Abstract
An accounting firm cannot sell more hours than it has. That changes what demand is worth to it, and it does not make demand worth less. Graduates who earned a bachelor's or master's degree in accounting fell to 55,152 in the 2023-24 academic year, down 6.6% from the year before1.
The firm's marketing problem is therefore to be found and verified by the clients it wants, to turn away the ones it does not, and to protect the hours it has. Asked for an accountant in 16 US cities, three AI assistants named 272 businesses and none was named by all three2. Each question was asked once, through each assistant's API2. The AI phrases for accounting are searched far more than its marketing phrases, at about 6,600 a month for "ai accounting" against 260 for "accountant marketing"3, and the evidence on how well AI does accounting work is mixed.
Six chapters cover capacity and how clients choose, SEO for accountants and AI answers, the advertising rules and the rules on client and tax return information, intake and busy season, AI for accountants, and measurement with a first 90 days. The paper is not tax or legal advice, and a firm's own counsel decides how any rule applies.
What this paper argues
- A firm that sells hours should market for fit before volume, because a client who takes more hours than the fee covers uses capacity the firm cannot easily replace.
- Asked for an accountant in 16 cities, three AI assistants named 272 businesses and agreed on none2.
- The phrase "cpa marketing" overstates interest in marketing for CPA firms, because CPA is also an advertising abbreviation, for cost per action: "what is cpa marketing" alone averages 720 searches a month3.
- In our reading, tax return information sets the hardest limit on a firm's tools: a preparer may not disclose or use it before obtaining the taxpayer's written consent, unless section 7216 or section 301.7216-2 of the regulations authorizes it4.
- On different tests the models' record in accounting is mixed: ChatGPT as of January 2023 averaged 47.5% on assessment questions where students averaged 76.7%, and a later configuration passed four professional exams with an average of 85.1%5,6.
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
CPA marketing when capacity is the constraint
What our study found when it asked about accountants
None of the businesses that three AI assistants named as accountants in our study was named by all three. We asked ChatGPT, Gemini and Perplexity who a good accountant is in each of 16 US cities2. The assistants named 272 businesses, 17 per question. Of those, 0 (0%) were named by all three, and 90.4% were named by only one2.
The accountant row sits inside a larger study: 199 local questions put to the three assistants across 16 cities and 23 kinds of business on 2026-11-10, which collected 597 answers, each through the assistant's own API with web search on2. Each question 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 sees2. 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. Our study, with its data, and the method behind it are published in full.
On the day we asked, most accountants appeared on only one list. We think a firm cannot judge its standing in AI answers by checking a single assistant, or assume that a good showing in one carries to the others.
The row describes the front of the path, where a prospective client looks for a firm. A firm that is found still has to be able to do the work, so this chapter starts at the other end, with the hours a firm has to sell.
The CPA firm as a business limited by hours: jobs, graduates and the CPA pipeline
A firm that sells accounting, tax or bookkeeping sells the hours of qualified people. Those hours are its capacity, and in this profession capacity is a real limit. None of the figures below measures one firm's capacity.
The field and the people
The Bureau of Labor Statistics counts private-sector establishments covered by unemployment insurance. Its 2024 annual averages show 63,177 offices of certified public accountants, 37,615 tax preparation offices and 71,408 offices of other accounting services7,8,9. These are offices, not firms: a firm with five offices appears five times, and an owner with no employees does not appear. Most CPA offices are small. In the first quarter of 2024, 44,800 of 62,828 CPA offices had fewer than 5 employees and 9,333 had 5 to 910,11,12. That is 71.3% (44,800 / 62,828) and 14.9% (9,333 / 62,828), or 86.2% with fewer than 10 (71.3% + 14.9%), counted by employees at one location and not in the whole firm.
NASBA reports 650,667 actively licensed CPAs on August 27, 2026, from 53 of the 55 licensing jurisdictions13. The IRS lists 886,338 individuals with current preparer tax identification numbers for 2026, as of September 1, 2026, among them 209,076 CPAs, 71,966 enrolled agents and 26,256 attorneys14. That count covers paid preparers of any kind, and the credential categories overlap. Neither count is of firms. A count of CPA firms is NF in NASBA's and the AICPA's published figures and in the Census Bureau's firm-size files; the nearest evidence is the office count above7. Because every firm with employees has at least one office, ~63,177 CPA firms with employees is a ceiling (from that office count; high, because a firm with several offices is counted once for each). It leaves out owners with no employees.
The U.S. Bureau of Labor Statistics counts 1,595,200 jobs for accountants and auditors in 2025 and projects 5% growth from 2025 to 203515. It reports a 2025 median pay of $83,680 a year, or $40.23 an hour, for the same occupation15. These count jobs in an occupation, not firms or offices, so they cannot say how many firms compete for the same clients or the same staff.
The supply of new accountants
AICPA figures describe the entry end. Graduates who earned a bachelor's or master's degree in accounting fell to 55,152 in the 2023-24 academic year, down 6.6% from the year before, according to the AICPA's 2025 Trends report1. Provisional federal data for 2023-24 show 40,817 accounting bachelor's degrees, down 3.3%, and accounting or taxation master's degrees down about 15% to 14,3351. The two counts add to the 55,152 total (40,817 + 14,335).
