covered legal queries, local queries or any named assistant that a client would use, so this paper cannot say whether any of its methods works for a law firm in a city. We think the evidence shows that visibility in generated answers can move, that it moves differently by subject, and that a firm has little direct control, which is a thin basis for a service line.
A second finding bears on how a firm should read an answer that mentions it. Liu, Zhang and Liang audited commercial generative search engines and found that, on average, "a mere 51.5% of generated sentences are fully supported by citations and only 74.5% of citations support their associated sentence"62. Those were the engines of 2023, and the products audited may have changed since. A cited page does not prove that an assistant said something true about the firm, and an assistant's description of a firm may match no page, so a firm that checks its answers should check the facts stated, not only the presence of its name.
Searches for generative engine optimization average about 4,400 a month at $35.04 a click, and for answer engine optimization about 2,400 a month at $21.3138. These are phrases a marketer or a vendor would type, so a firm may meet vendors who use them. A firm weighing such an offer can ask what was measured, on which queries, in which assistants and on what dates, and can treat a guarantee of an AI placement as a reason for doubt.
What a law firm controls: profiles, directories, practice-area pages and reviews
A firm controls its inputs, not the outputs. Google says there is no way to request or pay for a better local ranking40, and OpenAI says placement in ChatGPT is not guaranteed44. What a firm controls is the accuracy, completeness and consistency of what it publishes and of what is published about it.
The first input is a profile. Because Google bases local results on relevance, distance and popularity40, we think a profile that states the practice areas, location, hours and telephone number the same way the website does is an inexpensive way to supply relevance. Directories and review platforms are the second input. The most-cited sites in our study were directories, review platforms, list sites, publishers and a forum1, so an error in a listing on one of them can travel. A firm does not control those sites, but it controls whether its own entries are correct.
Paying for a listing raises a separate question. ABA Model Rule 7.2(b) bars a lawyer from compensating, giving or promising anything of value to a person for recommending the lawyer's services, with listed exceptions that include the reasonable costs of permitted advertisements3. The Model Rules are a model, and each state adopts its own version. Whether a given paid listing is advertising or a paid recommendation is a question for the firm's counsel.
Practice-area pages are the third. A page built to appear as a supporting link in Google's AI features would cover one kind of matter, name the firm and give contact details, and put the answer to the question it targets in its opening 40 to 60 words, in a form an assistant could quote, as we write the pages of our own site. In our study, websites cited in only one city, where a local business's own site would fall, drew 39% of citations, so such sites are not shut out of the answers, though the group may include other kinds of site1.
Reviews are the fourth input. Google says reviews and positive ratings can help local ranking40. A review program is a process the firm controls, though what any single reviewer writes it does not. Chapter 3 treats asking for, paying for and answering reviews, starting with ABA Formal Opinion 496, which says that a negative online review, alone, does not meet the requirements of permissible disclosure in self-defense under Model Rule 1.6(b)(5)63.
A monthly answer check
A monthly answer check is a design, not a result. This paper reports no results from such checks, and the steps below promise nothing about placement.
- Fix a list of questions once. Use one for each practice area and each office city, and add a second wording for each, because our study used one fixed wording and clients use many1.
- Run every question in each assistant the firm cares about, and in Google where an AI summary appears, on one day each month. Run each question three times in each assistant, because repeated runs of one prompt can return different lists59. Record the assistant, the date, the exact wording, and whether the session was signed in, because answers through an API are a baseline and not what every signed-in user sees1.
- For each answer, record the businesses named and their order, whether the firm is among them, and the websites cited.
- Check the facts stated about the firm against the firm's own records: address, telephone number, practice areas and anything that reads as a claim of specialty or results. Pass any such claim to counsel for review under the rules in Chapter 3.
- Keep each answer verbatim, as text or a screenshot, so that a later month can be compared with it.
- Enter no client information in any assistant, because a question about the firm's own marketing needs none. ABA Model Rule 1.6(c) requires a lawyer to make reasonable efforts to prevent the inadvertent or unauthorized disclosure of, or unauthorized access to, information relating to the representation of a client64; how it applies to a given tool is for the firm's counsel.
Suppose a firm tracks 12 questions on three assistants and asks each three times. That is 108 answers a month (12 questions x 3 assistants x 3 runs), and a single name moving in or out of one of them is not a trend, because one answer is one draw, not a standing result. Wrong facts are the first thing to act on, because the firm can correct them. Changes in position are a reason to look again next month. Chapter 6 places the check beside the other counts a firm keeps, and the long paper's chapter on measurement and attribution for a small practice goes further.
A repeat-asking study of AI recommendations for local businesses or professional practices is NF; the nearest evidence is the SparkToro† and Gumshoe.ai† study above, whose examples of limited options included Los Angeles Volvo dealers59. We expect lists of lawyers in a city, where the options are far more numerous, to repeat no more often than those lists did. If so, its under-1-in-100 figure is likelier an upper bound than a lower one for lawyers, and a firm should expect its name to come and go between runs. Repeating the questions on one day and across months, as the check above does, would show how stable the answers are for one firm in one city.
Chapter 3
Attorney advertising rules and what the texts say about law firm marketing
Attorney advertising rules are a set of texts: two Supreme Court decisions of 1976 and 197737,65, the American Bar Association's Model Rules, an ABA formal opinion on responding to online criticism, a federal rule on consumer reviews, and the rules and ethics opinions of each state. This chapter describes what the model texts say, reads four states' rules and one state's rewrite against them, and reduces the texts to design requirements for a site, a profile and a review program.
The Model Rules are a model that each state adopts in its own version, and a formal opinion is advisory. Where the chapter says a design should do something, that is our design requirement, not a statement of what the rules allow or forbid.
In US search data from November 2026, "attorney advertising rules" is searched about 70 times a month, with no cost per click reported38, and "law firm marketing" about 1,300 times a month38. The second phrase is searched roughly 19 times as often as the first (1,300 / 70). The phrases are ones a firm owner or a vendor would type, and volume shows interest, not purchase. We think the gap shows where a firm's attention goes first.
Attorney advertising rules at a glance: decisions, models, an opinion and a federal rule
Four kinds of text are described here, and the kind decides how much weight a text carries. Two are Supreme Court decisions: Virginia State Board of Pharmacy v. Virginia Citizens Consumer Council, from 197665, and Bates v. State Bar of Arizona, decided June 27, 197737. Four groups are ABA Model Rules and their comments: Rule 7.1 on false or misleading communications, Rule 7.2 on recommendations, specialists, the responsible lawyer and lead generators, Rule 7.3 on solicitation, and Rules 5.3 and 1.6(c) on nonlawyer assistance and client information. One is an advisory ABA formal opinion, Formal Opinion 496, dated January 13, 2021, on a negative online review63. One is a federal trade regulation rule, 16 CFR part 465, effective October 21, 2024, on consumer reviews and testimonials66. A fifth kind, the rules and ethics opinions of the states, is read in a later section. The two decisions are read as of their dates, without checking later history, so counsel should confirm what later decisions add.
From Bates to the Model Rules: why a law firm may advertise at all
In Bates v. State Bar of Arizona, decided June 27, 1977, the Supreme Court held that the State could not prevent publication of a lawyer's truthful advertisement of the availability and terms of routine legal services, because applying the disciplinary rule against it violated the First Amendment37. The Court framed the question narrowly, saying the constitutional issue was "only whether the State may prevent the publication in a newspaper of appellants' truthful advertisement concerning the availability and terms of routine legal services"37.
Bates cites the Court's decision of the year before. The syllabus to Virginia State Board of Pharmacy summarizes the Court as holding that commercial speech is not wholly outside the protection of the First and Fourteenth Amendments65, and Bates cites that case for the point that commercial speech "is entitled to some First Amendment protection"37. The Virginia decision also says untruthful speech, commercial or otherwise, has never been protected for its own sake65.
Bates also says that "Advertising that is false, deceptive, or misleading of course is subject to restraint"37, and that there may be reasonable restrictions on the time, place and manner of advertising37. Two further observations bear on legal advertising in particular. Because the public lacks sophistication concerning legal services, the Court observed, misstatements that might be overlooked or deemed unimportant in other advertising may be found quite inappropriate in legal advertising37. And about a matter it said it did not address, it noted that claims as to the quality of legal services are not susceptible of measurement or verification and so may be so likely to be misleading as to warrant restriction37. That remark concerns a matter the Court set aside and is not a ruling on quality claims.
Darby and Karni, whose article we know through a quoted passage, distinguished credence qualities, which are "expensive to judge even after purchase"21. Akerlof listed the licensing of doctors, lawyers and barbers among the practices that reduce quality uncertainty25. We think rules on what a lawyer may say belong to the same family of institutions, because a client cannot easily check a claim in advance. The long paper's chapter on the theory of choosing a professional argues that case in full.
Neither decision contains a rule for a website. They set a boundary: truthful information of the kind the Bates advertisement carried may not be restrained, and false, deceptive or misleading advertising may be. We think the Model Rules write the second half of that boundary into text.
Rule 7.1: false or misleading communications in lawyer marketing
ABA Model Rule 7.1 says a lawyer shall not make a false or misleading communication about the lawyer or the lawyer's services. A communication is false or misleading if it contains a material misrepresentation of fact or law, or omits a fact necessary to keep the statement, taken as a whole, from being materially misleading2.
Comment [2] reaches statements that are true. It says "Misleading truthful statements are prohibited by this Rule" and that a truthful statement is misleading if it omits a fact necessary to make the lawyer's communication, considered as a whole, not materially misleading67. A second test is that a truthful statement is misleading if a substantial likelihood exists that it will lead a reasonable person to formulate a specific conclusion about the lawyer or the lawyer's services for which there is no reasonable factual foundation67.
Comment [3] addresses claims and comparisons. It says an unsubstantiated claim about a lawyer's or law firm's services or fees, or an unsubstantiated comparison of those services or fees with those of other lawyers or law firms, may be misleading if presented with such specificity that a reasonable person would conclude it can be substantiated. The same comment says a truthful report of past results may mislead if it creates an unjustified expectation of the same results for other clients67.
A marketing page contains all three things these comments name: factual claims, comparisons and reports of results. Whether a given sentence crosses any of the lines the comments draw is for the firm's counsel, under the state's version of the rule. A firm can keep a record without answering that question. We think the minimum is a claims register: for every factual claim, comparison and reported result on a site, who wrote it, what supports it and when it was last checked. A site with one page per practice area and city has many more claims to substantiate than a site with five pages, so the register matters more as pages multiply.
Rule 7.2: paying for recommendations, specialist claims and the responsible lawyer
Paragraph (b) of Rule 7.2 bars a lawyer from compensating, or giving or promising anything of value to, a person for recommending the lawyer's services. It lists exceptions, including the reasonable costs of permitted advertisements, the usual charges of a lawyer referral service and nominal gifts of appreciation3. A firm that pays a person in connection with marketing should ask counsel whether the payment is for a recommendation and, if so, which exception fits.
Paragraph (c) concerns specialists. It bars a lawyer from stating or implying certification as a specialist unless an organization approved by the state or accredited by the ABA has certified the lawyer and the communication clearly names that organization3. The operative words are "state or imply." Which words imply certification is counsel's call. The design response is a short approved word list that every page, profile and advertisement draws from.
Paragraph (d) requires any communication made under the rule to include the name and contact information of at least one lawyer or law firm responsible for its content3. What counts as a communication made under the rule is for counsel. The design response is a footer template that carries a responsible lawyer's name and contact details on every page the firm controls.
Comment [5] covers lead generators. It says a lawyer may pay others for generating client leads, including internet-based leads, as long as the lead generator does not recommend the lawyer, the payment is consistent with Rules 1.5(e) and 5.4, and the generator's communications are consistent with Rule 7.168. A lawyer must not pay a lead generator that states, implies or creates a reasonable impression that it is recommending the lawyer, is making the referral without payment from the lawyer, or has analyzed a person's legal problems when determining which lawyer should receive the referral68.
"Law firm lead generation" is searched about 90 times a month at a cost per click of $137.2038, and "legal lead generation" about 70 times at $67.6038. Cost per click is what advertisers pay for a click, not what a lead fee or a client is worth. This chapter does not describe Rule 1.5(e) or Rule 5.4. Chapter 6 treats lead generators as a measurement question: what to count when a firm buys leads.
Rule 7.3: solicitation and the limit on live contact
Rule 7.3(a) defines solicitation as a communication initiated by or on behalf of a lawyer or law firm, directed to a specific person the lawyer knows or reasonably should know needs legal services in a particular matter, that offers to provide legal services for that matter69. Rule 7.3(b) bars a lawyer from soliciting professional employment by live person-to-person contact when a significant motive is pecuniary gain, unless the contact is with a lawyer, a person with a family, close personal, or prior business or professional relationship with the lawyer or firm, or a person who routinely uses such legal services for business purposes69.
We turn the definition into three design questions for any system that sends messages or places calls for a firm. Who initiated this contact, the person or the firm? Is the message addressed to one named person or to anyone who reads it? Did the person ask?
The questions matter more as software does the contacting. A follow-up sequence triggered by a form submission, a text to a lead bought from a generator and a call placed by an automated voice may each raise the question whether it is a contact made "on behalf of" a firm, in the definition's words. Chapter 4 takes up intake and the FCC ruling on calls made with AI-generated voices70. The requirement here is a log of every outbound contact the firm's systems make, with the trigger that caused it and the person's prior request, if any.
Reviews and replies: Formal Opinion 496 and the FTC rule on fake and bought reviews
ABA Formal Opinion 496, dated January 13, 2021, says a negative online review, alone, does not meet the requirements of permissible disclosure in self-defense under Model Rule 1.6(b)(5)63. It also says that, as a best practice, lawyers should consider not responding to a negative post or review, because responding may draw more attention to it, and that lawyers may ask the host website or search engine to remove it63.
A reply can carry information. In health care, the Manasa Health Center resolution agreement says the center impermissibly disclosed the protected health information of four patients in response to their negative reviews on Google Reviews, a contrast from another profession and another rule71. For a law firm, the design requirement is a reply template, approved by a lawyer before use, that contains nothing about any client's matter, and a rule about who may post it.
The FTC's Trade Regulation Rule on the Use of Consumer Reviews and Testimonials (16 CFR part 465), a federal rule, was published in the Federal Register on August 22, 2024 and took effect October 21, 202466. Under 16 CFR 465.2(a), it is an unfair or deceptive act or practice for a business to write, create or sell a consumer review, consumer testimonial or celebrity testimonial that materially misrepresents that the reviewer exists, used the product, or what their experience was72. The FTC said the rule addresses reviews that misrepresent that they are by someone who does not exist, such as AI-generated fake reviews73.
Section 465.4 makes it an unfair or deceptive act for a business to provide compensation or other incentives in exchange for, or conditioned on, consumer reviews "expressing a particular sentiment, whether positive or negative"74. Section 465.7 makes it a violation for anyone to use an unfounded or groundless legal threat, a physical threat, intimidation or a knowingly false public accusation in response to a consumer review in an attempt to prevent it being written or to have it removed75.
The FTC says the rule authorizes courts to impose civil penalties for knowing violations76, and in warning letters of December 22, 2025 it said violations can bring civil penalties of up to $53,088 per violation77. A firm that hands its review program to an agency should ask the agency which of these sections its methods touch. Chapter 1 reports what surveys say about reviews, and Chapter 2 what Google says about them.
Asking a client for a review is a separate question from answering one, and two state bar opinions speak to it. Texas Ethics Opinion 685, dated January 2020, concludes that the Texas Disciplinary Rules do not prohibit a lawyer from encouraging current and former clients to leave positive reviews or ratings online, provided the lawyer does not encourage statements that are false, misleading or unfounded78. North Carolina's 2018 Formal Ethics Opinion 1, adopted in July 2018, says a lawyer may ask present or former clients to post reviews on the lawyer's directory profile as long as there is no quid pro quo, and that under no circumstances may a lawyer solicit, encourage or assist in the posting of fake, false or misleading reviews79. Each is an opinion of one state's bar and not a rule, and neither speaks for other states; whether either bears on a firm elsewhere is open. Both turn on what is attached to the request and what the review says, the same points the FTC rule makes about incentives and fake reviews.