In 2021-22, 47,067 students earned a bachelor's degree in accounting, down 7.8% from the year before, and 18,238 earned a master's degree, down 6.4%16. Those add to 65,305 (47,067 + 18,238), against 55,152 in 2023-24, a fall of 10,153 (65,305 - 55,152), or 15.5%, over two school years. The later report counts master's degrees in accounting or taxation, so the comparison shows a direction and not an exact rate. Graduates had already fallen in 2019-20, by 2.8% at the bachelor's level and 8.4% at the master's level17; counts for that year are NF, and those rates are the nearest evidence. For the missing year, 2022-23, the count is ~59,000 (55,152 / 0.934, the one-year fall of 6.6% turned back into a starting count). Rounding in the 6.6%1 moves it by about 30 either way.
CPA Exam candidates moved more sharply. In 2023, 42,626 new candidates entered the pipeline, the highest tally since 2016, and the AICPA says such a surge typically comes the year before a major exam change1. In 2024 the count was 28,0821. That is a fall of 14,544 (42,626 - 28,082), or 34.1%. Because 2023 was a surge, we think the fall probably overstates any underlying trend. The AICPA's count of unique candidates sitting the exam, a different measure from new candidates, fell from 72,271 in 2021 to 67,335 in 202216, which is 6.8% and came before the 2023 surge. Candidate counts for 2019 and 2020 are NF; the nearest evidence is the 2021 and 2022 counts, and we do not line the two measures up as one series.
Public accounting firms that responded to the AICPA survey reported hiring 11,985 new graduates in 2024, but the AICPA says that, because of a low response rate by firms, the national total could not be projected with confidence1. The figure counts only the firms that answered, and it does not tell us whether firms are finding the people they want. Table 1 sets the figures side by side.
| Measure | Figure | Year | Source |
|---|---|---|---|
| Jobs for accountants and auditors | 1,595,200 | 2025 | 15 |
| Projected growth in those jobs | 5% | 2025 to 2035 | 15 |
| Median pay for accountants and auditors | $83,680 a year ($40.23 an hour) | 2025 | 15 |
| Accounting bachelor's degrees | 47,067, down 7.8% | 2021-22 | 16 |
| Accounting master's degrees | 18,238, down 6.4% | 2021-22 | 16 |
| Accounting graduates, bachelor's and master's | 55,152, down 6.6% | 2023-24 | 1 |
| Accounting bachelor's degrees (provisional) | 40,817, down 3.3% | 2023-24 | 1 |
| Accounting or taxation master's degrees (provisional) | 14,335, down about 15% | 2023-24 | 1 |
| Unique CPA Exam candidates | 72,271 | 2021 | 16 |
| Unique CPA Exam candidates | 67,335 | 2022 | 16 |
| New candidates entering the CPA Exam pipeline | 42,626 | 2023 | 1 |
| New candidates entering the CPA Exam pipeline | 28,082 | 2024 | 1 |
| New graduates hired by responding public accounting firms | 11,985 | 2024 | 1 |
Table 1. The supply of accountants: jobs and pay from the U.S. Bureau of Labor Statistics, and graduates, CPA Exam candidates and firm hires from the AICPA. Unique candidates and new candidates are different measures, and the last row counts only the firms that responded to the AICPA survey and is not a national total.
Graduates and new CPA Exam candidates are fewer in the latest years the AICPA reports, in an occupation whose jobs the Bureau of Labor Statistics projects to grow. The table does not show a shortage in any firm: it counts supply, and this paper does not measure vacancies or staff turnover. We think the safe inference is smaller than a claim of shortage. A firm that wins more work than its staff can do should not assume it can hire quickly to meet it, and Chapter 4 returns to this for busy season.
The working year
Sweeney and Summers followed public accountants at one national firm and reported an average workload of 49 hours a week before the busy season and about 63 hours during it18. The study is from 2002, so it shows a pattern and not today's hours. A survey of over 700 auditors found that they worked, on average, five hours a week above the level at which they believe audit quality begins to deteriorate, and often 20 hours above it at the peak of busy season19. It records perceptions, and neither study is a time-record study of a small practice.
The IRS had received 99,802,000 individual income tax returns by April 3, 2026, 114,333,000 by April 10 and 144,992,000 by May 820,21,22. Of the returns received by May 8, 68.8% were in by April 3 (99,802,000 / 144,992,000) and 78.9% by April 10 (114,333,000 / 144,992,000). The counts show when returns arrive and not when a firm's hours are spent, and they cover all filers, whether or not they use a preparer. We think they describe a year in which a firm's hours are scarcest in the first quarter and may be less scarce afterward. This paper does not measure how many hours firms leave unsold outside busy season.
How to get accounting clients: why choosing an accountant is hard
A prospective client can have trouble judging an accountant before hiring one, and sometimes afterward as well. Akerlof (1970) described an information asymmetry: sellers know more about the quality of a car than buyers do23. He argued that in the used-car market most cars traded will be the lemons and good cars may not be traded at all23. He also named institutions that counteract the uncertainty, among them guarantees, and licensing, including the licensing of doctors, lawyers and barbers, which certifies proficiency23. Dulleck and Kerschbamer (2006) survey the economics of credence goods and ask under which conditions experts such as doctors, mechanics and computer specialists have an incentive to exploit the informational problems of diagnosis and treatment24. Nakayama, Sutcliffe and Wan (2010) attribute to Darby and Karni (1973) the category of credence goods, which "cannot be evaluated in normal use" even after purchase25.