Marketing agencies and vendors: Rules 5.3 and 1.6(c)
A law firm's marketing runs through other people's hands: an agency that writes pages, a host that stores form submissions, a call-tracking service, a directory, a CRM. ABA Model Rule 5.3 requires a partner or lawyer with comparable managerial authority to make reasonable efforts to ensure the firm has measures giving reasonable assurance that a nonlawyer employed, retained by or associated with the lawyer behaves compatibly with the lawyer's professional obligations. It makes the lawyer responsible for a nonlawyer's conduct that would violate the Rules if the lawyer orders or ratifies it80.
Rule 1.6(c) requires a lawyer to make reasonable efforts to prevent the inadvertent or unauthorized disclosure of, or unauthorized access to, information relating to the representation of a client64. Whether a marketing vendor, or an automated tool, is a nonlawyer "employed or retained by or associated with" a lawyer in the sense of Rule 5.3 is for counsel80. Both paragraphs ask for reasonable efforts and measures and do not mention a contract form. In health care, the HIPAA Rules permit a covered entity to disclose protected health information to a business associate if it obtains satisfactory assurances in a contract or other written arrangement81, and our check of 28 software tools found 9 that sign a business associate agreement without requiring a special tier, according to the vendors' own pages82.
"Law firm marketing agency" is searched about 880 times a month at $135.97 a click38, and "legal marketing agency" about 320 times at $150.9838. We are a marketing practice and so have an interest in this section. We think a firm's lawyer should read every page, profile and advertisement a marketing partner drafts before it is published.
The design response is a vendor inventory. For each vendor, list what it receives from the firm (form data, call audio, messages, client names), who at the firm approved it, what the agreement says about use and deletion, and who holds the accounts. Accounts for listings, advertising and analytics belong in the firm's name, so that changing agencies does not mean losing the record.
A table of design requirements for a page, a profile, an advertisement and a review request
Table 1 reduces the model texts, the FTC rule and two state opinions to what we would build. Each cell says what a page, a profile, an advertisement or a review request would carry or log so that a lawyer can review it. "Not addressed" means the text in that row does not speak to that surface.
| Text | A page or site | A profile or listing | An advertisement or paid placement | A review request or reply |
|---|---|---|---|---|
| Rule 7.1 and comments2,67 | A claims register: each claim, comparison and result has an owner, a basis and a review date | The same reviewed wording in every listing | Each claim traces to the register before the advertisement runs | Quoted reviews are shown whole or with omissions marked |
| Rule 7.2(b)3 | Any recommended-by element traced to who is paid and for what | Each paid listing recorded with what the payment buys | Each paid placement classified for review | No gift or payment attached to a request without approval |
| Rule 7.2(c)3 | An approved word list for describing practice areas | Directory specialty fields filled from the same list | The list governs headlines and copy | Client reviews that use the listed words are read by a lawyer before they are featured |
| Rule 7.2(d)3 | Responsible lawyer's name and contact in every page template | The same in each listing that allows it | Name and contact in the advertisement or its landing page | The request names the firm |
| Comment [5] to Rule 7.268 | Forms record the source of each inquiry so generator leads can be told apart | Generator profiles read for words that suggest a recommendation or an analysis of the caller's problem | Generator contract and copy go to a lawyer before signing | Not addressed |
| Rule 7.369 | A form states what contact the visitor will receive and from whom | Not addressed | A log of every contact directed to one person, with its trigger and the person's prior request | Requests go to people who have been clients, kept apart from any offer of services |
| Formal Opinion 49663 | Embedded reviews are the client's words; replies are not embedded unreviewed | An approved reply template with no client information, and a rule on who may post | Not addressed | Private complaints go to a named person before any public reply |
| 16 CFR part 46566,72,74,75 | Only real clients' reviews appear; none written by staff, relatives or software, none bought | The same, and no staff account posts a review | A testimonial traces to a real client who used the service | One text for every client; no incentive tied to a sentiment; no threat in any message or reply |
| Texas Ethics Opinion 685 and North Carolina 2018 Formal Ethics Opinion 178,79 | Not addressed | A request for a review on a directory profile follows an approved script | Not addressed | A request offers nothing of value for a review, and nobody on the team encourages or assists a false, misleading or unfounded one |
| Rules 5.3 and 1.6(c)64,80 | An inventory of every vendor that receives form data, calls or messages, with who approved it | Listing accounts held in the firm's name | Advertising accounts and landing pages owned by the firm | The tool that sends requests holds only the contact data it needs |
Table 1. Design requirements by text and surface. These are our requirements, not statements of what any rule permits.
Read down the columns, the table describes a few pieces of plumbing. A claims register and a word list serve the page, the profile and the advertisement at once, so the three should draw from one source. A vendor inventory serves every row that involves a tool, and a contact log serves Rule 7.3. Each can start as a spreadsheet, and a firm that adds them late has to find, after the fact, where each claim and each contact came from.
State attorney advertising rules: how California, Texas, Florida, Washington and New York differ from the Model Rules
The rules a firm answers to are its own state's. Table 2 sets paragraphs from four states' rules, read on November 10, 2026, beside the model paragraphs described above, and records the change to New York's. It covers the paragraphs read and not each state's whole rule set, and no statute or bar opinion beyond the two review opinions named earlier.
| State | Responsible lawyer or firm named | Label and filing | Results and testimonials | Edition read and caution |
|---|---|---|---|---|
| Model Rules | Rule 7.2(d): the name and contact information of at least one responsible lawyer or law firm3 | The model paragraphs described above, 7.3(a) and 7.3(b), mention no label and no filing69 | Comment [3] to Rule 7.1: a truthful report of past results may mislead if it creates an unjustified expectation of the same results for other clients67 | A model that each state adopts in its own version |
| California | Rule 7.2(c): the name and address of at least one lawyer or law firm responsible for the content83 | Rule 7.3(c): the word "Advertisement" or words of similar import on a written, recorded or electronic communication soliciting a person known to need legal services in a particular matter, with listed exceptions83. No filing rule is described here | Comment [4] to Rule 7.1: a truthful report of achievements, or a testimonial or endorsement, may mislead if it creates an unjustified expectation of the same results for other clients in similar matters83 | Rules PDF headed "Current Rules 2026", read November 10, 2026. Rules only: no statute or bar opinion was read |
| Texas | Rule 7.02(a): the name of a responsible lawyer and the lawyer's primary practice location84 | Rule 7.04(a): unless exempt under Rule 7.05, file a copy of each advertisement, a completed application and a fee with the State Bar's Advertising Review Committee no later than ten days after dissemination84. Rule 7.05(b) exempts information and links posted on a law firm website, except the contents of the homepage84. Texas's separate solicitation rule is not described here | No Texas rule on results or testimonials is described here | State Bar PDF headed as including amendments effective January 2022, read November 10, 2026. Rule 7 was renumbered in 2021, so older sites show Rule 7.07. Changes after an October 2024 amendment order were not checked |
| Florida | Rule 4-7.12(a): the name of at least one lawyer or the law firm responsible, and the city, town or county of one or more bona fide office locations of the lawyer who will perform the services85 | Rule 4-7.19(a): subject to the exemptions in Rule 4-7.20, file a copy of each advertisement with The Florida Bar at least 20 days before first dissemination85. Rule 4-7.20(g) exempts information on the lawyer's website from that filing85 | Rule 4-7.13(b)(9): a testimonial is deceptive or inherently misleading if, among other things, the lawyer wrote or drafted it, it was given in exchange for something of value, it does not represent what clients generally experience, or it lacks a disclaimer that the prospective client may not obtain the same or similar results85 | Chapter 4 of the Rules Regulating The Florida Bar, compilation headed October 2026, read November 10, 2026. Rules 4-7.19 and 4-7.20 were last amended effective July 2025. The website exemption is from filing, not from the content rules |
| Washington | RPC 7.2 is marked reserved, so the state has no separate advertising rule86 | RPC 7.3(a) permits solicitation unless it is false or misleading, the person cannot exercise reasonable judgment because of their physical, emotional or mental state, the person has said they do not want to be solicited, or it involves coercion, duress or harassment87. No label or filing rule is described here | Comment [3] to RPC 7.1: an advertisement that truthfully reports a lawyer's achievements for clients or former clients may mislead if it creates an unjustified expectation of the same results for other clients in similar matters88 | Washington rules PDF, read November 10, 2026. That RPC 7.1 governs advertising is an inference from its text, not a statement on the reserved 7.2 page. Only paragraph (a) of RPC 7.3 is described here |
| New York | Not read: the new text could not be opened | The New York State Bar Association says that effective June 1, 2026 the Appellate Division replaced the former detailed requirements, including attorney advertising labels, record retention and solicitation filing, with a simpler, principles-based approach closer to the ABA's 2018 rules89. This paper does not state the former requirements as current law | Not read | A bar association's account, which is secondary. The court's order could not be read. A public reproduction read on November 10, 2026 still showed the former text, so older New York checklists and pages may describe a rule that no longer applies |
Table 2. Selected paragraphs of four states' lawyer advertising rules, and New York's change, beside the model paragraphs described above (texts read November 10, 2026). These are descriptions, not statements of what any rule permits.
Three differences matter for a firm's templates. The first is what an advertisement must name: California, Texas and Florida each ask for a location detail beyond the model's name and contact information, and Washington's reserved Rule 7.2 leaves it with no counterpart. A footer that carries a responsible lawyer's name, street address, office city and telephone number would give one place to adjust, though whether it meets any state's rule is not settled here.
The second is filing. Texas asks for a copy within ten days after an advertisement is disseminated, and Florida asks for one at least 20 days before. A filing calendar therefore needs a clock that runs forward from the publication date for one state and backward from it for the other. Both treat a website differently from other media, and not alike: Texas's exemption leaves out the contents of the homepage, while Florida's exempts website information from filing but not from its content rules. Whether a given page, profile or paid placement is an advertisement that must be filed is for counsel, and a log of what was published, where and on what date is the record that answer needs.
The third is results and testimonials. We think California's and Washington's comments to Rule 7.1 follow the model's comment on past results, and California's names testimonials and endorsements as well. Florida states its testimonial conditions in a rule paragraph of its own. Whether a client's online review is a testimonial under Florida's rule is for counsel. For a review program the design point is the one in Table 1: the firm does not draft reviews and attaches nothing of value to them, and what disclaimer, if any, a page that shows reviews needs is part of the same question.
What this chapter does not settle
The chapter reads four states and the change to a fifth, and no other state's text, so Table 2 is not a count of how states differ and every row of Table 1 needs to be read against the firm's own state. It describes selected paragraphs of Rules 7.1, 7.2 and 7.3 and not the rest of those rules, and it does not describe Rule 5.5.
This paper does not measure how advertising rules are enforced, how often lawyers are disciplined for marketing, or how compliance relates to results, and it does not test whether following any row of Table 1 helps or hurts a firm's ranking, its review count or its inquiries. The table is a design argued from the texts.
The firm's lawyer can read the state's version of each rule and any state bar opinion on websites, directories and reviews, and the firm can compare Table 1 with its present pages and listings and mark each gap. The four rulebooks compared sets the legal rules beside those for dental, medical and accounting practices, for readers who want to see how much of this plumbing carries across professions.
Chapter 4
Law firm intake, website design and the legal answering service
Law firm intake is the first contact between a person with a legal problem and the firm: a call, a form, a chat message or an email, up to the point where a consultation is booked or the person is told the firm cannot help. A firm that was found and chosen can still lose the client at this stage. This chapter covers what the evidence says about responsiveness, what Rule 1.18 and related texts ask of intake, and how a website, an answering service, an AI voice agent and a CRM fit together.
Table 1 gives US monthly searches and cost per click for the phrases that name intake tools and services, from Google Ads keyword data of November 2026. The phrases are ones a firm owner or a vendor would type, volume shows interest and not purchase, and cost per click is what advertisers pay, not what a client is worth.
| Phrase | US searches a month | Cost per click |
|---|---|---|
| ai answering service (not specific to law) | 4,40038 | $58.4838 |
| law firm website design | 1,60038 | $96.0638 |
| legal answering service | 72038 | $214.5138 |
| law firm crm | 39038 | $108.0438 |
| legal intake software | 14038 | $106.4438 |
| law firm intake | 7038 | $92.4138 |
Table 1. Phrases for intake tools and services, US Google Ads data, November 2026.
Of the legal phrases in our keyword data, "legal answering service" has the highest cost per click, $214.5138, and "law firm intake" is searched less than any other phrase in Table 1.
Law firm intake as a conversion problem: what a vendor's secret shop found
The best-known tests of law firm responsiveness are secret shops run by vendors. In the 2024 secret-shopper study by Clio† (legal practice-management software), 500 law firms were emailed and phoned with a client inquiry, and the results are in Table 290. Only 40% of firms answered the phone call, down from 56% in 201990. Six firms in ten did not answer (100% - 40%) and, since 52% either answered or phoned back, 12 percentage points of firms did not answer but phoned back (52% - 40%). Clio†'s own gloss is that 48% were "essentially unreachable by phone"90. According to Clio†'s press release, firms that responded to emails did so promptly, but only 2% referenced similar legal cases as requested by shoppers91.
| Measure, as the vendor reports it | 2019 | 2024 |
|---|---|---|
| Firms that responded to an emailed inquiry | 40%90 | 33%90 |
| Firms that answered the phone call | 56%90 | 40%90 |
| Firms that answered or phoned back | Not given | 52%90 |
| Firms essentially unreachable by phone | Not given | 48%90 |
| Among firms that replied by email, share that replied within eight hours | Not given | 84%91 |
| Among firms that replied by email, share that gave clear next steps or cost information | Not given | 18%91 |
| Shoppers unlikely to recommend the firms they contacted | Not given | 73%91 |
Table 2. Clio†'s 2024 secret-shopper results for 500 law firms. The 2019 figures are Clio†'s restatement of its earlier study inside the 2024 report.
The figures are from Clio†'s highlights page and a press release, not the full report. The highlights page does not say how the 500 firms were chosen, so the figures are indicative and not a count of US firms. The release does not make plain what its 2% for similar legal cases is a share of91, so we do not compute with it. A shopper's inquiry is not a person with a live problem. Clio†'s survey of 1,028 US legal professionals is a different instrument90, and Chapter 5 uses it for AI.
Other tests point the same way, each with its own measure. Clio†'s 2019 report describes its first shop as 1,000 US law firms emailed, 500 of them also phoned. Of the calls, 56% were answered, 39% went to voicemail and 5% went unanswered, and 57% of the voicemail calls were not returned within 72 hours29. About 22 calls in 100 ended in voicemail with no call back inside three days (39% x 57%). A bar-association magazine (secondary) reports a 2016 secret shop by an American Bar Association committee in which one in three callers did not speak to a person and callers were put on hold 45% of the time92.
Hennessey Digital† (law firm marketing) reports that in 2025 it contacted 1,333 US law firms through their online lead forms, that 74% responded within seven days, up from 59% in 2021, and that the median response time to an online lead was 13 minutes93. Its firms were drawn from marketing lists, so the share is not a rate for US law firms at large. "Did not answer", "did not speak to a person" and "did not respond within seven days" are different measures, from different years and channels, so we do not average them or set them in one line.
In the tests by Clio† and Hennessey Digital† (500 to 1,333 firms), then, a large share of firms did not answer a call, an email or a web form, and Clio† reports that just 18% of firms that replied by email gave clear next steps or cost information90,91,93. Our argument from them is that a firm that has earned visibility or a referral (Chapters 1 and 2) has produced an inquiry, and an inquiry that meets silence wastes that work.
Speed of response: what the general lead research says, and what it does not
Oldroyd, McElheran and Elkington wrote in Harvard Business Review in 2011 that "Our research shows that most companies are not responding nearly fast enough"94. The article body is behind a paywall, so the detailed figures come from a news report that reproduces its text, and we give them as the article is reported.
The authors audited 2,241 US companies, measuring how long each took to respond to a web-generated test lead95. In the same report, 37% of companies responded within an hour, 16% within one to 24 hours and 24% took more than 24 hours, while 23% never responded at all95; the four shares sum to 100% (37% + 16% + 24% + 23%). Among companies that responded within 30 days, the average response time was 42 hours96.