Our application to accounting is an inference. A client can rarely tell whether a return was prepared well or a set of books is right, so the client looks for signals that stand in for quality. A professional credential is one such signal. We think it certifies a floor and says little about fit, responsiveness or the kinds of client a firm serves well. The full argument is in the economics of choosing a professional.
Two studies of professional services markets add to this, though the passages we read do not mention accountants. Day and Barksdale (1992) write that purchasing professional services is different from buying physical products and many other services, and that because professional services are typically intangible and customized, "evaluation prior to purchase is difficult"26. Gluckler and Armbruster (2003) analyze the market for management consulting and argue that its defining features are institutional and transactional uncertainty, and that competitiveness there is based on "experience-based trust and networked reputation"27. Both sit at the lower grade of evidence in our files, and the second concerns consulting. We treat them as consistent with the economics above and rest no claim about accountants on either alone.
A software vendor and a trade publication have also surveyed clients and firms on how clients find an accountant. TaxDome† (practice management software) surveyed more than 350 US business decision-makers in 2025. Of them, 58% said they found their current accounting firm through a referral, 17% through online search and 3% by responding to an advertisement28. Those are business executives and not individual taxpayers, and 85% of TaxDome†'s respondents called responsiveness "very" or "extremely" influential when choosing a firm, a stated attitude and not observed behavior28.
Accounting Today and ADP† (payroll) surveyed accounting firms in 2019 about their clients. Unsolicited client referrals, and referrals from fellow professionals, were each named a top source of new clients by 79% of firms, solicited client referrals by 43% and advertising by under 10%29. Firms could name several sources, they gave opinions and not records, and the page gives no sample size. CPA.com's benchmark of client advisory services reports that 67% of responding practices named existing tax clients and 64% named external referrals as a top source of advisory leads, against 19% for the firm's website and 4% for advertising30. Those shares are for advisory leads only and count a different thing from TaxDome†'s, so we do not set them side by side.
The IRS's 2021 Comprehensive Taxpayer Attitude Survey, made up of 2,099 surveys of the public by telephone and online random sampling, lists paid tax professionals (88%) among the most valuable sources of tax advice and information31. That measures how taxpayers value professionals as a source of advice and does not by itself show how they choose one. The surveys on sources put referrals and existing clients first and paid advertising last, and we think that does not show that marketing is beside the point. A referral is a prospect who has heard of the firm, and this paper does not measure whether that prospect looks the firm up before calling.
Beyond responsiveness, these surveys give no ranking of what clients value in a firm, so the argument that follows still runs partly from the mechanism. If a client cannot judge the work, the marketing task is to put checkable facts in front of the client before the first call: who does the work, what credentials they hold, what the firm does and does not do, how an engagement starts and what it is likely to involve. Each fact also works as a filter. A page that says which clients the firm serves helps the right prospect recognize the firm and helps the wrong one move on.
Accounting firm marketing and capacity: fit before volume
This section is our argument, and its figures are cited. A firm short of hours still gains from demand, in a particular form. It cannot sell more hours than it has, but it can sell each hour for more and sell hours that would otherwise go unsold. Both need more inquiries than the firm will accept, because a firm that must take whoever calls cannot choose, cannot raise its minimum fee and has no one to call in a slow month. A firm short of hours therefore gains from demand in three ways: it chooses better-fitting and better-paying clients, it raises its minimum fee, and it fills the months outside busy season.
The profession has been raising price. The AICPA's MAP survey reported a median hourly billing rate, across all positions, of $159 in fiscal year 2022, up 16% from $137 in fiscal year 202032. Its 2025 edition reported a median 6.7% increase in firms' net client fees over the prior year, against 9.1% in the 2023 edition33. These are medians of firms that chose to take part, and they do not separate price from volume.
When hours are short, the measure that matters is the fee a client pays divided by the hours the client takes. An inquiry rarely shows the hours, so Table 2 solves for them: it divides each cited fee by the median billing rate of $159 an hour32 to give the hours the fee pays for at that rate. A client who takes more hours than that is billed below the median rate. The last column sets the same fees against the lowest plan from Technical Marketing plus the agent. That is $1,298 a month, or $15,576 a year (12 x $1,298): the $999 plan34 and the $299 agent34. The agent includes 250 call minutes, about 62 four-minute calls (250 / 4), and charges $0.39 a minute after that; the column takes it at $299, with its calls inside the included minutes. The column gives the new clients a year that cover that cost on first-year fees alone.