A separate study cited in the same article covered 1.25 million sales leads at 29 B2C and 13 B2B companies, and found that companies that attempted to contact potential customers within an hour were more than 60 times as likely to qualify the lead as companies waiting 24 hours or longer95,96. The report adds that firms contacting potential customers within an hour of a query were nearly seven times as likely to qualify the lead as those that tried even an hour later95. A 2007 study by Oldroyd and InsideSales.com† found a 21-fold decrease in the odds of qualifying a prospect if the response time stretched from 5 to 30 minutes97, and a 100-fold increase in contact ratios from calling within 5 minutes versus 3097.
None of these studies, in the claims cited here, concerns law firms: the audited companies and the recipients of the leads were US companies95, and none of the claims says that law firms were among them. The comparisons are between companies, and a company that responds quickly may differ in other ways from one that responds slowly, so the ratios cannot be read as what speed would do for a law firm.
The mechanism is plausible for law, and the survey evidence is about stated behavior. In Clio†'s 2019 consumer survey, 57% of those who had ever shopped for a lawyer said they contacted more than one law firm29, and 42% said that if they like the first lawyer they speak with they will not need to speak with any others29. The 42% is a share who would stop looking, not a share who hire whoever answers first. In the same survey 10% expected a reply to a message within an hour, 24% within a few hours and 45% within 24 hours29.
People with a legal problem also move quickly. In a 2016 Insurance Research Council poll, nearly two-thirds of claimants who contacted an attorney about an auto injury claim did so within a week of the accident98. FindLaw† (legal marketing) reports that 56% of respondents to its 2024 survey took action toward their legal need within a week or sooner28. A measurement of the effect of response time on booking in law, dentistry, medicine or accounting is NF; the nearest evidence is the general lead research above and the stated behavior in the surveys. That the first firm to answer is likely to have the conversation is therefore our argument, and we expect it to hold more strongly for urgent matters than for planning work.
Rule 1.18 and why law firm intake collects little
ABA Model Rule 1.18 treats a person who consults a lawyer about the possibility of forming a client-lawyer relationship with respect to a matter as a prospective client. It bars the lawyer from using or revealing information learned from a prospective client even when no relationship ensues, subject to Rule 1.9's exceptions for former clients4. The rule is a model that each state adopts in its own version.
Intake design has to manage a tension. Intake wants information, to reach the person, route the matter and check conflicts, and each fact collected from a person who never becomes a client may be information the rule speaks of. We think the sensible design collects four things: a way to reach the person, the kind of matter, whatever the firm's lawyers decide a conflict check needs, and a booked time. Everything else waits for the consultation.
Software sharpens the question of who counts as consulted. Florida Ethics Advisory Opinion 24-1, which the Florida Bar's Board of Governors voted unanimously to approve on January 19, 202499, says website chatbots used for client intake present additional risks, including that a prospective client relationship or even a lawyer-client relationship has been created without the lawyer's knowledge100. We read the opinion from a copy hosted by a law school, because the Bar's own page showed no text. It is an advisory opinion and Florida's. Whether an intake call to a staff member or a voice agent is a consultation under a state's version of Rule 1.18 is for counsel.
Law firm website design as intake design
"Law firm website design" is searched about 1,600 times a month in the US38, the most of any legal phrase in Table 1. A website built as a billboard ends at the phone number. A website built as part of an intake system ends in a handoff: a consultation booked, a message routed to a named person, or a clear decline. The second takes more work at the start, because the framing and plumbing come before the finish: decide what the firm takes, who answers, what is collected and where it goes, then design the pages that carry a visitor through that.
A visitor to a law firm's site should be able to do five things without searching:
- Tell from the first screen whether the firm handles this kind of matter and serves this place.
- Call, from a phone, with one tap on a number that a person or the agent answers.
- Book a consultation in a slot the firm has opened for that kind of matter.
- Send a message through a form whose fields the firm chose and whose destination someone watches.
- Learn what the firm does not take, so those visitors do not spend the firm's time.
The form is where the Rule 1.18 tension shows. Chapter 3 described Rule 7.2(d), on a responsible lawyer's name and contact information3, and Rules 5.3 and 1.6(c), on nonlawyer assistance and client information64,80, which a firm may weigh for the vendors that host a form. The fields are the firm's choice, and we offer a minimum: a name and a way to reach the person, matter type from a list the firm wrote, and a preferred contact time. Add a free-text box only if the firm's lawyers want a narrative, since a narrative is where the most information enters, and place near the fields a notice, drafted by a lawyer, about what sending the form does and does not do.
The destination matters as much as the fields, because a form that sends to an inbox no one watches gives a visitor silence. Clio†'s shop found that 33% of firms responded to an email90. Hennessey Digital†'s 2025 test of online lead forms found that 74% responded within seven days, and the study states that 26% of law firms do not respond to online leads93. Neither says what became of the form on the firm's side. We think the form's route should be written down: where the submission goes, who is alerted, how fast a first reply is expected and who covers when the person on duty is away.
The legal answering service and the AI receptionist: what a voice agent may do and what stays with a lawyer
A legal answering service picks up a firm's calls when no one at the desk can, collects the details the firm chose and passes them on. People can staff one, and so can software. We think what stays with a lawyer, whoever or whatever answers, is everything that calls for judgment about the law or the matter: whether the caller has a case, whether the firm will take it, what an existing matter looks like, what a fee will be beyond what the firm has published, and whether a conflict exists. The front desk in depth covers the general case across professions.
Florida Opinion 24-1 also says generative AI chatbots that communicate with clients or third parties must comply with restrictions on lawyer advertising and must include a disclaimer indicating that the chatbot is an AI program and not a lawyer or employee of the law firm100. Whether a telephone voice agent is a chatbot in the opinion's sense is for counsel. We think the script should be reviewed like the firm's advertising, and its first sentence should say what the agent is.
We think a firm should plan as though it will answer for what its agent says. The nearest case cited here is not about lawyers. In Moffatt v. Air Canada, the Civil Resolution Tribunal found Air Canada liable for negligent misrepresentation because its website chatbot told a customer he could apply for bereavement fares retroactively101. The tribunal wrote that "It should be obvious to Air Canada that it is responsible for all the information on its website. It makes no difference whether the information comes from a static page or a chatbot"101. For a law firm the case is an analogy and not a holding.
Call recording, AI disclosure and the AI voice: the texts
Three families of texts bear on a voice agent on a firm's line: recording laws, AI-disclosure statutes and the federal ruling on AI voices. Table 3 sets a few texts of each kind side by side with a limit on each. It does not cover the law of every state.
| Text | What it says, as the claim prints it | A limit on what we say |
|---|---|---|
| 18 U.S.C. 2511(2)(d) | A private person may intercept a wire, oral or electronic communication where the person is a party to it or one of the parties has given prior consent, unless the purpose is to commit a criminal or tortious act102 | Federal; the next two rows are state texts that speak of all participants or parties |
| RCW 9.73.030(1), Washington | It is unlawful to intercept or record a private communication transmitted by telephone or other device without first obtaining the consent of all the participants103 | One state; others differ |
| California Penal Code 632(a) | Punishes a person who, intentionally and without the consent of all parties to a confidential communication, uses an electronic recording device to record it, including by telephone; the punishment includes a fine not exceeding $2,500 per violation or imprisonment, or both, with a cap of $10,000 per violation for a person previously convicted under certain sections104 | One state; we do not set out the statute's definition of a confidential communication |
| FCC Declaratory Ruling 24-17 (February 2024) | Confirms that the TCPA's restrictions on an "artificial or prerecorded voice" encompass current AI technologies that generate human voices, so such calls require the called party's prior express consent to initiate, absent an emergency purpose or exemption70 | Speaks of calls initiated with an AI voice; whether it reaches an agent that answers an inbound call is for counsel |
| California Business and Professions Code 17941 | Makes it unlawful to use a bot to communicate or interact with another person in California online with the intent to mislead the person about its artificial identity, to knowingly deceive about the content of the communication to incentivize a purchase or sale of goods or services or to influence a vote; a person using a bot is not liable under that section if it discloses that it is a bot104 | Speaks of a bot "online"; whether a telephone agent is covered is a legal question our source does not answer |
| Utah Code 13-77-103 and 13-77-104 | A supplier using generative AI in a consumer transaction must disclose that the person is dealing with AI and not a human if the person asks; a regulated-occupation provider must prominently disclose AI use in a high-risk interaction, verbally at the start of a verbal interaction; clear and conspicuous disclosure at the outset and throughout gives a safe harbor105,106. The high-risk definition includes personalized advice that could reasonably be relied upon to make significant personal decisions, including legal advice or services107 | A regulated occupation is one regulated by the Department of Commerce that requires a license or state certification107; the page does not say which professions that covers, so we do not say whether it reaches lawyers |
Table 3. Recording and AI-disclosure texts that bear on a voice agent, with a limit on each.
The rule depends on the state: the federal text turns on the consent of one of the parties102, and the Washington and California texts on the consent of all participants or parties103,104. In our service, whether calls are recorded is the firm's decision, and the agent's opening sentence, which names it as an AI assistant, is the natural place for any recording notice that applies.
Law firm CRM and legal intake software as record keeping
We use "CRM" to mean any system, from a spreadsheet to purpose-built software, in which a firm records each person who contacts it and what happened. Without one a firm cannot say how many inquiries arrived, how quickly each was answered, which source sent it or why a person was declined, and the measurement in Chapter 6 depends on those answers. This paper does not measure whether any CRM or intake software improves a firm's results; Table 1 shows only that the tools are searched for.
The fields we would keep are few, and each answers a purpose:
- Date, time and channel of first contact, and the time of the first reply, so that the number the secret shop measured is the firm's own.
- Source of the inquiry (search, referral, directory, lead generator, advertisement, an AI assistant's answer), so that inquiries can be counted by source.
- Matter type, picked from the firm's list, so that routing and declines follow a rule.
- Who handled it, and the status of the conflict check, set by whoever the firm's lawyers designate.
- Outcome: consultation booked, referred out, declined or signed, and for a decline the reason.
- Any outbound contact made to the person, with its trigger, so that Chapter 3's Rule 7.3 log is the same record.
The record holds information from people who may never become clients, which is the kind of information Rule 1.18 speaks of4, and it may sit with a vendor, where Rules 1.6(c) and 5.3 speak of reasonable efforts and supervision64,80. Whether and how they apply is open. We think the record belongs in the vendor inventory from Chapter 3, with the same questions about access and deletion. Measurement and attribution for a small practice treats what a small firm can learn from such a record.
Our own agent: what it does, what it will not do and what it costs
Our service is an AI voice agent. It tells callers at the start that it is an AI assistant, collects the details the firm decides on, such as a name, a number and the matter type, books consultations in slots the firm opens, takes messages for the lawyer or paralegal on a file, and transfers an urgent call, such as a hearing, an arrest or a filing deadline, to the lawyer the firm names. It never gives legal advice, never says whether a caller has a case and never accepts a matter for the firm. It does not confirm who the firm's clients are or discuss an existing matter. A caller who seems to be in danger hears a safety line, approved by the firm, that says to dial 911 before anything else.
Setup is the firm's work as much as ours: the firm lists the intake fields, the practice areas it takes and the matters it turns away, and a lawyer at the firm reviews the script before it goes live. We test it with the firm's common and difficult calls and, once it is switched on, go through the call summaries with a lawyer at the firm. The agent answers inbound calls.
The price is $299 a month with 250 call minutes included, then $0.39 a minute39, added to a plan. Plans are $999, $1,999 and $2,999 a month per location, or $9,990, $19,990 and $29,990 for a 12-month term paid in full, two months free. The $2,999 plan carries a $3,000 setup fee, $1,400 on a 12-month term, and the two lower plans a one-time onboarding fee of $499 and $999, waived on a 12-month term39. Our page on the legal answering service lists the call types, and the pricing page lists the plans.
This paper does not measure how our agent performs, and we make no claim that it raises bookings. A randomized or field study of an AI agent answering inbound calls in a legal, dental, medical or accounting setting is NF, and so is a peer-reviewed containment rate for inbound calls. The nearest peer-reviewed figure is for different work: at one hospital, an AI voice assistant making outbound calls before a procedure completed 86.4% of its calls in the study's first phase108, which is a guide and not a measure of a law firm's front desk. We think the agent adds little for a firm whose phone a person already answers every time, and a caller who needs a human at once is the reason the transfer path is part of the script.
A break-even for an answering agent at a law firm: a worked example from cited fees
To see what the agent must produce to cover its cost, a firm needs the cost and what a collected hour or a signed matter brings. The agent costs $299 a month with 250 call minutes included, which is about 62 four-minute calls (250 / 4), and $0.39 a minute after that39. The example takes the monthly cost as $299 for a firm that already pays for a plan, which holds while the agent's calls stay inside the included minutes; a four-minute call past them costs $1.56 (4 x $0.39). A firm with no plan takes the lowest, $999 a month39, for a cost of $1,298 a month ($999 + $299). A year of the first comes to $3,588 (12 x $299), and a year of the second to $15,576 (12 x $1,298).
Each fee is matched to what it buys. The plan produces inquiries and the agent answers them, so the plan is set against new matters and the agent against answered calls and the hours they bring. A ledger for the agent alone leaves the plan out of both sides, and one that charges the whole plan to the agent counts the cost of the demand and drops the demand. Table 4 therefore prices two purchases: the agent added to a plan the firm already has, set against the calls it answers, and the lowest plan plus the agent, set against the matters the plan brings and the agent catches.
For an hour, the State Bar of Michigan's 2023 survey of 2,191 attorneys found a median billing rate of $300 an hour18; the survey is Michigan's and self-reported. In Clio†'s 2025 data the average realization rate was 88% (the share of billable work invoiced) and the average collection rate 93% (the share of invoices paid)19. An hour billed at $300 therefore brings in $245.52 ($300 x 88% x 93%).
For a matter, a Spring 2020 survey by the Washington State Office of Public Defense found an average attorney cost of $3,500 for a DUI, excluding investigators and experts15; it is one state's survey of private defense attorneys. For an injury case, the Insurance Research Council reports that average auto bodily-injury claim payments rose to more than $20,000 in 202216, and the State Bar of California says a personal injury contingency fee will often be one-third17. A one-third fee on $20,000 would be ~$6,667 (from the two sources above; the payment figure is what insurers paid, not a lawyer's revenue).
| What is bought | Monthly | A year | Collected hours a month to cover it | DUI matters a year | Injury cases a year |
|---|---|---|---|---|---|
| Agent added to a plan the firm already has | $299 | $3,588 | 1.2 | 1.0 | 0.5 |
| Lowest plan plus agent | $1,298 | $15,576 | 5.3 | 4.5 | 2.3 |
Table 4. What each purchase must produce to cover its cost (calc.): collected hours at $245.52 an hour, DUI matters at $3,500 and injury cases at $6,667.
The lowest plan plus agent is the larger bar, and a firm that takes few DUI matters in a year relies on the plan's own demand to supply them. Divorce and estate-plan fees are not in the table: survey-based fees for divorce, wills and probate matters are NF from a source we could open; the nearest evidence is the hourly rate in Clio†'s data, an average of $343 for family law lawyers and $370 for wills and estates19. For those practice areas the hours column is the one to read, but 59% of firms billed flat fees alone or beside an hourly rate in 202419, so an hourly rate is a benchmark and not what a client pays.
A firm is seldom short of hours. In Clio†'s 2025 data the average law firm utilization rate, the share of an eight-hour day put toward billable work, was 38%, so five hours of a lawyer's day go unbilled19; a lawyer at 38% bills 3.0 of the eight hours (38% x eight hours). In a 2023 Clio† report the average for solo lawyers in 2022 was 25%, about two hours of billable work a day20. Unbilled hours are not all idle: in the State Bar of Michigan's survey, full-time private practitioners reported a median of 10 hours a week on non-billable activities18. Even so, the figures say the limit for these firms is billable work and not lawyer time, which is why we think the inquiries that already arrive are the first place to look: a matter caught is worked in hours the firm already pays for.
A DUI matter at $3,500 is 14.3 hours billed at $245.52 collected an hour ($3,500 / $245.52). The 4.5 matters in the last row of Table 4 therefore take about 63 such hours in a year (calc.), which is about thirteen days of one lawyer's unbilled hours at five a day. We estimate that as the room, from the utilization rate above, and it likely errs high for a matter that takes more lawyer time than its fee pays for at the median rate.
A full schedule changes the form of the gain and does not remove it. A firm with more inquiries than hours can choose better-paying matters, raise its minimum fee or add a lawyer, and the agent's cost is then covered by better matters rather than additional ones.
Two links in the chain are unmeasured, and the example solves for them. The first is how many calls and forms a firm receives. A measurement of law firm call volume is NF; the nearest evidence is for other professions. Patient Prism† (dental call analysis) reports 2025 data that work out to about 116 patient calls per location a month, and Call Box† (call tracking and answering) reports 2020 data on health care practices that give 299 to 505 calls per office location109,110. We carry neither over to law, so the example counts matters and not calls.
The second link is the share of answered inquiries that sign a client. We estimate that it is no higher than about one in two and probably well below. The basis is the adjacent professions. Patient Prism† classed 34 of every 100 patient calls as real booking opportunities and 21 as booked, which is 62% of the opportunities (21 / 34)109, and Invoca† (call tracking) reports that 45% of the answered calls that were leads converted on the call111. Neither is a signing rate, the two measures differ and are not averaged, and a consultation must still become a signed engagement, so the estimate likely errs high for law.
Table 5 shows the inquiries a year that must be answered and signed as DUI matters to produce the counts in Table 4, at three values of that share with one in two as the ceiling. At the $6,667 injury value the counts fall to about half ($3,500 / $6,667).
| Share of answered inquiries that sign | Agent added to a plan the firm already has | Lowest plan plus agent |
|---|---|---|
| 1 in 10 | 10.3 inquiries a year | 44.5 inquiries a year |
| 1 in 4 | 4.1 inquiries a year | 17.8 inquiries a year |
| 1 in 2 | 2.1 inquiries a year | 8.9 inquiries a year |
Table 5. Inquiries a year that must be answered and signed as DUI matters to cover each purchase (calc.): $3,588 or $15,576 divided by $3,500, then by the share that sign. The shares are a range, and we estimate the ceiling at one in two. For the lowest plan plus agent the inquiries can come from the plan's new demand as well as from calls the agent catches.
Neither Table 4 nor Table 5 prices errors or the lawyer's time to review the script and the call summaries. The margin an error would have to use up is the fees from matters above the counts shown: each DUI matter beyond the 1.0 in the first row of Table 4 adds $3,500, and an error would have to cost the firm more than all of those fees to turn the purchase negative. We think the review time comes out of hours that the average firm in Clio†'s data has unbilled.
The system loses money in one corner: a firm that is not growing and whose intake is already sufficiently optimized, with nearly every call and form answered in time. That corner is the exception and not the base case. In Clio†'s 2024 shop 40% of the 500 firms answered the call, so 60% did not (calc., above)90, and in Hennessey Digital†'s 2025 test 26% of 1,333 firms did not respond to an online lead93, so the firms in those vendor samples were a long way from it.
The count that settles the question for one firm is in its own log: how many new-matter inquiries went unanswered last month, and how many of those signed. Neither vendor sample is a rate for that firm, and Chapter 6 shows how to keep the count. The full law firm ledger works the numbers at greater length, and AI risk and law sets out the cost of error that Tables 4 and 5 omit.
Chapter 5
Legal AI for law firms: what the evidence shows and where it pays
The phrase legal ai is searched more often than any law firm marketing phrase in our keyword data38. This chapter recommends no tool.
Legal AI, AI for lawyers and AI for law firms: what is searched
In our keyword data, legal ai averages about 18,100 US searches a month, ai for lawyers about 2,400 and ai for law firms about 1,900 (Google Ads keyword data, November 2026)38. The most searched law firm marketing phrase in the same data is law firm seo, at 3,600 a month38. Legal ai is searched roughly five times as often (18,100 / 3,600, about 5).
The usual limits on search figures apply. The phrases are ones a person in the trade would type, and we cannot tell whether a search came from a lawyer, a law student, a vendor or a client. Volume shows interest, not purchase. Cost per click is what advertisers pay. It does not show what a client is worth or whether a tool will pay: advertisers paid an average of $31.62 a click on legal ai and $58.21 on ai for lawyers38.
Who reports using AI in legal work, and why the surveys cannot be added together
Three surveys of legal work report AI use, and each counts a different thing (Table 1). We do not read them as a trend or as a count of how many firms or lawyers have adopted AI.
| Survey | Fieldwork | Who was counted | What counted as use | What it reported |
|---|---|---|---|---|
| Clio† (legal practice-management software), Legal Trends Report | June 5 to 23, 202490 | 1,028 U.S. legal professionals: lawyers and support staff such as paralegals and administrators90 | Using AI in some capacity in their practice90 | 79% in some capacity; 25% had adopted AI widely or universally90 |
| ABA, 2024 Legal Technology Survey | Not given (the 2024 edition) | 512 respondents answering the online research volume112 | Attorneys saying their offices were currently using AI-based technology tools112 | 30.2% of attorneys; 17.7% of solo practitioners; 47.8% at firms of 500 or more lawyers112 |
| Thomson Reuters Institute† (sells AI products), as reported by LawSites (secondary) | January and February 2025113 | 1,702 respondents across legal, tax, accounting, audit, corporate risk, fraud and government; legal professionals were 41% of them113 | Legal organizations actively using gen AI113 | 26% of legal organizations113 |
Table 1. Three surveys of AI use in legal work: who each counted, what counted as use, and what it reported. The figures are not comparable with one another.
The rows cannot be compared, because both the units and the questions differ. Setting Clio†'s 79% beside the 26% in the Thomson Reuters Institute† report compares different questions put to different people and supports no statement about the profession90,113.
Clio† says that when it last researched AI in 2023, 19% of law firms were using AI in their practice, and it sets that against the 2024 share of surveyed legal professionals90. The units differ, so we do not treat the pair as a rise. Comparisons inside one instrument are safer, because the question and the sample are the same. In the ABA survey, accuracy was the concern named most often, by 74.7% of those surveyed, ahead of reliability at 56.3% and data privacy and security at 47.2%112. For a small firm, the one useful detail is the ABA size split in Table 1: in its 2024 sample, solo practitioners reported less office use than firms of 500 or more lawyers did112.
A fourth instrument sits outside the legal sector. The Census Bureau's Business Trends and Outlook Survey (BTOS) found that overall business use of AI hovered between 17% and 20% in its December 2025 to May 2026 data114. It counts businesses of every kind, so it cannot stand in for a figure about a profession.
What legal AI gets wrong: measured error and court sanctions
The evidence on errors is of two kinds: measured rates in tests and orders in real cases. We use hallucination in the sense the National Institute of Standards and Technology gives confabulation in NIST AI 600-1: the production of confidently stated but erroneous or false content115. The AI-profit paper's chapter on risk, law and the cost of error takes the wider question.
Dahl, Magesh, Suzgun and Ho report in a 2024 preprint that legal hallucinations are "alarmingly prevalent," occurring between 58% of the time with ChatGPT 4 and 88% with Llama 2 when the models are asked specific, verifiable questions about random federal court cases116. That tests language models answering questions about cases, not any firm's workflow.
A second study tested tools sold for legal research. Magesh and co-authors found that the AI research tools made by LexisNexis (Lexis+ AI) and Thomson Reuters (Westlaw AI-Assisted Research and Ask Practical Law AI) each hallucinate between 17% and 33% of the time, less than general-purpose GPT-4 but contrary to the providers' claims of being hallucination-free117. The team also scored accuracy. Lexis+ AI was the highest-performing system tested, answering 65% of queries accurately, while Westlaw's AI-Assisted Research was accurate 42% of the time and hallucinated nearly twice as often as the other legal tools tested118. Accurate and hallucinated are separate measures, so the two sets of figures do not add to 100%. The benchmark held 202 queries, which its authors call small in comparison with other evaluations118, and the products were early-2024 versions, so it does not tell a firm what any product does today.
A third test came out differently. Vals AI, an AI-evaluation company, reports that in its October 2025 benchmark of 200 legal research questions all the AI products it tested scored 74% to 78% on an average weighted score, within 9 points of a lawyer baseline of 69% and above that baseline on every criterion119. It used different questions, scoring and products from the Stanford test. Table 2 sets the three tests side by side; they are not averaged, and none is a current error rate.
| Test | Date and size | What was tested | What it reported |
|---|---|---|---|
| Dahl and co-authors, a preprint | 2024; specific, verifiable questions about random federal court cases116 | General-purpose language models, ChatGPT 4 and Llama 2 among them116 | Hallucination between 58% (ChatGPT 4) and 88% (Llama 2)116 |
| Magesh and co-authors, Stanford | Early 2024; 202 queries118 | Legal research tools from LexisNexis and Thomson Reuters117 | Hallucination 17% to 33%117; accurate on 65% of queries (Lexis+ AI) and 42% (Westlaw AI-Assisted Research)118 |
| Vals AI, an AI-evaluation company | October 2025; 200 questions119 | A different set of AI products119 | Weighted score 74% to 78% against a lawyer baseline of 69%119 |
Table 2. Three tests of AI on legal research questions. Each used different questions, products and scoring, so the figures are not comparable.
Courts have responded to filings with fabricated citations. Table 3 gives three orders as the claims print them.
| Case | Court and date | What the order found | Sanction as printed |
|---|---|---|---|
| Mata v. Avianca | S.D.N.Y., June 22, 2023120 | Two lawyers and their firm filed non-existent cases generated by ChatGPT120 | $5,000 penalty, jointly and severally120 |
| Wadsworth v. Walmart | D. Wyo., order of February 24, 2025121 | Motions in limine cited nine cases, of which eight did not exist121 | One attorney's pro hac vice admission revoked and a $3,000 fine; two other attorneys fined $1,000 each121 |
| Johnson v. Dunn | N.D. Ala., July 23, 2025122 | Three attorneys confirmed that citations in two motions were ChatGPT hallucinations122 | Public reprimand, disqualification from the case, publication of the order and referral to the Alabama State Bar122 |
Table 3. Three court orders on fabricated citations, with the sanction each imposed.
The Mata order said "there is nothing inherently improper about using a reliable artificial intelligence tool for assistance," and that existing rules "impose a gatekeeping role on attorneys to ensure the accuracy of their filings"120. It found that the respondents abandoned their responsibilities by submitting non-existent judicial opinions with fake quotes and citations created by ChatGPT, and then continued to stand by those opinions after judicial orders called their existence into question120. In Johnson v. Dunn the court said fabricating legal authority is serious misconduct that demands a serious sanction122.
Three orders are examples, not a sample, and say nothing about how often such filings occur. Damien Charlotin's AI Hallucination Cases database, which tracks legal decisions where a court found or implied that a party relied on AI-hallucinated content, listed 2,289 cases when we read it on November 10, 2026, and its own text calls it a work in progress123. It is one researcher's running count, a secondary source that changes as cases are added, so we rest no conclusion on it.
At least one district has written disclosure into its local rules. The Northern District of Texas requires, under Local Civil Rule 7.2(f), that a brief prepared using generative AI disclose that fact on its first page under the heading "Use of Generative Artificial Intelligence"124. The district adopted the rule by orders of June 10, 2024, effective September 3, 2024125. California's guidance, covered next, says an attorney should check for rules or orders in the jurisdiction that require disclosure of AI use126.
What the guidance says: ABA Formal Opinion 512, California and Florida
ABA Formal Opinion 512, dated July 29, 2024, says lawyers using generative AI tools must fully consider their applicable ethical obligations, including their duties to provide competent legal representation and to protect client information5. The ABA Model Rules are a model and each state adopts its own version, so what follows describes the texts and does not say what any state requires; the firm's counsel reads the state's own rules.
On competence, Opinion 512 says lawyers need not become experts in a tool but must have a reasonable understanding of its capabilities and limitations5. Comment [8] to ABA Model Rule 1.1 says a lawyer should keep abreast of changes in the law and its practice, including the benefits and risks associated with relevant technology127. On verification, the opinion says a lawyer's reliance on, or submission of, a generative AI tool's output without an appropriate degree of independent verification or review could violate the duty of competent representation under Rule 1.15. It adds that uncritical reliance can result in inaccurate legal advice to clients or misleading representations to courts and third parties5.
On confidentiality, the opinion says that because many of today's self-learning tools are designed so that their output could lead directly or indirectly to disclosure of client information, a client's informed consent is required before information relating to the representation is input5. It says that adding general, boilerplate provisions to engagement letters is not sufficient to obtain informed consent5. On communication, it says client disclosure of AI use may be unnecessary depending on the circumstances, but that lawyers must disclose their practices if a client asks how the work was done or whether AI was used, or if the client requires disclosure5. For outside providers, it says lawyers have additional supervisory obligations when they rely on others outside the firm to employ AI tools in the representation5.
The State Bar of California issued its original practical guidance in November 2023. A 2026 version, approved by the State Bar Board of Trustees on May 14, 2026, replaces it and addresses agentic AI126. It describes two related duties of competence when using AI: baseline technological competence about the system's capabilities, data sources, limitations and risks, and independent professional judgment in reviewing, verifying and correcting outputs126. It says the duty of candor to the tribunal cannot be delegated to AI126, and that an attorney must consider disclosing to the client an intention to use AI in the representation, including how it will be used and its benefits and risks126. It also speaks to the kind of system a voice agent is, saying attorneys must not deploy agentic systems in a manner that allows the system to make substantive legal determinations, communicate legal advice, prepare and file pleadings, or otherwise act in a representative capacity without meaningful attorney supervision and review126.
Florida's Board of Governors voted unanimously on January 19, 2024 to approve Ethics Advisory Opinion 24-1, according to the Florida Bar News99. The opinion says lawyers may use generative AI in the practice of law but must protect the confidentiality of client information, provide accurate and competent services, avoid improper billing practices and comply with restrictions on lawyer advertising, and it opens by saying "Advisory ethics opinions are not binding"128. Check the Bar's own version of the text.
The Professional Ethics Committee of the State Bar of Texas issued Opinion 705 in February 2025 on the ethical issues a lawyer's use of generative AI raises under the Texas Disciplinary Rules129. It says that a lawyer should always verify the accuracy of any responses received from a generative AI tool129. The New York State Bar Association's Task Force on Artificial Intelligence issued a report in April 2024 that the Association's House of Delegates approved on April 6, 2024130. It is a bar-association policy report and not an ethics opinion, and it says the duty of confidentiality under Rule 1.6 also extends to what client information a lawyer may share when using certain generative AI tools130.
Table 4 sets the five texts' wording on client information side by side. We do not say which controls for any firm.
| Text | Wording on client information | What triggers it |
|---|---|---|
| ABA Formal Opinion 512 | Informed consent is required5 | Inputting information relating to the representation into a self-learning generative AI tool5 |
| State Bar of California, 2026 guidance | An attorney must not input any confidential client information, absent informed client consent126 | A generative AI solution that may present material risks to confidentiality or security126 |
| Florida Ethics Opinion 24-1 | Informed consent is recommended100 | Using a third-party generative AI program if that would involve disclosing any confidential information100 |
| Texas Ethics Opinion 705, February 2025 | Rule 1.05 will likely be implicated129 | The lawyer intends to provide client confidential information to the program129 |
| New York State Bar Association task force report, April 2024 | The duty of confidentiality also extends to what client information a lawyer may share130 | Using certain generative AI tools130 |
Table 4. How five texts word the treatment of client information in a generative AI tool. The New York text is a policy report and not an ethics opinion, and the California text replaced a 2023 version.