| Kind of client | Fee a year | Hours the fee pays for at $159 an hour | New clients a year to cover $15,576 |
|---|---|---|---|
| Advisory client of a CPA firm, median, 202435 | $17,867 | 112.4 | 0.9 |
| Monthly bookkeeping client, most common vendor band, 202536 | $3,000 to $5,988† | 18.9 to 37.7 | 2.6 to 5.2 |
| Business tax return, most common vendor band, 202536 | $1,000 to $1,499† | 6.3 to 9.4 | 10.4 to 15.6 |
| Individual return, most common vendor band, 202536 | $400 to $599† | 2.5 to 3.8 | 26.0 to 38.9 |
| Base Form 1040, NATP average, 202537 | $236 | 1.5 | 66.0 |
Table 2. For five kinds of client: the fee, the hours that fee pays for at the AICPA median billing rate32, and the new clients a year that cover the lowest plan plus the agent (calc.). The hours divide each fee by $159, the clients divide $15,576 by each fee, and the bookkeeping fee is the monthly band times 12; the $15,576 is 12 months of the $999 plan and the $299 agent34. The advisory figure is the median of 206 self-selected practices that sell the service well. The 1040 figure is the average among tax preparers whom NATP could reach. The three fees marked † are the most common bands among 219 customers of Ignition† (billing software), a sample that includes solo bookkeepers, so the low end of those bands partly reflects who answered. All fees are firm-reported price lists in different years and not what clients paid.
At the median rate, a base Form 1040 at NATP's average fee pays for an hour and a half of work, and an advisory client at the survey median pays for 112 hours, which is 76 times as many ($17,867 / $236). The number of new clients needed to cover the lowest plan and the agent therefore depends on which clients the firm markets for, which is the case for marketing for fit. The cost is also under a tenth of the fees a professional brings in: $15,576 a year is 8.2% of the median $189,695 in net client fees per full-time professional32.
The advisory client is not a one-year client. The same survey gives a median client turnover of 10%35. We infer from it a stay of about 10 years (1 / 0.10) and about $178,670 in fees over that time (10 x $17,867), using two medians. Elsewhere in the profession, 56% of clients had been with their firm for eight or more years and 34% for a decade or longer, in the Accounting Today survey of firms29. NATP's chart shows 87% returning clients and 13% new among the preparers it surveyed, though it does not say whether it counts clients or revenue38. TaxDome†'s survey reports that 26% of business clients leave their accounting firm because they have outgrown it28. Because clients stay, a first-year fee understates what a client brings, and a poor fit stays too.
A base Form 1040 and an advisory engagement draw on the same professional's hours at very different fees (Table 2), and a firm with more inquiries than hours can take the second kind first. In the CAS survey the median practice served 67 advisory clients and reported 14 new ones, against a median client turnover of 10%35. At 10% turnover, 6.7 of the 67 clients leave (10% x 67), so a practice that wanted to stand still would need about seven new clients, and the median reported about twice that. The survey's table gives no period for the 14.
A firm can also raise its minimum. Among tax preparers who charge a minimum fee plus a charge based on complexity, half charge between $150 and $328 for a base Form 1040 with Schedules 1-3, and the average is $23637. The middle half spans $178 ($328 - $150), so a firm's minimum is the firm's own choice. A firm with more inquiries than hours can raise it with less risk of an empty week, and a firm that must take every inquiry cannot. We think this is one of the main ways demand pays in a profession short of hours.
The third gain is in the months outside busy season. A firm whose hours are scarcest in the first quarter has hours to sell in the rest of the year, and work billed monthly or planned across the year does not wait for the filing deadline. Customers of Ignition† most often cited more than $2,500 a month for CFO and controller services (23% of firms) and more than $2,000 a year for tax planning and advisory services (25%)39. Both are open-ended top bands and not averages. The CAS survey's typical monthly fee is $3,00035. We think these are the clients who can fill the slower months, and they are the clients a plan has to reach.
A misfit client has a cost that the ratio does not show. The hours it absorbs come from other clients or from staff time outside working hours, and the entry-end figures in Table 1 give a firm no reason to assume it can buy them back quickly. Such a client may be displacing a better one who was never taken on, a cost the firm may never see.
Fit is partly visible before an engagement. A firm learns the type of entity, the complexity of the work, the state of the records, the timing of the request and the prospect's expectation about price in the first conversation, and several can be stated on a page so that the prospect sorts itself first. Chapter 4 describes what a page and a form may ask at the inquiry stage.
The lowest plan and the agent lose money in one corner: a firm that wants no new clients, would not choose among more of them, and whose intake is already sufficiently optimized, with nearly every call and form answered in time. That is narrower than a full calendar, which leaves the firm able to choose, to raise its minimum and to fill slow months. The cited base rates put the median firm away from the corner on growth: its fees grew 6.7% in a year33, and the median advisory practice needs about seven new clients a year to replace those it loses. Two sources point toward the corner. In the Accounting Today survey, 63% of firms said 6% or less of their clients were new each year29. And 52% of NATP's responding preparers (100% - the 48% who said their practice was growing38) did not report growth.
On intake, the estimate in the last section of this chapter puts the share of first calls that a firm without a full-time receptionist leaves unanswered well above zero. For a firm in the corner on both counts, the lowest plan and the agent are a cost with no product. This paper does not claim that most firms are there.
The strongest objection is that a small firm needs volume to survive and cannot afford to be selective. We think the objection is right for a firm with spare hours, where volume matters and fit matters less, and wrong for a firm at capacity, where the marginal client is served with hours taken from someone else. Choosing needs a choice, though: a firm at capacity that receives no more inquiries than it can serve cannot raise its minimum or turn down a poor fit.
A second objection is that fit cannot be known before the work begins. That is partly true, because a client can look simple and turn out not to be. It argues for a short, honest intake and for reviewing fit after the first year of work, not for dropping the filter. This paper does not measure how well any signal predicts hours. A firm's own records of hours by client would settle the question for that firm, and Chapter 6 describes how to count them.
What is searched for: CPA marketing, accounting firm marketing and AI accounting
Table 3 lists searches about accounting firm marketing and about AI in accounting: US monthly averages from Google Ads keyword data for November 2026, with the average cost per click where Google reported one.
| Search phrase | US searches a month | Cost per click | Source |
|---|---|---|---|
| ai accounting | 6,600 | $38.95 | 3 |
| ai for accountants | 6,600 | $38.95 | 3 |
| ai in accounting | 6,600 | $38.95 | 3 |
| cpa marketing | 1,300 | $12.31 | 3 |
| accountant marketing | 260 | $24.97 | 3 |
| accounting marketing | 260 | $24.97 | 3 |
| marketing for accountants | 260 | $24.97 | 3 |
| accounting firm marketing | 170 | $42.10 | 3 |
| seo for accountants | 170 | $10.08 | 3 |
| accounting practice management | 90 | $47.20 | 3 |
| ai for cpas | 90 | $49.33 | 3 |
| accounting firm website design | 70 | $13.22 | 3 |
| accounting lead generation | 70 | none reported | 3 |
| bookkeeping marketing | 70 | $27.63 | 3 |
| cpa firm marketing | 50 | $52.68 | 3 |
| how to get accounting clients | 50 | $7.85 | 3 |
| cpa seo | 40 | none reported | 3 |
| tax preparer marketing | 30 | $39.19 | 3 |
| accounting firm seo | 10 | none reported | 3 |
| how to get cpa clients | 10 | $4.78 | 3 |
Table 3. Searches about accounting firm marketing and AI in accounting, by US monthly volume, from Google Ads keyword data, November 2026. The volume for cpa marketing includes an unknown share of searches that concern the advertising model and not accountants.
The marketing phrases are small, and the largest is ambiguous. CPA is also an advertising abbreviation, for cost per action, and three further phrases in our data read like searches for that meaning: what is cpa marketing at 720 searches a month, cpa meaning marketing at 720 and cpa marketing network at 4803. An unknown share of the 1,300 monthly searches for cpa marketing3 therefore concerns the advertising model and not accountants, so we do not call it the leading phrase for accounting firms.
Three variants carry no such ambiguity: accountant marketing, accounting marketing and marketing for accountants draw 260 each3. The phrases are small throughout: nine of the 20 in Table 3 draw 70 a month or fewer, including how to get accounting clients at 50 and how to get cpa clients at 103. The phrase closest to this chapter's theme, accounting practice management, draws 903.
The three AI phrases that share a volume are far larger. The 6,600 is roughly 25 times the 260 for accountant marketing (6,600 / 260)3. Set against the 1,300 for cpa marketing, 6,600 is roughly 5 times as large, but that factor is a floor: because part of the 1,300 is not about accountants, the ratio for accounting firm marketing alone would be higher3. Only the comparison with the 260 variants avoids the ambiguity.
Those three AI phrases, ai accounting, ai for accountants and ai in accounting, show the same volume and the same cost per click, so they may be one pool of searches counted three times, and we do not add them. The three phrases at 260 searches and $24.97 a click show the same pattern3 and get the same treatment. We do not know how the keyword tool groups close variants, and the identical figures are the only sign that it does.
Three limits apply to all of Table 3. First, the phrases read as ones a practice owner or a vendor would type, not ones a prospective client would, so the table says nothing about the size of the demand a firm is trying to reach. Second, volume shows interest, not purchase. Third, cost per click is what advertisers pay for a click on an ad and not what a client is worth: among the phrases for which Google reported one, it runs from $4.78 for how to get cpa clients to $52.68 for cpa firm marketing3.
We cannot say who types the AI phrases: an owner, a staff accountant, a student and a software vendor are all possible. The profession's evidence on AI in accounting work is a separate question, taken up in Chapter 5. The table supports only this: anyone writing about accounting firm marketing is writing for a small searched audience, and anyone writing about AI for accountants is writing for a much larger one, whoever its members are.
Our interest, and the map of this paper
We state our interest in this subject once. Technical Marketing sells marketing plans for service businesses at $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 term34. It also sells an AI voice agent, added to a plan, at $299 a month with 250 call minutes included, then $0.39 a minute34. Our page for accountants and our pricing page say more about each. We gain if a firm buys either, and a reader should weigh what follows with that in mind. We have tried to write the conclusions so that they hold for a firm that buys nothing from us, and we say where the evidence is thin.
Each fee is set against the product it buys. The plan produces demand and the agent answers it, so Table 2 sets them together against new clients, and the chapters on intake and AI set the agent alone against answered calls and staff hours.
Our thesis is that an accounting firm is limited by hours before it is limited by demand, because it cannot sell more hours than it has. That limit changes what demand is worth to the firm and does not remove it, so fit, capacity and confidentiality should shape both its marketing and its use of AI.