On insurance, ALPS, a legal malpractice insurer, published an article by its risk manager on August 21, 2025 saying that lawyers' professional liability policies typically do not exclude coverage for negligence claims arising from the use of generative AI, though coverage may depend on whether the conduct meets the policy's definition of professional services131. The same article says that if firm lawyers allow the AI to make critical legal judgments without attorney oversight, that could be viewed as the unauthorized practice of law, which such policies typically exclude131. It is an insurer's risk manager writing, not policy wording, and "typically" is his generalization, so a firm reads its own policy. A bar statement on malpractice coverage of AI errors is NF; the insurer's article is the nearest evidence. We think it is one more reason for a voice agent to make no legal judgments, which is how ours is built.
What studies of other work show, and where they stop
None of the productivity studies below tested lawyers. In an experiment on mid-level professional writing tasks, access to ChatGPT cut completion time by 37%, or 10 minutes, relative to a control group that took an average of 27 minutes132.
A field experiment with consultants separated tasks by whether they fell inside the technological frontier. For tasks inside it, consultants using AI completed tasks 25.1% more quickly and produced results of more than 40% higher quality than a control group133. For a task selected to be outside the frontier, consultants using AI were 19 percentage points less likely to produce correct solutions than those without AI133. A study of customer support workers found that access to a generative AI assistant raised productivity, measured as issues resolved per hour, by 14% on average, including 34% for novice and low-skilled workers but with minimal impact on experienced and highly skilled workers134. That assistant was a conversational tool used by customer support agents, not an AI agent replacing them and not a voice system135.
In an April 2025 working paper, Humlum and Vestergaard report that AI chatbots had no significant impact on earnings or recorded hours in any occupation, with confidence intervals ruling out effects larger than 1%136. The same paper reports modest productivity gains, with average time savings of 3%, combined with weak wage pass-through136. Time saved need not show up in earnings, which is the point Table 5 makes below for one firm.
Taken with the error tests in the previous section, the frontier result matters most for legal work. The tests point both ways: the Stanford team measured hallucination in tools that providers had called hallucination-free, and the Vals benchmark found products scoring above a lawyer baseline. We think legal research with citations still resembles a task near the edge of the frontier, because the errors are confidently stated115 and a reader cannot tell from the surface of an answer which ones are wrong. That is why the verification duty in the guidance matters. Brynjolfsson and Hitt (2000) add that the benefits of IT investment are often intangible and disproportionately difficult to measure137.
What vendor surveys report about AI and a firm's results
Clio† and Thomson Reuters† have published findings that link AI use to firm results. Both sell to law firms, both report associations between groups that described themselves, and neither shows what a tool did. Clio†'s 2025 press release says firms with wide AI adoption are nearly three times more likely to report revenue growth than firms that have not adopted AI138. Growing firms may also be the firms that adopt more, and the comparison cannot separate the two. The same release says 77% of the firms that increased revenue with AI attributed it to improved operations such as document generation, workflow automation and client communication138. The base is firms that said they increased revenue with AI, and the attribution is their own.
Thomson Reuters† publishes similar findings from a separate survey, its 2025 Future of Professionals study of 2,275 professionals worldwide. More than half (53%) said their organizations are already seeing ROI as a direct or indirect result of AI adoption, and 22% said their business has a visible, defined AI strategy139. Organizations with visible AI strategies were twice as likely as those with informal or ad hoc adoption to experience revenue growth as a direct or indirect result139. The respondents also predicted that AI would save professionals an average of 5 hours a week over the next year139. That is a prediction and not a measured saving, and the sample is global rather than small-firm lawyers. As with Clio†'s, the figures show an association and do not say which comes first.
Set against the labor-market paper above, we think the surveys are a reason to run a small pilot and log it, as the last section describes; they give no basis for expecting a particular return.
Where legal AI pays in a small law firm: intake before drafting
We use the profit identity from the AI-profit paper's theory chapter in a plain form here: a use of AI pays when the revenue it adds or protects, less the cost of running it, checking its work and carrying its risk, is positive. Drafting and research carry a checking cost that the guidance makes explicit and a measured error rate, while intake addresses a different loss, the inquiry that never reached a lawyer. We think a small firm should look at capture at intake before AI-assisted drafting.
The first reason concerns fees. Opinion 512 says that generative AI tools may provide lawyers a faster and more efficient way to render legal services, but that lawyers who bill clients an hourly rate for time spent on a matter must bill for their actual time5. It says a lawyer may not charge a client to learn how to use a generative AI tool the lawyer will regularly use for clients5. It also says that to the extent a tool functions like equipping and maintaining a legal practice, a lawyer should consider its cost to be overhead and not charge the client for it absent a contrary disclosure in advance5. Texas Opinion 705 says a lawyer billing hourly may charge for the actual time spent using a generative AI program but may not charge hourly fees for the time that was "saved" by using it129, and Florida's opinion says that, though such programs may make a lawyer's work more efficient, the increase in efficiency must not result in falsely inflated claims of time128.
Take an hour at the median rate. The State Bar of Michigan's 2023 survey of 2,191 attorneys found a median billing rate of $300 an hour18; the survey is Michigan's and self-reported. Clio†'s 2025 data put the average realization rate at 88% and the average collection rate at 93%19. An hour billed at $300 then brings in $245.52 ($300 x 88% x 93%). Under hourly billing at actual time, an hour that AI-assisted drafting takes out of a task is an hour not billed. Nobody has measured how many hours a task loses, so Table 5 shows a range of one to four hours.
| Hours AI-assisted drafting removes from one task | Billed fees not earned at $300 an hour | Collected fees not earned at $245.52 an hour |
|---|---|---|
| 1 | $300 | $245.52 |
| 2 | $600 | $491.04 |
| 3 | $900 | $736.56 |
| 4 | $1,200 | $982.08 |
Table 5. Fees a lawyer does not earn on one task billed hourly at actual time, for each hour that AI-assisted drafting removes (hours x $300 and hours x $245.52). The hours are a range for an unmeasured link.
Each hour removed costs the firm its fee unless the hour is filled with billable work. In Clio†'s data the average lawyer bills 38% of an eight-hour day19, or 3.0 hours (eight hours x 38%), so hours are seldom what a law firm lacks. A firm that charges a fixed fee for the task faces a different calculation: Clio†'s data show 59% of firms billing flat fees alone or beside an hourly rate in 202419. The opinions cited here speak to hourly billing, and counsel answers what they say about other arrangements. A tool that saves lawyer time in drafting does not by itself add revenue to a firm billing hourly at actual time. It adds revenue when matters arrive to fill the hours.
The second reason concerns where the freed hours would come from. Filling them needs more matters, and more matters need more inquiries that reach a lawyer. Clio†'s secret-shopper study points to losses at that stage. In its 2024 study of 500 law firms, only 40% of firms answered the phone call, down from 56% in 2019, and 52% either answered or called back90. Clio† does not say how it drew the 500 firms. Chapter 4 weighs that study and the other tests.
We estimate, as Chapter 4 does from the adjacent professions, that no more than about one in two law firm inquiries signs a client (likely high). An unanswered inquiry is then worth at most about half a matter's fee: $1,750 for a DUI matter at $3,500 ($3,500 / 2), and about $3,300 for an injury case at $6,667 ($6,667 / 2). Chapter 4's Table 4 shows that the lowest plan plus the agent cost 5.3 collected hours a month. An unanswered inquiry cannot become a client, and a tool that answers it carries a different risk from a tool that drafts a brief.
Intake carries a risk of its own. Florida's opinion says website chatbots used for client intake present additional risks, including that a prospective-client relationship, or even a lawyer-client relationship, has been created without the lawyer's knowledge100. Chapter 4 takes up that question under Rule 1.18. Outside the law of lawyers, the Civil Resolution Tribunal found Air Canada liable for negligent misrepresentation in Moffatt v. Air Canada, 2024 BCCRT 149, because its website chatbot told a customer he could apply for bereavement fares retroactively101. The tribunal wrote that the airline is "responsible for all the information on its website"101. It is one tribunal's decision, and we cite it only as a warning about statements made by a tool.
Our design intent for a law firm's line is that the agent tells the caller it is an AI assistant, collects the intake fields the firm chose, books a consultation or takes a message, transfers an urgent call to the lawyer the firm names, and gives no legal advice. Whether a given script is acceptable is for the firm's counsel.
The full ledger for a law firm, with every input shown, is in the AI-profit paper's chapter on dental, medical, legal and accounting practices. The evidence for this ordering is thin. A randomized or field study of an AI agent answering inbound calls at a law firm is NF; the nearest peer-reviewed figure, Chapter 4's outbound hospital calls, is for different work. What an unanswered inquiry costs a firm rests on the one-in-two estimate above.
What stays with a lawyer
The texts above place several duties on the lawyer, not the tool: verifying anything filed or sent5,120, candor to the tribunal126, the decision to put client information into a tool and the client's informed consent (Table 4), what to tell the client126 and the fee5,129. Supervision of the people and vendors who run a tool for the firm is another. Opinion 512 says lawyers have additional supervisory obligations when they rely on others outside the firm to employ AI tools5, and Model Rule 5.3 requires a partner or lawyer with comparable managerial authority to make reasonable efforts to ensure the firm has measures giving reasonable assurance that a nonlawyer behaves compatibly with the lawyer's professional obligations80. This describes what the texts say and is not a checklist of what any firm must do.
Chapter 3 sets out the texts on communications about a lawyer's services. How they reach a page or profile that a tool drafted, and whether anything described on our pages on AI for law firms and the AI receptionist for law firms fits a firm's intake rules, are for counsel.
What we do not know about legal AI
Three questions stay open. Beyond the ABA survey's size split (17.7% office use among solo practitioners against 47.8% at firms of 500 or more lawyers112), this chapter does not measure AI use in small law firms. It does not measure the time lawyers spend verifying AI output either. The tests and the guidance set the direction: a tool that was accurate on 42% to 65% of queries in one test118 leaves the lawyer to check each answer, and checking time takes away from any saving, such as the 5 hours a week that respondents to Thomson Reuters†' survey predicted139. Nor does it measure the effect of AI on a firm's revenue: the Clio† and Thomson Reuters† surveys report associations, and the labor-market paper found no significant effect on earnings.
The error tests are from 2024 and 2025 and are not current error rates. The guidance is advisory and varies by state, and California's replaced an earlier version in 2026, so any of these texts may have changed since we read it. What would settle the question for one firm is its own pilot: a logged count of hours saved, hours spent checking, errors caught before filing, and inquiries reached, kept for long enough to show more than noise. Chapter 6 shows how to keep such a count and how much a small number can tell a firm.
Chapter 6
Measuring law firm marketing, what small numbers cannot show, and a first 90 days
A law firm that wants to know whether its marketing works faces two problems. Even major retailers and brokerages, in field experiments, found advertising returns hard to measure140. And a small firm's counts are small. This chapter covers both, then gives an order of work for the first 90 days that speaks to the reader directly.
Law firm marketing ROI: why returns are hard to measure
Lewis and Rao (2015) report on twenty-five large field experiments with major US retailers and brokerages, most reaching millions of customers, and conclude that measuring the returns to advertising is difficult140. The median confidence interval on return on investment across their experiments is over 100 percentage points wide140. The authors give volatility as the reason: individual-level sales are very volatile relative to the per capita cost of advertising, so informative experiments can easily require more than 10 million person-weeks140.
Observational methods, which a firm would use if it could not run an experiment, have their own problem. Lewis and Rao conclude that selection bias, due to the targeted nature of advertising, is a crippling concern for widely employed observational methods140. Gordon and co-authors compared experiments with observational methods using 15 US advertising experiments at Facebook, comprising 500 million user-experiment observations and 1.6 billion ad impressions141. They found that the observational methods often fail to produce the same effects as the randomized experiments, even after conditioning on extensive demographic and behavioral variables141.
The clearest single case is eBay's. Blake, Nosko and Tadelis turned off brand-keyword paid search in an experiment, and 99.5% of the forgone click traffic was immediately captured by natural, unpaid search traffic from the platform142. Typical OLS methods had given a return on investment of over 4,100% without time and geographic controls and over 1,400% with them, while the experiment found a return of -63%, with a 95% confidence interval of -124% to -3%142. The authors report that brand-keyword ads had no measurable short-term benefits142.
None of these results is about a law firm. What may transfer is the mechanism: a paid ad shown to someone who typed the firm's own name reaches a person who already had the name, perhaps from a referral, and the ledger below can show how many inquiries arrive that way. The size of any effect may not transfer, and this paper cites no test of whether a law firm's brand searches behave as eBay's did.
Attribution rules add a second problem. Berman (2018) shows, in a game-theoretic model of multi-publisher advertising, that the popular last-touch method over-incentivizes ad exposures, often resulting in lower advertiser profits143. That is a theoretical result under the paper's assumptions, not an empirical estimate of last-click error. An estimate of how far last-click attribution misstates a firm's results is NF; the nearest evidence is the eBay and Facebook comparisons above, which show that one method can overstate a return by a wide margin in a given setting and give no general rate. Credit given to the last thing a prospective client clicked is a convention, and a ledger that asks the client how they found the firm records something the click trail cannot.
A small firm should still measure, but the claim it can support is smaller: a count shows what happened, in what order and from what source, and does not by itself show what a given advertisement caused. A wide interval around a return is compatible with a return of zero and with a large positive one, and the same holds for a firm's own counts. The long paper's chapter on measurement and attribution for a small practice takes the question further.
What a law firm can count without statistics: a ledger in a spreadsheet or a law firm CRM
A firm can count seven things without a statistician (Table 1). Each answers one question and can live in a spreadsheet or a law firm CRM. The first three move in the first month, which matters two sections on. Chapter 4 treats CRM and intake software as record keeping, and the long paper's chapter on the front desk, response and the AI receptionist covers response time in depth.
| Count | The question it answers | How to record it | Where it misleads |
|---|---|---|---|
| Calls answered and calls missed | Of the calls that reached the firm, how many were answered by a person or the agent? | The phone system or agent log: calls in, calls answered, calls that rang out or went to voicemail, and whether a message was returned | Counts the answer, not what was said |
| Inquiries replied to, and time to first reply | How many inquiries got a reply, and how long did a person wait? | Minutes from the inquiry to the first reply by a person or an agent, by channel: call, form, email, chat | Counts the reply, not its quality |
| Consultations booked and held | How many inquiries became a conversation with a lawyer? | Date booked, date held, and no-shows | Many bookings and few held consultations point to scheduling, not reach |
| Inquiries by source | Where did the people who contacted the firm come from? | Ask each person how they found the firm and pick from a fixed list; add a line for the name of a former client who referred them | A person may name the last thing they saw, not the first |
| Matters opened (signed) | Which inquiries became clients? | Source, practice area and date signed | Small numbers; see the next section |
| Declined matters and the reason | Were the inquiries a fit? | One category from a fixed list: outside the practice area, outside the area served, conflict, no capacity, not a matter the firm takes | Measures the audience, not the quality of the response |
| Fees collected per matter, by source | What is a matter worth, by source? | Add after the matter closes, or at fixed intervals | Lags by months or years |
Table 1. The seven counts of a law firm marketing ledger. The table holds no figures, so it carries no citations.
Ask the source question the same way every time. We think the most dependable record is one the person taking the inquiry fills in at the time; a website analytics report is a second source, not a replacement. In Pew Research Center's analysis of US adults' browsing data, users who encountered a Google AI summary clicked a traditional search result link in 8% of visits, against 15% of visits without an AI summary47. That finding says nothing specific about local or professional-services queries. We infer it still gives a reason to doubt that sessions capture every way a person came to know a firm, since a person who reads an answer may not click.
Keep the ledger thin. ABA Model Rule 1.18 treats a person who consults a lawyer about the possibility of forming a client-lawyer relationship as a prospective client, and bars the lawyer from using or revealing information learned from that person even when no relationship ensues, subject to Rule 1.9's exceptions for former clients4. We design a ledger to hold the source, the dates, a category for the reason and the outcome, and not the facts of the person's matter; the firm's counsel decides what a given log may hold. Chapter 4 covers the rule.
A tool that records calls or holds inquiry details is a vendor that sees client-related information. Model Rule 1.6(c) requires a lawyer to make reasonable efforts to prevent the inadvertent or unauthorized disclosure of, or unauthorized access to, information relating to the representation of a client64. Rule 5.3 requires a partner or lawyer with comparable managerial authority to make reasonable efforts to ensure the firm has measures giving reasonable assurance that a nonlawyer retained by or associated with the lawyer behaves compatibly with the lawyer's professional obligations80. Chapter 3 sets out the texts on vendors.