The through-line is found, chosen, reached, within the rules, measured. Found is Chapter 2, on how a firm appears in local search and in AI answers and what it controls. Chosen is this chapter and Chapter 4, because a client chooses a firm and the firm chooses its clients. Reached is Chapter 4, on first contact, the website as a filter, and busy season. Within the rules is Chapter 3, which describes the professional and federal texts on advertising and client information and reduces them to design requirements. Measured is Chapter 6, with a first 90 days. Chapter 5 weighs the evidence on AI in accounting work and where it pays.
A measure of how often accounting firms miss calls is NF. Two neighbors are the nearest evidence and bound it, each in its own units, and we do not average them. In a 2024 secret-shopper test by Clio† (legal practice software), 40% of 500 law firms answered one phone call40. Law offices and CPA offices have similar sizes: 88.3% of offices of lawyers and 86.2% of CPA offices had fewer than 10 employees in the first quarter of 2024 (calc.)10,11,12,41. Among the health care clients of Invoca† (call tracking), 54% of inbound calls were answered by a person42.
Invoca†'s figure puts the unanswered share at 46% of calls (100% - 54% answered), and Clio†'s at 60% of firms on one call (100% - 40% answered). We estimate that an accounting firm without a full-time receptionist leaves a large minority of first calls unanswered by a person (from those two figures; high for a firm with a full-time receptionist). For conversion, a measure at accounting firms is NF; the nearest is Invoca†'s health care data, in which 43% of calls answered by a person were leads and 45% of those leads converted on the call42. We do not carry the health care figure over.
Chapter 2
SEO for accountants and what AI assistants answer
How a firm is found: SEO for accountants and local search
Google says that local results are mainly based on relevance, distance and popularity43, and that there is no way to request or pay for a better local ranking43. We think that sets the job for an accounting firm: be the relevant answer, be reachable by the searcher, and be a firm other people link to and review.
Reviews and links enter Google's description through one of the three factors. Google's help page says that this factor is also 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 ranking43. The page is a company statement about Google's own system, not an independent measurement of ranking weights. Its wording changes over time, and because it says results are mainly based on three factors, it does not claim to list every one.
We infer that the three factors differ in how far a firm controls them. Relevance is largely in its hands, because it depends on what the profile and the pages say. Popularity is partly in its hands, because reviews and links come from other people. Distance is not in its hands, since it is a fact about where the searcher and the office are. A firm that serves clients remotely and a firm that depends on nearby clients therefore stand in different places in local results. This paper does not measure what share of accounting clients are served each way.
Searches for help with SEO for an accounting firm are few: our keyword data show 170 searches a month for seo for accountants and 10 for accounting firm seo3. Those are phrases a firm or a vendor would type, not a client, so a table of such phrases cannot size the demand a firm wants to reach. The searches that matter to a firm's pipeline are the ones its prospective clients type, and this paper does not measure them. The nearest evidence is a survey of business buyers by TaxDome† (practice management software), whose limits Chapter 1 gives: of more than 350 US business decision-makers, 58% said they found their current accounting firm through a referral and 17% through online search28.
Accounting firm SEO and AI Overviews: what Google says
Google states 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."44 It also says that no specific optimization is required for either feature and that existing SEO fundamentals "continue to be worthwhile"44. A second Google Search Central page, dated 2024 in our files, says that SEO best practices remain relevant for these features and that there are no additional requirements to appear in them45.
These are Google's statements about its own products, not tests. They say nothing about how ChatGPT or Perplexity choose what to name, and our study, reported below, finds that the three assistants we asked name different firms from one another.
We think the work for Google's AI features is the work of ordinary search visibility: pages that are indexed and can be shown with a snippet, an accurate profile, and pages that answer the question a prospective client asks. A firm that is told it needs a separate technique to appear in Google's AI features should ask what the claim rests on, given that Google says none is required. That does not settle the question for other assistants, and the generative engine optimization section below gives the research on them.
What happens to clicks when an AI summary appears
In its 2025 analysis of US adults' browsing data, Pew Research Center found that users who encountered a Google AI summary clicked a traditional search result link in 8% of visits, against 15% of visits without one46. Pew's wording is that those who did not encounter a summary clicked on a search result "nearly twice as often". Users also ended their browsing session entirely after 26% of search pages with an AI summary, compared with 16% of pages with only traditional results46.
The same analysis shows how common summaries were. In Pew's dataset of 68,879 unique Google searches, 12,593 produced an AI summary46, which is 18% (12,593 / 68,879) and matches Pew's report that 18% of all the Google searches in its study generated one46. Of the summaries, 88% cited three or more sources and 1% cited a single source46. A summary that cites several sources can name several firms or pages at once.
These results cover US adults' searches in general and are not specific to local or professional-services queries. How often AI Overviews appear for accounting or tax searches is NF; the nearest evidence is from neighboring fields. Whitespark† (local SEO) ran 540 queries by hand in three cities for six industries, among them dentists, medical clinics and personal injury lawyers, and found AI Overviews on 15% of local-intent queries and on 92% of informational-intent queries47. Conductor† (SEO platform) reports them on 25.8% of the Google queries it analyzed for the Financials industry, which covers banks, insurance and financial services48. SE Ranking† (SEO software) found them on 125 of 300 finance keywords, or 41.67%, in September 2024 data49.