What small numbers cannot tell a firm
Counts of signed matters are small, and small counts move for no reason. Table 2 shows how far. Suppose signed matters arrive independently at a steady average rate. This simplifies real months, which also move with seasons and with marketing. It is the Poisson model of a count, under which the standard deviation of a month's count is the square root of its average. The usual rule of thumb puts about 95 of every 100 months within two standard deviations of the average; computed exactly for the Poisson model, the ranges in Table 2 hold 98, 97 and 97 of every 100 months (calc.). Firms differ in how many matters they sign, so Table 2 shows three averages and a firm substitutes the average from its own ledger.
| Average signed matters a month | Two standard deviations | Range a month can span by chance alone |
|---|---|---|
| 4 | 4 | 0 to 8 |
| 9 | 6 | 3 to 15 |
| 16 | 8 | 8 to 24 |
Table 2. The range a month's count of signed matters can span by chance alone, if matters arrive independently at a steady average rate. The three averages are a range for a quantity that differs by firm; two standard deviations is two times the square root of the average.
Take the middle row. On that assumption, a firm that averages 9 signed matters a month should expect months as low as 3 and as high as 15 with nothing about its marketing changed. Suppose a change adds 2 matters a month. To compare two averages of n months each, at 9 a month, we use the standard error of their difference, which is the square root of (18 / n), with the normal approximation to the counts. A difference of 2 reaches two standard errors only when that value falls to 1, which happens at n = 18: about 18 months of the old way and 18 of the new, with nothing else changing, which a practice would find hard to hold constant. Even then a true difference of 2 would only just reach the threshold, so about half of such comparisons would fall short of it.
The same arithmetic applies to a purchase. In Chapter 4's break-even, the lowest plan plus the agent is covered by 4.5 DUI matters or 2.3 injury cases a year, a little over a third of a matter a month (4.5 / 12 = 0.375). Against months that swing by six either way, no comparison of total signed matters will show an effect that small, though the effect may still be there. The total is the wrong instrument for it, and the next section says what to use.
A small count can catch large effects. A source that produced no inquiries in three months, a first reply time measured in days, or a block of declined matters that share one reason will show up in a ledger without statistics. A rise in signed matters from 9 to 10 will not. We suggest acting on what the ledger shows is broken and leaving alone what chance could explain. Because the spread of a count shrinks relative to its size as the count grows, a quarterly reading is steadier than a monthly one. Suppose the average is 9 a month, or 27 a quarter. Two standard deviations is then about 10 on 27, or 38% of the average, against 6 on 9, or 67% (calc.).
What a firm can count from the first month, and what an hour and a matter are worth
A firm can start judging a purchase in the first month, because the counts upstream of signed matters are no smaller than the count of signed matters and move first. Every signed matter began as an inquiry, and every inquiry began as a call, a form or a message. Calls answered, inquiries replied to and consultations booked change the week a firm changes how it answers, and an answering agent or a plan changes them directly. Matters opened and fees collected arrive later, while the cost of the system starts on day one. The judgment therefore runs in two steps.
The first starts in week one. Record the share of calls answered, the share of inquiries replied to and the time to first reply, and compare each with the same count before the change. Outside figures are reference points, not benchmarks: Clio† (legal practice-management software) found in its 2024 shop that 40% of 500 firms answered the call90, and Hennessey Digital† (law firm marketing) found in its 2025 test of 1,333 firms' web forms that 74% responded within seven days93.
The second runs as matters close. Trace each matter opened to its source, and mark whether its first contact would have been missed without the change: a call after hours that the agent answered, or a form reply sent within minutes. Mark it when the contact happens, not afterward, because it is the firm's judgment. Add up the fees collected from the matters marked yes. The purchase is covered when that sum reaches its cost. This is arithmetic on a handful of matters, not a statistical test, so it need not wait for the total of signed matters to move beyond chance.
Table 3 gives the values the fees column is read against. They come from the same cited figures as Chapter 4, each with its limit.
| Unit | Value | Source and limit |
|---|---|---|
| A collected hour | $245.52, which is $300 x 88% x 93% | The Michigan median rate of $30018, with realization of 88% and collection of 93% in Clio†'s data19; Michigan's survey is self-reported |
| A DUI matter | $3,500 | Average attorney cost in a 2020 Washington State Office of Public Defense survey15; one state, excludes investigators and experts |
| An injury case | $6,667 | One-third of the more than $20,000 average bodily-injury claim payment16, and the one-third contingency fee that the State Bar of California says is often used17; our inference, and the payment is what insurers paid |
Table 3. The values a law firm's fee column can be read against; where a value is computed, the working is in the table. Replace each with the firm's own average as soon as the ledger has one.
Against these values, Chapter 4's purchases are covered at the counts that chapter gives. The agent added to a plan the firm already has takes 1.2 collected hours a month, 1.0 DUI matters a year or 0.5 of an injury case. The lowest plan plus the agent takes 5.3 hours a month, the 4.5 DUI matters or 2.3 injury cases a year given above. For practice areas such as divorce and estate planning, fees from a source we could open are NF; the nearest we hold are the DUI and injury values in Table 3, so the hours are the unit to use, with the firm's own rate. When declines for no capacity are growing, the firm is filling its schedule, and the gain from more inquiries takes another form: better-paying matters chosen, a higher minimum fee or another lawyer, each of which the fees column will show.
Law firm lead generation and Rule 7.2: where purchased leads enter the count
A purchased lead is an inquiry from a source, and it belongs in the same ledger as any other. Comment [5] to ABA Model Rule 7.2 says a lawyer may pay others for generating client leads, such as Internet-based client leads, as long as the lead generator does not recommend the lawyer, any payment is consistent with Rules 1.5(e) and 5.4, and the generator's communications are consistent with Rule 7.168. It also says a lawyer must not pay a lead generator that states, implies or creates a reasonable impression that it is recommending the lawyer, is making the referral without payment from the lawyer, or has analyzed a person's legal problems when determining which lawyer should receive the referral68. Rule 7.2(b) separately bars a lawyer from compensating or giving or promising anything of value to a person for recommending the lawyer's services, with listed exceptions that include the reasonable costs of permitted advertisements3.
These are model texts about what a generator may say and how it may be paid. States adopt their own versions, and how they apply to a given vendor or contract is for the firm's counsel. Chapter 3 sets them out in full. A firm that buys leads should compare the vendor's count of leads delivered with its own ledger's counts of consultations held and matters signed from that source. The break-even is the monthly cost divided by the fee collected per matter. For every $1,000 a month spent on a lead source, the source must produce 0.29 of a DUI matter a month at $3,500 ($1,000 / $3,500), or 0.15 of an injury case at $6,667 ($1,000 / $6,667), the values in Table 3. A cost per lead says little until it is carried through to the last line of Table 1 and set against figures like these, or against the firm's own.
Client value, retention and referred work
A signed matter is worth more than its first fee when clients return or refer others. Gupta, Lehmann and Stuart (2004) define the value of a customer as the expected sum of discounted future earnings144. Valuing five firms from public data, they found that a 1% improvement in retention, margin or acquisition cost improves firm value by 5%, 1% and 0.1% respectively144. They also found that a 1% improvement in retention has almost five times greater impact on firm value than a 1% change in discount rate or cost of capital144. The 5% for retention comes from a model fitted to five firms and depends on the model's assumptions144, and none of these results is about professional practices.
Reichheld and Sasser's 1990 article, in the summary that Bain & Company republishes, states that companies can boost profits by almost 100% by retaining just 5% more of their customers145. We rest nothing on this secondary copy of the authors' statement.
We think whether retention matters to a law firm depends on the practice area. Where clients seldom return, the lever is referral, and where they do return, it is repeat work. We estimate that the referred share of new matters is below one half (from FindLaw†'s 2024 survey; likely high). FindLaw† (legal marketing) reports that 48% of respondents who learned about the attorney they contacted used referrals from family, friends or trusted coworkers27. That counts people who used referrals as one source among several, so as a share of clients who arrived by referral it errs high. The repeat share in firm records is NF, and nothing near it is cited here.
Chapter 1 weighs the vendor and poll figures on how clients find a lawyer. The ledger gives the firm its own number: add two columns to the last line of Table 1, for later matters from the same client and for matters from the people that client referred. After a year or more, a firm can say what share of its fees, by source, came from repeat and referred work, which no outside figure can tell it about that firm.
Measuring the monthly AI answer check
Chapter 2 describes a monthly answer check: a fixed list of questions put to each assistant, with the answers recorded. Our study (the report, with its data, and its method in the long paper's chapter on how assistants recommend professionals) is the guide to what such a check can measure, and its limits come first. Each question in our study was asked once, names were matched across assistants by spelling, up to eight names were read per answer, and answers through an API are a baseline, not what every signed-in user sees1. The study therefore says nothing about run-to-run variation. Whether the same assistant, asked the same question an hour later, would name the same firms is unmeasured, and a single monthly reading cannot separate a real change from that variation.
A design that respects the limit asks each question several times in one sitting, for each assistant, and records the spread. Our question form was "Who is a good dentist in Atlanta, GA? Name specific businesses." with the city and the kind of business changed1. A firm would put its own city and practice area in the same form. For each answer, record whether the firm was named, where in the list, which other businesses were named and which sites the answer cited. Treat a change between months as real only if it exceeds the spread across repeats within one sitting. Our study did not measure how many repeats are enough.
Read each assistant separately. For family law attorneys, asked about in 16 cities, the assistants named 240 businesses, of which 13 (5.4%) were named by all three and 75.8% by only one1. A firm named by one assistant on one day says little about the other two.
Record the sites cited as well as the names. The 25 most-cited websites in our study drew 29% of all citations, and they were directories, review platforms, list sites, publishers and a forum, not any one local business's site1. A firm's profile on those kinds of site is something to count. Our study does not show that a cited page caused a name to appear, and Chapter 2 sets out what a firm controls.
A first 90 days for law firm marketing strategies, in order
This section speaks to you. It puts the work of Chapters 2 to 5 in order, with the count first, because a count started after the work cannot show what the work changed. Table 4 gives the order.
| Weeks | Work | What you count | Where to read more |
|---|---|---|---|
| 1 to 2 | Set up the ledger in Table 1. Write down how inquiries are answered today. Run the answer check several times for each assistant. | A baseline for every line of Table 1, calls answered first, and for the answer check | This chapter; Chapter 2 |
| 1 to 4 | Profiles and directories: claim and complete the firm's listings, starting with the Google Business Profile | Listings complete; inquiries by source | Chapter 2 |
| 3 to 6 | Practice-area pages: one for each practice area the firm wants, approved by the firm's own counsel before it goes live | Inquiries by source and practice area | Chapters 2 and 3 |
| 5 to 8 | Intake and answering: decide who answers each channel and how fast; if you add an AI agent, callers are told it is one | Calls answered; time to first reply; consultations booked and held | Chapters 4 and 5 |
| 7 to 10 | Review requests: ask every client the same way, in wording approved as the pages are | Reviews received per signed matter | Chapter 3 |
| 11 to 13 | First reading: read the ledger and the answer check, fix what is broken, leave alone what chance could explain | Every line of Table 1 | This chapter |
Table 4. A first 90 days in order, with the count running from week 1. The weeks overlap on purpose.
For profiles, Google's Business Profile help page says local results are mainly based on relevance, distance and popularity40, and that there is no way to request or pay for a better local ranking40. We think work on profiles is therefore work on the information those factors draw on, not a purchase of position. For practice-area pages, ABA Model Rule 7.1 says a lawyer shall not make a false or misleading communication about the lawyer or the lawyer's services2. Chapter 3 has the comments on unsubstantiated claims.
For intake, the ledger's response-time line shows how fast inquiries are answered today, and an AI agent, if you add one, should tell callers it is an AI assistant, as ours does. Chapter 4 covers what an agent may do on a firm's line.
Ask for reviews once intake is stable. The FTC said its final reviews rule prohibits businesses from providing compensation or other incentives conditioned on the writing of consumer reviews expressing a particular sentiment, positive or negative73. Chapter 3 covers that rule and Formal Opinion 496 on negative reviews.
For reference, our plans are $999, $1,999 and $2,999 a month per location, or $9,990, $19,990 and $29,990 for a 12-month term paid in full, two months free. The $2,999 plan carries a $3,000 setup fee, $1,400 on a 12-month term, and the two lower plans a one-time onboarding fee of $499 and $999, waived on a 12-month term39. Our AI voice agent is an add-on at $299 a month with 250 call minutes included, about 62 four-minute calls (250 / 4), and $0.39 a minute after that39.
Take the lowest plan with the agent, as Chapter 4 does, with the agent's calls inside the included minutes. That is $1,298 a month ($999 + $299), or $3,894 over three months (3 x $1,298), a quarter of the yearly cost in Chapter 4's Table 4, so the quarter's break-even is a quarter of the yearly counts: 15.9 collected hours (3 x 5.3), about 1.1 DUI matters or 0.6 of an injury case (a quarter of 4.5 and of 2.3). The plan is set against the new matters it brings and the agent against the calls it answers. You can run the same 90 days with a spreadsheet and your own staff. Whether to buy any of it from us is your call, and our page for law firms and our pricing page describe what we offer.
At day 90, read the ledger in the order of Table 1. First compare calls answered, inquiries replied to and time to first reply with the baseline from weeks 1 to 2. Then read consultations held, matters opened and fees collected, each by source, and add up the fees from the matters whose first contact would have been missed before the change. Treat any rate built on a few consultations with care: Table 5 shows how wide the doubt is around a rate near one half, such as a signing rate, for different numbers of consultations held.
| Consultations held | Two standard errors around a rate near one half |
|---|---|
| 10 | about 32 percentage points |
| 20 | about 22 percentage points |
| 50 | about 14 percentage points |
| 100 | about 10 percentage points |
Table 5. The width of the doubt around a rate near one half, by number of consultations held. Each width is two times the square root of (0.25 / the number held), with the normal approximation, which is rough at small numbers.
A block of declined matters that share one reason points first at your listings and pages, not at intake. Decide three things from the first reading. Which source produced nothing, and should it be stopped or fixed? Does response time, or declines for fit, explain the gap between inquiries and signed matters? Which assistants never name the firm, and which sites do they cite instead? Repeat the reading each quarter, not each month, for the reason in Table 2's section.
Limits of this paper and what would settle them
- How often law firms miss calls and emails. The tests are Clio†'s 2024 shop of 500 firms and its 2019 shop, a 2016 shop reported second-hand, and Hennessey Digital†'s 2025 test of 1,333 firms' web forms, limited as Chapter 4 says. A study with a stated sampling frame, or a firm's own call log, would settle it for one firm.
- The share of clients who come by referral. The figures are survey answers, attributed in Chapter 1 and above. The repeat and referral share in firm records is NF; the nearest evidence is those survey answers. Many firms' source columns would settle it.
- The effect of reviews on the choice of a lawyer. The survey figures in Chapter 1 are stated behavior, and this paper cites no study that tracked reviews against signed matters. A study with a stated sample would settle it.
- Counts of law firms, conversion rates and what a matter is worth. The Bureau of Labor Statistics counts 188,520 private establishments in offices of lawyers in 202411. They are locations with employees and not firms, and sole practitioners with no employees are not counted. Of first-quarter establishments, 75.5% had fewer than 5 employees (141,518 / 187,335)12,13. Inquiry-to-client rates at law firms are NF; the nearest evidence is the adjacent professions' rates, from which Chapter 4 estimates a ceiling. A matter's worth rests on the fees in Chapter 4, which cover DUI and injury matters only, and a firm's own ledger is the source for its own.
- Run-to-run variation in AI answers. A repeat-asking study of AI recommendations for local businesses or professional practices is NF; our own study, which asked each question once, is the nearest evidence and cannot show variation. Repeated runs of a fixed question list across several weeks would measure it.
- How far last-click attribution misstates a law firm's results. An estimate is NF, as above; the nearest evidence is the eBay and Facebook comparisons. A randomized test would settle it, but Lewis and Rao say informative advertising experiments can be costly and potentially infeasible for many firms140.
- Surveys of AI use. They count different things and cannot be combined. One instrument, asked with the same wording of a stated sample over time, would show a trend.
- The rules. The ABA texts are models, and states adopt their own versions. The state's own text settles what applies.
References
Before the paper cited a source, the page was fetched and the passage we rely on was matched against it, except where a reference is marked otherwise. The small numbers in the text point here.
- Technical Marketing. Who AI recommends locally: a study of local answers. Technical Marketing, 2026-11. technicalmarketing.ai/reports/ai-local-recommendations/ Our own data.
- American Bar Association. Rule 7.1: Communications Concerning a Lawyer's Services (ABA Model Rules of Professional Conduct). American Bar Association, 2024. americanbar.org/groups/professional_responsibility/publications/model_
- American Bar Association. Rule 7.2: Communications Concerning a Lawyer's Services: Specific Rules (ABA Model Rules). American Bar Association, 2024. americanbar.org/groups/professional_responsibility/publications/model_
- American Bar Association. Rule 1.18: Duties to Prospective Client (ABA Model Rules of Professional Conduct). American Bar Association, 2020. americanbar.org/groups/professional_responsibility/publications/model_
- ABA Standing Committee on Ethics and Professional Responsibility. ABA Formal Opinion 512: Generative Artificial Intelligence Tools. American Bar Association, 2024-07-29. americanbar.org/content/dam/aba/administrative/professional_responsibi
- American Bar Association. ABA Profile of the Legal Profession. American Bar Association, 2025. americanbar.org/news/profile-legal-profession/
- American Bar Association. U.S. lawyer population up significantly for the first time since 2020, ABA report finds. American Bar Association, 2025. americanbar.org/news/abanews/aba-news-archives/2025/12/aba-2025-profil
- American Bar Association. ABA Profile of the Legal Profession 2024 Demographics. American Bar Association, 2000-2024. americanbar.org/news/profile-legal-profession/demographics/
- U.S. Bureau of Labor Statistics. Occupational Outlook Handbook: Lawyers. U.S. Bureau of Labor Statistics, 2025. bls.gov/ooh/legal/lawyers.htm
- U.S. Census Bureau. Quarterly Selected Services Estimates, Fourth Quarter 2025 (advance), Table A4. U.S. Census Bureau, 2025. www2.census.gov/services/qss/2025/qssq4-25advpr.pdf
- U.S. Bureau of Labor Statistics. QCEW open data: 2024 annual averages, NAICS 54111 Offices of lawyers (all areas). U.S. Bureau of Labor Statistics, 2024. data.bls.gov/cew/data/api/2024/a/industry/54111.csv
- U.S. Bureau of Labor Statistics. QCEW open data: 2024 first quarter, NAICS 54111 Offices of lawyers (all areas). U.S. Bureau of Labor Statistics, 2024-Q1. data.bls.gov/cew/data/api/2024/1/industry/54111.csv
- U.S. Bureau of Labor Statistics. QCEW open data: 2024 first quarter by establishment size class 1, all industries and areas. U.S. Bureau of Labor Statistics, 2024-Q1. data.bls.gov/cew/data/api/2024/1/size/1.csv
- U.S. Bureau of Labor Statistics. QCEW open data: 2024 first quarter by establishment size class 2, all industries and areas. U.S. Bureau of Labor Statistics, 2024-Q1. data.bls.gov/cew/data/api/2024/1/size/2.csv
- Washington State Office of Public Defense. Private Representation Costs - Misdemeanors. Washington State Office of Public Defense, 2020. opd.wa.gov/sites/default/files/2023-08/00915-2020_MPrivateRates.pdf
- Insurance Research Council. Insurance Research Council Study Finds Rising Medical Costs and Attorney Involvement Are Driving Up Auto Injury Claim Costs. Insurance Research Council (The Institutes), 2022. insurance-research.org/news/insurance-research-council-study-finds-ris
- State Bar of California, Standing Committee on Professional Responsibility and Conduct. Arbitration Advisory No. 2025-01: Determination of a Reasonable Fee. State Bar of California, 2025. calbar.ca.gov/sites/default/files/portals/0/documents/mfa/2025/2025-01
- State Bar of Michigan. 2023 Economics of Law Survey. State Bar of Michigan, 2023. michbar.org/file/pmrc/pdfs/EoL_Report23.pdf
- Clio (Themis Solutions Inc.). 2025 Legal Trends Report. Clio; PDF copy hosted by the Alaska Bar Association, 2024. alaskabar.org/wp-content/uploads/2025-Legal-Trends-Report.pdf Secondary source.
- Clio (Themis Solutions Inc.). 2023 Legal Trends for Solo Law Firms. Clio, 2022. clio.com/wp-content/uploads/2022/10/2023-Legal-Trends-for-Solo-Law-Fir Secondary source.
- Darby,Karni. Free Competition and the Optimal Amount of Fraud. University of Chicago Press, 1973. doi.org/10.1086/466756 Held through its record or abstract; passage not matched.
- Uwe Dulleck; Rudolf Kerschbamer. On Doctors, Mechanics, and Computer Specialists: The Economics of Credence Goods. Journal of Economic Literature (American Economic Association), 2006. doi.org/10.1257/002205106776162717
- Mike Bendixen; Graham Gault. Consumer evaluation of perceived risks for goods and services. South African Journal of Business Management, 1995. sajbm.org/index.php/sajbm/article/download/827/761 Secondary source.
- Ellen Day and Hiram C. Barksdale Jr.. How firms select professional services. Elsevier / Industrial Marketing Management, 1992. doi.org/10.1016/0019-8501(92)90002-b Held through its record or abstract; passage not matched.
- George A. Akerlof. The Market for "Lemons": Quality Uncertainty and the Market Mechanism. The Quarterly Journal of Economics (Oxford University Press), 1970. doi.org/10.2307/1879431
- George J. Stigler. The Economics of Information. Journal of Political Economy (University of Chicago Press), 1961. doi.org/10.1086/258464
- FindLaw (Thomson Reuters). The 2024 U.S. Consumer Legal Needs Survey: What attorneys need to know. FindLaw, 2024. findlaw.com/lawyer-marketing/blog/the-2024-u-s-consumer-legal-needs-su Secondary source.
- FindLaw (Thomson Reuters). Key takeaways from the 2024 U.S. Consumer Legal Needs Survey. FindLaw, 2024. findlaw.com/lawyer-marketing/blog/key-takeaways-from-the-2024-us-consu Secondary source.
- Clio (Themis Solutions Inc.). Legal Trends Report 2019. Clio, as hosted by the District of Columbia Bar, 2019. dcbar.org/getmedia/a7bbe4b0-eec7-4fc7-8c85-485853c235ab/Clio-2019-Lega Secondary source.
- Illinois Supreme Court Commission on Professionalism / Clio. Clio Legal Trends Report Finds Law Firms Aren't Responsive to Potential Clients. Illinois Supreme Court Commission on Professionalism, 2019. 2civility.org/clio-legal-trends-report-finds-law-firms-arent-responsiv Secondary source.
- ABA Journal. How People Find Lawyers: Referrals Are Popular, Blogs Not So Much, Poll Finds. ABA Journal, 2010-09. abajournal.com/news/article/how_people_find_lawyers_referrals_are_popu Secondary source.
- BrightLocal. Local Consumer Review Survey 2023. BrightLocal, 2023. brightlocal.com/research/local-consumer-review-survey-2023/ Secondary source.
- BrightLocal. Local Consumer Review Survey 2024. BrightLocal, 2024. brightlocal.com/research/local-consumer-review-survey-2024/ Secondary source.
- Michael Luca. Reviews, Reputation, and Revenue: The Case of Yelp.Com. Harvard Business School Working Paper 12-016, 2011. econpapers.repec.org/RePEc:hbs:wpaper:12-016
- Anderson and Magruder. Learning from the Crowd: Regression Discontinuity Estimates of the Effects of an Online Review Database. Oxford University Press (OUP), 2012. doi.org/10.1111/j.1468-0297.2012.02512.x
- Judith A. Chevalier; Dina Mayzlin. The Effect of Word of Mouth on Sales: Online Book Reviews. Journal of Marketing Research (American Marketing Association), 2006. doi.org/10.1509/jmkr.43.3.345
- U.S. Supreme Court (Blackmun, J.). Bates v. State Bar of Arizona, 433 U.S. 350. Legal Information Institute, Cornell Law School (text of the opinion), 1977-06-27. law.cornell.edu/supremecourt/text/433/350
- Technical Marketing. What service businesses search for: 2026 demand data. Technical Marketing, 2026-11. technicalmarketing.ai/reports/service-business-search-demand/ Our own data.
- Technical Marketing. Pricing. Technical Marketing, 2026-11. technicalmarketing.ai/pricing/ Our own data.
- Google. Tips to improve your local ranking on Google (Google Business Profile Help). Google Business Profile Help, 2026-11. support.google.com/business/answer/7091?hl=en
- Anindya Ghose, Avi Goldfarb, and Sang Pil Han. How Is the Mobile Internet Different? Search Costs and Local Activities. Information Systems Research / INFORMS, 2013. doi.org/10.1287/isre.1120.0453
- Google Search Central. AI features and your website. Google, 2025. developers.google.com/search/docs/appearance/ai-features
- Google Search Central. Google Search's AI Overviews and Your Website. Google, 2024. developers.google.com/search/docs/appearance/ai-overviews
- OpenAI. Searching the web with ChatGPT. OpenAI Help Center, 2026. help.openai.com/en/articles/9237897-chatgpt-search
- Johannes Beus. Why (almost) everything you knew about Google CTR is no longer valid. SISTRIX, 2020. sistrix.com/blog/why-almost-everything-you-knew-about-google-ctr-is-no Secondary source.
- Brian Dean. Here's What We Learned About Organic Click-Through Rate. Backlinko, 2023. backlinko.com/google-ctr-stats Secondary source.
- Pew Research Center. Google users are less likely to click on links when an AI summary appears in the results. Pew Research Center, 2025-03. pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to
- Rosie Murphy. Local Services Ads Click Study. BrightLocal, 2018. brightlocal.com/research/local-services-ads-click-study/ Secondary source.
- SE Ranking. AI Overviews Research: Analyzing Google's approach to YMYL topics. SE Ranking, 2024. seranking.com/blog/ai-overviews-and-ymyl-topics-research/ Secondary source.
- Miriam Ellis. New Research: The Prevalence of AI Overviews in Local Search. Whitespark, 2025. whitespark.ca/blog/case-study-the-prevalence-of-ai-overviews-in-local- Secondary source.
- Rand Fishkin. 2024 Zero-Click Search Study: For every 1,000 EU Google Searches, only 374 clicks go to the Open Web. In the US, it's 360.. SparkToro, 2024. sparktoro.com/blog/2024-zero-click-search-study-for-every-1000-us-goog Secondary source.
- Chatterji et al.. How People Use ChatGPT. National Bureau of Economic Research, 2025. doi.org/10.3386/w34255
- Pew Research Center. Americans and AI 2026: Chatbots, Smart Devices and Views on Impact. Pew Research Center, 2026. pewresearch.org/internet/2026/06/17/americans-and-ai-2026-chatbots-sma
- BrightLocal Research Team. Local Consumer Review Survey 2025. BrightLocal, 2025. brightlocal.com/research/local-consumer-review-survey-2025 Secondary source.
- Rosie Murphy. Local Consumer Review Survey 2026: Star Ratings Keep Rising, Old Reviews Don't Cut It. BrightLocal, 2026. brightlocal.com/research/local-consumer-review-survey/ Secondary source.
- Conductor. The 2026 AEO / GEO Benchmarks Report. Conductor, 2025. conductor.com/academy/aeo-geo-benchmarks-report/ Secondary source.
- Microsoft Clarity. AI Traffic Converts at 3x the Rate of Other Channels (Study). Microsoft, 2025. clarity.microsoft.com/blog/ai-traffic-converts-at-3x-the-rate-of-other Secondary source.
- OpenAI. Setting up your business on ChatGPT. OpenAI Help Center, 2026. help.openai.com/en/articles/20001635-setting-up-your-business-on-chatg
- Rand Fishkin. NEW Research: AIs are highly inconsistent when recommending brands or products; marketers should take care when tracking AI visibility. SparkToro, 2025. sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recom Secondary source.
- Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan, Ameet Deshpande. GEO: Generative Engine Optimization. arXiv, 2023. arxiv.org/abs/2311.09735
- Aggarwal et al.. GEO: Generative Engine Optimization. arXiv, 2023. arxiv.org/pdf/2311.09735
- Liu, Zhang, and Liang. Evaluating Verifiability in Generative Search Engines. Association for Computational Linguistics, 2023. arxiv.org/abs/2304.09848
- ABA Standing Committee on Ethics and Professional Responsibility. ABA Formal Opinion 496: Responding to Online Criticism. American Bar Association, 2021-01-13. americanbar.org/content/dam/aba/administrative/professional_responsibi
- American Bar Association. Rule 1.6: Confidentiality of Information (ABA Model Rules of Professional Conduct). American Bar Association, 2024. americanbar.org/groups/professional_responsibility/publications/model_
- Justice Blackmun, for the Court. Virginia State Board of Pharmacy v. Virginia Citizens Consumer Council, Inc., 425 U.S. 748. Supreme Court of the United States (text via Cornell Legal Information Institute), 1976-05-24. law.cornell.edu/supremecourt/text/425/748
- Federal Trade Commission. Trade Regulation Rule on the Use of Consumer Reviews and Testimonials (89 FR 68034). Federal Register (govinfo.gov), 2024-10-21. govinfo.gov/content/pkg/FR-2024-08-22/pdf/2024-18519.pdf
- American Bar Association. Comment on Rule 7.1 (ABA Model Rules of Professional Conduct). American Bar Association, 2024. americanbar.org/groups/professional_responsibility/publications/model_
- American Bar Association. Comment on Rule 7.2 (ABA Model Rules of Professional Conduct). American Bar Association, 2024. americanbar.org/groups/professional_responsibility/publications/model_
- American Bar Association. Rule 7.3 Solicitation of Clients (ABA Model Rules of Professional Conduct). American Bar Association, 2019. americanbar.org/groups/professional_responsibility/publications/model_
- Federal Communications Commission. Implications of Artificial Intelligence Technologies on Protecting Consumers from Unwanted Robocalls and Robotexts, Declaratory Ruling, FCC 24-17. Federal Communications Commission, 2024-02-08. docs.fcc.gov/public/attachments/FCC-24-17A1.pdf
- HHS Office for Civil Rights and Manasa Health Center. Resolution Agreement and Corrective Action Plan, Manasa Health Center, LLC (HHS TN 20-380176). U.S. Department of Health and Human Services, 2023-03-28. hhs.gov/hipaa/for-professionals/compliance-enforcement/agreements/mana
- Federal Trade Commission. 16 CFR 465.2 Fake or false consumer reviews, consumer testimonials, or celebrity testimonials. Legal Information Institute, Cornell Law School (Code of Federal Regulations text), 2024-10-21. law.cornell.edu/cfr/text/16/465.2
- Federal Trade Commission. Federal Trade Commission Announces Final Rule Banning Fake Reviews and Testimonials. Federal Trade Commission, 2024-08-14. ftc.gov/news-events/news/press-releases/2024/08/federal-trade-commissi
- Federal Trade Commission. 16 CFR 465.4 Buying positive or negative consumer reviews. Legal Information Institute (copy of the e-CFR text), 2024-10-21. law.cornell.edu/cfr/text/16/465.4
- Federal Trade Commission. 16 CFR 465.7 Review suppression. Legal Information Institute (copy of the e-CFR text), 2024-10-21. law.cornell.edu/cfr/text/16/465.7
- Federal Trade Commission. The Consumer Reviews and Testimonials Rule: Questions and Answers. Federal Trade Commission, 2024-10-21. ftc.gov/business-guidance/resources/consumer-reviews-testimonials-rule
- Federal Trade Commission. FTC Warns 10 Companies About Possible Violations of the Agency's New Consumer Review Rule. Federal Trade Commission, 2025-12-22. ftc.gov/news-events/news/press-releases/2025/12/ftc-warns-10-companies
- Professional Ethics Committee for the State Bar of Texas. Opinion 685. Texas Center for Legal Ethics, 2020-01. legalethicstexas.com/resources/opinions/opinion-685/
- North Carolina State Bar. 2018 Formal Ethics Opinion 1: Participation in Website Directories and Rating Systems that include Third Party Reviews. North Carolina State Bar, 2018-07-27. ncbar.gov/for-lawyers/ethics-and-governing-rules/ethics-opinions/opini
- American Bar Association. Rule 5.3: Responsibilities Regarding Nonlawyer Assistance (ABA Model Rules of Professional Conduct). American Bar Association, 2024. americanbar.org/groups/professional_responsibility/publications/model_
- HHS Office for Civil Rights. Business Associates Frequently Asked Questions (FAQs). U.S. Department of Health and Human Services, 2026. hhs.gov/hipaa/for-professionals/privacy/guidance/business-associates/i
- Technical Marketing. Which software will sign a HIPAA BAA? 28 common tools, checked. Technical Marketing, 2026-11. technicalmarketing.ai/reports/hipaa-baa-availability/ Our own data.