For accountants, AI Overviews appear on ~15% of local-intent searches (from Whitespark†'s 15% of local-intent queries across six industries). We estimate that they appear on more than a quarter of informational searches about tax and accounting questions, taking Conductor†'s share of all analyzed queries in the neighboring Financials industry as the floor; the figure likely errs low if tax questions behave like the legal and health keywords in SE Ranking†'s data, where the shares were 77.67% and 65.33%49. Each vendor chose its own queries, so none of these figures is comparable with another or with Pew's 18% of all searches46. Pew's comparison is also between visits with and without a summary, so it does not by itself show how any one search would have gone with the summary removed.
Two further studies describe clicks in general. SISTRIX† (SEO software) reported in a 2020 study an average click-through rate of 28.5% for the first organic position in Google and 2.5% for the tenth50. SparkToro† (audience research) analyzed clickstream panel data from Datos and reported that 58.5% of American Google searches in 2024 resulted in zero clicks51. Neither study is restricted to local or professional searches, the SISTRIX† figures are from 2020, and we rest no conclusion on either alone.
Many adults use chatbots, and search is still the default. In Pew's February 2026 survey of 5,119 US adults, 49% said they ever use AI chatbots and 42% said they ever use them to search for information52. In a Bain survey of 1,500 consumers in September 2025, 56% said they mostly or always use search engines as their default search tool, and 16% said they mostly or always use chatbots53. The two surveys count different things, ever-use and default-use, so they are not a trend. Both are survey answers, and neither shows how many people then contacted a firm.
A prospective client may also put a tax question to a chatbot before looking for a firm. A 2025 paper in Artificial Intelligence and Law tested language models on 390 questions, including perturbed versions, from the IRS-endorsed Volunteer Income Tax Assistance certification. Top models such as GPT-4 and Claude 3.5 Haiku reached 83.33% accuracy on basic multiple-choice questions, against 54.17% for GPT-3.554. These are certification questions and not a client's situation, and the result says nothing about how an assistant chooses an accountant. Chapter 5 weighs the evidence on AI in accounting work.
The strongest objection is that fewer clicks make SEO pointless. Pew's comparison is 8% of visits against 15%46, a lower rate of clicking and not none, and Google says that existing SEO fundamentals remain worthwhile for its AI features44. We think the fundamentals are the entry ticket to both the click and the citation and do not compete with AI visibility.
We think two consequences follow. A firm's name, location and credentials as they appear inside an answer matter more when fewer searchers reach the page. And a firm that measures only visits to its website will undercount the effect of being named. AI referrals to an accounting practice's site are NF, as is how often a name in an AI answer leads to a call or an engagement at any professional practice; the nearest evidence for the referrals is below.
Conductor† puts AI referrals at 1.08% of all traffic to its enterprise customers' sites across 10 industries and at 0.48% for the Financials industry48,55. Microsoft Clarity† (analytics) reports less than 1% on more than 1,200 publisher and news sites56. AI referrals are ~1% or less of an accounting firm's site visits (from Conductor†'s 1.08% and 0.48% and Clarity†'s less than 1%). We cannot say which way the estimate errs for a small firm, because every dataset is of larger sites. A referral count also covers clicks only: OpenAI's help page says a mention counts when ChatGPT shows a result for a local business, even if no one opens the listing57, so a name that is shown and not clicked is not counted.
For conversion, the nearest evidence is outside professional services. Clarity† found sign-up conversion of 1.66% for LLM traffic against 0.15% for search on publisher sites56, and Adobe† (analytics) reported that AI-referred traffic to US retail sites converted 90% better than other traffic in September 202658. Neither is an accounting firm, and we do not carry either figure over. Chapter 6 takes up measurement.
What AI assistants said about accountants and bookkeepers in our study
Each question in our study had the form "Who is a good dentist in Atlanta, GA? Name specific businesses.", with the city and the kind of business changed2. Our study, with its data, is published in full, and the study's method is described in the professions paper. Table 1 sets the accountant row beside the bookkeeper row and the study totals.
| Group | Businesses named | Named by all three | Named by only one | Source |
|---|---|---|---|---|
| Accountants, 16 cities | 272 | 0 (0%) | 90.4% | 2 |
| Bookkeepers, 3 cities (small sample) | 50 | 3 (6%) | 84% | 2 |
| Ten kinds of business asked in all 16 cities (160 questions) | 2,414 | 84 (3.5%) | not reported | 2 |
| All 199 questions | 2,975 | 122 (4.1%) | 79.7% | 2 |
Table 1. Accountants and bookkeepers against the study totals in our study of local AI answers. The bookkeeper row rests on three cities, a small sample.
For accountants, the assistants named 272 businesses in 16 cities, 17 per question, and none was named by all three2. That is below both totals on agreement: 3.5% for the ten kinds of business asked about in all 16 cities2, and 4.1% across all 199 questions2. On the share named by only one assistant, 90.4%, the accountant row is above the 79.7% for all 199 questions2. The published totals for the 160-question group do not report that share, so the comparison is with the all-question figure. The group's share lies between 75.0% and 98.2% (calc.), because the other 561 names (2,975 - 2,414) came from the other 39 questions.
Per question, the assistants named 17 distinct businesses against about 15 for the study as a whole (2,975 names / 199 questions)2. A longer combined list can lower the overlap mechanically, and we have not tested whether that accounts for any of the gap. Financial advisors, a neighboring row, show a similar pattern: 5 of the 281 businesses named (1.8%) were named by all three and 87.5% by only one2.