- State Bar of California / California Supreme Court. California Rules of Professional Conduct, Current Rules 2026, rule 7.2 (Advertising). State Bar of California, 2026. calbar.ca.gov/Portals/0/documents/rules/Rules-of-Professional-Conduct.
- Supreme Court of Texas / State Bar of Texas. Texas Disciplinary Rules of Professional Conduct (Including Amendments Effective January 31, 2022), Rule 7.02. State Bar of Texas, 2022-01-31. texasbar.com/AM/Template.cfm?Section=Home&ContentID=27271&Template=/CM
- Supreme Court of Florida / The Florida Bar. Rules Regulating The Florida Bar, Chapter 4 (RRTFB October 22, 2026), Rule 4-7.12 Required Content. The Florida Bar, 2026-10-22. www-media.floridabar.org/uploads/2026/11/2027_04-OCT-Chapter-4-RRTFB.p
- Washington Supreme Court. Washington Court Rules, RPC 7.2 [RESERVED]. Washington Courts, 2021-01-26. courts.wa.gov/court_rules/pdf/RPC/GA_RPC_07_02_00.pdf
- Washington Supreme Court. Washington Court Rules, RPC 7.3 Solicitation of Clients. Washington Courts, 2025-04-29. courts.wa.gov/court_rules/pdf/RPC/GA_RPC_07_03_00.pdf
- Washington Supreme Court. Washington Court Rules, RPC 7.1 Communications Concerning a Lawyer's Services, Comment [3]. Washington Courts, 2016-09-01. courts.wa.gov/court_rules/pdf/RPC/GA_RPC_07_01_00.pdf
- New York State Bar Association. Ethical Marketing of Your Law Practice: Understanding New York's New Rules on Lawyer Advertising. New York State Bar Association, 2026-06-01. nysba.org/events/ethical-marketing-of-your-law-practice-understanding- Secondary source.
- Clio. Highlights from the Legal Trends Report: the Legal Industry in 2024. Clio (a legal practice-management software vendor), 2024. clio.com/blog/highlights-from-2024-legal-trends-report/ Secondary source.
- Clio. AI-powered legal practices surge: Clio's latest Legal Trends Report reveals major shift (press release). Clio (a legal practice-management software vendor), 2024. clio.com/about/press/clio-latest-legal-trends-report/ Secondary source.
- North Carolina Bar Association (NC Lawyer / CPM column). Secret Shop Your Law Firm. North Carolina Bar Association, 2016. ncbar.org/?p=10300 Secondary source.
- Hennessey Digital. 2025 Lead Form Response Times & Trends. Hennessey Digital, 2025. hennessey.com/2025-lead-form-response-time-study/ Secondary source.
- James B. Oldroyd; Kristina McElheran; David Elkington. The Short Life of Online Sales Leads. Harvard Business Review, 2011. hbr.org/2011/03/the-short-life-of-online-sales-leads
- SmartCompany. Are you letting online leads go cold? (quoting Oldroyd, McElheran and Elkington, The Short Life of Online Sales Leads, Harvard Business Review, March 2011). SmartCompany (Australia), 2011. smartcompany.com.au/marketing/20110704-are-you-letting-online-leads-go Secondary source.
- Oldroyd. The Short Life of Online Sales Leads. Harvard Business Review, 2011. thedenmangroupselling.wordpress.com/wp-content/uploads/2011/03/the-sho
- Oldroyd. The Lead Response Management Study. Lead Response Management, 2007. leadresponsemanagement.org/lrm_study
- Insurance Research Council. Why Do Claimants Hire Attorneys?. Insurance Research Council, 2016. insurance-research.org/news/why-do-claimants-hire-attorneys
- Jim Ash. Board of Governors adopts ethics guidelines for generative AI use. The Florida Bar News, 2024-01-19. floridabar.org/the-florida-bar-news/board-of-governors-adopts-ethics-g
- The Florida Bar. Florida Bar Ethics Opinion 24-1 (January 19, 2024). The Florida Bar (copy hosted by University at Buffalo School of Law CLE materials), 2024-01-19. law.buffalo.edu/content/dam/law/content/cle/24mar27-materials/fl-bar-e
- Christopher C. Rivers, Tribunal Member, Civil Resolution Tribunal (British Columbia). Moffatt v. Air Canada, 2024 BCCRT 149. Civil Resolution Tribunal of British Columbia, 2024-02-14. decisions.civilresolutionbc.ca/crt/crtd/en/525448/1/document.do
- United States Congress. 18 U.S.C. 2511: Interception and disclosure of wire, oral, or electronic communications prohibited. Legal Information Institute, Cornell Law School (US Code text), 2026. law.cornell.edu/uscode/text/18/2511
- Washington State Legislature. RCW 9.73.030: Intercepting, recording, or divulging private communication - Consent required - Exceptions. Washington State Legislature, 2026. app.leg.wa.gov/rcw/default.aspx?cite=9.73.030
- California Legislature. California Penal Code section 632. California Legislative Information (leginfo.legislature.ca.gov), 2026. leginfo.legislature.ca.gov/faces/codes_displaySection.xhtml?lawCode=PE
- Utah Legislature. Utah Code 13-77-103 Required disclosures. Utah State Legislature (le.utah.gov), 2025-05-07. le.utah.gov/xcode/Title13/Chapter77/C13-77-S103_2025050720250507.html
- Utah Legislature. Utah Code 13-77-104 Safe harbor. Utah State Legislature (le.utah.gov), 2025-05-07. le.utah.gov/xcode/Title13/Chapter77/C13-77-S104_2025050720250507.html
- Utah Legislature. Utah Code 13-77-101 Definitions. Utah State Legislature (le.utah.gov), 2026-05-06. le.utah.gov/xcode/Title13/Chapter77/C13-77-S101_2026050620260506.html
- Annapoorna Kini et al.. Conversational artificial intelligence for pre-procedural patient preparation: implementation, validation and patient satisfaction. npj Digital Medicine, 2025. pmc.ncbi.nlm.nih.gov/articles/PMC13710839
- Patient Prism. The Dental Patient Access Report. Patient Prism, 2025. patientprism.com/report/dental-patient-access-report/ Secondary source.
- Call Box. The Top 5 Phone Trends of 2020 and How Healthcare Practices Are Taking Action: Data Study. Call Box, 2020-02. assets-cdn.carwars.com/websites/callbox/resources/dental/CB+Healthcare Secondary source.
- Invoca. The Healthcare Lead Conversion Benchmarks Report 2026. Invoca, 2026. invoca.com/reports/the-invoca-healthcare-lead-conversion-benchmarks-re Secondary source.
- Mark Calaguas. 2024 Artificial Intelligence TechReport. American Bar Association, Law Practice Division (ABA TechReport), 2024. americanbar.org/groups/law_practice/resources/tech-report/2024/2024-ar
- Bob Ambrogi. Thomson Reuters Survey: Over 95% of Legal Professionals Expect Gen AI to Become Central to Workflow Within Five Years. LawSites (LawNext), reporting a Thomson Reuters Institute survey, 2025. lawnext.com/2025/04/thomson-reuters-survey-over-95-of-legal-profession Secondary source.
- Khatiwoda. Large Firms With at Least 20 Employees Biggest AI Users. U.S. Census Bureau, 2026. census.gov/library/stories/2026/05/ai-use-businesses.html
- National Institute of Standards and Technology. Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, NIST AI 600-1. National Institute of Standards and Technology, 2024-07. nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf
- Matthew Dahl, Varun Magesh, Mirac Suzgun, Daniel E. Ho. Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models. arXiv (v2, last revised 21 Jun 2024), 2024. arxiv.org/abs/2401.01301
- Varun Magesh, Faiz Surani, Matthew Dahl, Mirac Suzgun, Christopher D. Manning, Daniel E. Ho. Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools. arXiv (submitted 30 May 2024), 2024. arxiv.org/abs/2405.20362
- Varun Magesh, Faiz Surani, Matthew Dahl, Mirac Suzgun, Christopher D. Manning, Daniel E. Ho. Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools. arXiv (published in the Journal of Empirical Legal Studies, 2025), 2024. arxiv.org/pdf/2405.20362
- Vals AI. VLAIR - Legal Research. Vals AI, 2025. vals.ai/industry-reports/vlair-10-14-25 Secondary source.
- Judge P. Kevin Castel, U.S. District Court, S.D.N.Y.. Mata v. Avianca, Inc., No. 22-cv-1461 (PKC), Opinion and Order on Sanctions (Doc. 54). U.S. District Court for the Southern District of New York (RECAP/CourtListener copy of the filed order), 2023-06-22. storage.courtlistener.com/recap/gov.uscourts.nysd.575368/gov.uscourts.
- U.S. District Court for the District of Wyoming. Wadsworth v. Walmart Inc., No. 2:23-CV-118-KHR, Order on Sanctions and Other Disciplinary Action (Doc. 181). U.S. District Court for the District of Wyoming (RECAP/CourtListener copy of the filed order), 2025-02-24. storage.courtlistener.com/recap/gov.uscourts.wyd.64014/gov.uscourts.wy
- U.S. District Court for the Northern District of Alabama. Johnson v. Dunn, No. 2:21-cv-1701-AMM, Sanctions Order (Doc. 204). U.S. District Court for the Northern District of Alabama (Docket Alarm copy of the filed order), 2025-07-23. docketalarm.com/cases/Alabama_Northern_District_Court/2--21-cv-01701/J
- Damien Charlotin. AI Hallucination Cases Database. Damien Charlotin (personal research database), 2026-11-10. damiencharlotin.com/hallucinations/ Secondary source.
- U.S. District Court for the Northern District of Texas. Local Civil Rules, N.D. Tex., LR 7.2(f) Disclosure of Use of Generative Artificial Intelligence. U.S. District Court for the Northern District of Texas, 2024-09-03. txnd.uscourts.gov/sites/default/files/documents/CIVRULES.pdf
- Chief Judge David C. Godbey, U.S. District Court for the Northern District of Texas. Special Order No. 2-98 (adopting new rules LR 7.2(f) and LCrR 47.2(e)). U.S. District Court for the Northern District of Texas, 2024-09-03. txnd.uscourts.gov/sites/default/files/orders/2-98.pdf
- State Bar of California, Committee on Professional Responsibility and Conduct. Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (Executive Summary). State Bar of California, 2026-05-14. calbar.ca.gov/Portals/0/documents/ethics/Generative-AI-Practical-Guida
- American Bar Association. Comment on Rule 1.1 (ABA Model Rules of Professional Conduct). American Bar Association, 2024. americanbar.org/groups/professional_responsibility/publications/model_
- The Florida Bar, Professional Ethics Committee. Florida Bar Ethics Opinion 24-1. The Florida Bar, 2024-01-19. www-media.floridabar.org/uploads/2024/01/FL-Bar-Ethics-Op-24-1.pdf
- Professional Ethics Committee for the State Bar of Texas. Opinion 705. Texas Center for Legal Ethics, 2025-02. legalethicstexas.com/resources/opinions/opinion-705/
- NYSBA Task Force on Artificial Intelligence. Report and Recommendations of the New York State Bar Association Task Force on Artificial Intelligence. New York State Bar Association, 2024-04-06. nysba.org/wp-content/uploads/2022/03/2024-April-Report-and-Recommendat
- Mark Bassingthwaighte, Risk Manager. Insurance Coverage Issues for Lawyers in the Era of Generative AI. ALPS Corporation, 2025. alpsinsurance.com/blog/insurance-coverage-issues-for-lawyers-in-the-er
- Noy and Zhang. Experimental evidence on the productivity effects of generative artificial intelligence. Science, 2023. doi.org/10.1126/science.adh2586
- Dell'Acqua et al.. Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality. Harvard Business School, 2023. mitsloan.mit.edu/sites/default/files/2023-10/SSRN-id4573321.pdf
- Brynjolfsson, Li, and Raymond. Generative AI at Work. National Bureau of Economic Research, 2023. doi.org/10.3386/w31161
- Erik Brynjolfsson, Danielle Li, Lindsey R. Raymond. Generative AI at Work. National Bureau of Economic Research (Working Paper 31161), 2023. nber.org/papers/w31161
- Anders Humlum, Emilie Vestergaard. Large Language Models, Small Labor Market Effects (BFI Working Paper 2025-56). Becker Friedman Institute, University of Chicago, 2025-04. bfi.uchicago.edu/wp-content/uploads/2025/04/BFI_WP_2025-56-3.pdf
- Brynjolfsson and Hitt. Beyond Computation: Information Technology, Organizational Transformation and Business Performance. Journal of Economic Perspectives, 2000. doi.org/10.1257/jep.14.4.23
- Clio. The Science Behind Smarter Law: Clio's 2025 Legal Trends Report Reveals How Technology Is Rewiring the Way Lawyers Work. Clio, 2025. clio.com/about/press/the-science-behind-smarter-law-clios-2025-legal-t Secondary source.
- Thomson Reuters. The AI Adoption Reality Check: Firms with AI Strategies are Twice as Likely to see AI-driven Revenue Growth; Those Without Risk Falling Behind. Thomson Reuters (press release, June 26, 2025), 2025. thomsonreuters.com/en/press-releases/2025/june/the-ai-adoption-reality Secondary source.
- Randall A. Lewis; Justin M. Rao. The Unfavorable Economics of Measuring the Returns to Advertising. The Quarterly Journal of Economics (Oxford University Press), 2015. doi.org/10.1093/qje/qjv023
- Brett R. Gordon; Florian Zettelmeyer; Neha Bhargava; Dan Chapsky. A Comparison of Approaches to Advertising Measurement: Evidence from Big Field Experiments at Facebook. Marketing Science (INFORMS), 2019. doi.org/10.1287/mksc.2018.1135
- Thomas Blake; Chris Nosko; Steven Tadelis. Consumer Heterogeneity and Paid Search Effectiveness: A Large-Scale Field Experiment. Econometrica (Econometric Society), 2014. doi.org/10.3982/ECTA12423
- Ron Berman. Beyond the Last Touch: Attribution in Online Advertising. Marketing Science (INFORMS), 2018. doi.org/10.1287/mksc.2018.1104
- Sunil Gupta; Donald R. Lehmann; Jennifer Ames Stuart. Valuing Customers. Journal of Marketing Research (American Marketing Association), 2004. doi.org/10.1509/jmkr.41.1.7.25084
- Frederick F. Reichheld; W. Earl Sasser, Jr.. Zero Defections: Quality Comes to Services. Harvard Business Review (excerpt hosted by Bain & Company), 1990. bain.com/insights/zero-defections-quality-comes-to-services-harvard-bu Secondary source.
Cite this paper
Mike Golliet. "Law firm marketing and SEO: what the evidence and the rules say." Technical Marketing, November 10, 2026. https://technicalmarketing.ai/papers/law-firm-marketing/
You may quote from this paper with a link to this page. It describes rules and research in general terms and is not legal, medical, tax or financial advice. Found an error? Tell us and we will correct it and note the change here.