The bookkeeper row rests on three cities. The assistants named 50 businesses, 16.7 per question2, so the row covers three questions (50 / 16.7), and a handful of businesses decides the 6%. We do not read the gap between 6% and 0% as a difference between bookkeepers and accountants. For the ten kinds asked about in every city, the share named by all three ranged from 1.9% in Charlotte to 7.9% in Seattle2. That range is not reported for accountants alone, and we mention it because even the highest city share is a small minority.
The strongest objection to taking any of this seriously is that answers this unstable are noise and can be ignored. Low agreement means no single assistant stands in for the others, so a firm that checks only one learns little about the rest. It does not show that the answers are random, because we asked each question once2, and it does not show that prospective clients act on them, since our study did not follow anyone from an answer to a call, and the conversion figures above come from retail and publisher sites. Whether to watch the answers is then a question of cost. A monthly check, described at the end of this chapter, uses staff time, which is the scarce resource, so it should be small, and a firm can decide on its own numbers whether the answers matter to it.
How the three assistants differ, and which sites they cite
The cited figures on how the assistants answer are study-wide averages, not broken out by kind of business, so they describe the assistants and not accountants in particular. Table 2 gives them across all 199 questions.
| Assistant | Businesses named per answer | Websites cited per answer | Answers citing Reddit | Source |
|---|---|---|---|---|
| ChatGPT (model gpt-6-luna) | 4.8 | 3.5 | 2% | 2 |
| Gemini (model gemini-3.8-flash) | 6 | 6.5 | 42% | 2 |
| Perplexity (model sonar) | 7.8 | 7.7 | 0% | 2 |
Table 2. How ChatGPT, Gemini and Perplexity answered across all 199 questions in our study, as averages per answer.
Any two assistants overlapped on between 11.9% and 14.3% of the businesses they named2. All three named the same business first in 9 of 199 questions, and at least two agreed on the first name in 752. The sites they cited differ even more. The 597 answers cited 1,455 different websites, for 3,530 citations in all2, and 88% of the websites cited were cited by only one assistant2, against 79.7% of the businesses named2.
The most-cited sites are not firms' own sites. 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 forum2. Websites cited in only one city, where a local business's own site would fall, drew 39% of citations2. The most-cited site overall was reddit.com, in 87 answers2. Gemini cited it in 83 of 199 answers2, ChatGPT in 2% and Perplexity in 0% (Table 2). Perplexity's most-cited site was bbb.org, in 58 of 199 answers2, and ChatGPT's most-cited site appeared in 32 of 1992.
Much of what the assistants cite is not the firm's own site, so we infer that a firm's name may reach an answer largely through other people's pages. The citation figures cover all 23 kinds of business, so we cannot say which directories or list sites matter for accountants. The method limits apply here as they do to the names2.
Why the assistants may disagree about accountants: a conjecture
We do not know why the assistants agree so little on accountants. The explanations below are our conjectures, and none was tested in our study.
One is different sources. The assistants largely cited different pages, since 88% of the websites cited were cited by only one assistant2. If an assistant names the firms its pages mention, different pages would produce different names. The evidence shows that the sources differ; that they produce the different names is our conjecture.
Another is a wide field of similar firms. If a city has many firms offering similar services, and few signals set a handful of them apart, each assistant's choice among them may be close to arbitrary. The Bureau of Labor Statistics counts 63,177 offices of certified public accountants across the country7, which are offices and not firms. Our study did not count firms city by city, so we cannot say how wide the field was in any of the 16 cities.
A third is the matching of names. Names were matched across assistants by spelling2, so a firm that appears under one form of its name in one list and another form in another would be counted as two businesses. Part of the 0% for accountants2 may then reflect spelling and not disagreement. Our study did not count how many names were split in this way in the accountant answers.
A fourth is single asking. Each question was asked once2. If an assistant gives different names when the same question is repeated, part of the gap between assistants may be a gap between runs, and our design cannot separate the two.
Other people have repeated questions, though not about accountants. SparkToro† and Gumshoe.ai† (AI tracking) had 600 volunteers run 12 prompts through ChatGPT, Claude and Google's AI, 2,961 runs in all, and reported less than a 1 in 100 chance that a tool asked 100 times would give the same list of brands in any two responses, and less than 1 in 1,000 that it would give the same list in the same order59. That study is not peer reviewed, its author discloses a possible conflict of interest, and its prompts were not about accountants or the other professions. A repeat-asking study of AI recommendations for local businesses or professional practices is NF; the nearest evidence is the study just described.
We expect repeated asking to change our accountant lists too. Our single-asking figure of 90.4%2 therefore probably errs high as a measure of disagreement between assistants, because run-to-run change is counted as disagreement, but we do not know by how much.
A repeat would settle it: ask each question several times per assistant on different days, match the names by hand against the firms' own sites, and record each model and date. The agreement between runs of one assistant could then be set against the agreement between assistants, which would show how much of the 90.4% named by only one assistant2 is disagreement between assistants and how much is noise.
Generative engine optimization for CPA firms: what the research found
Generative engine optimization, or GEO, is the practice of changing content so that a generative search engine is likelier to use and cite it. The paper that introduced it reports that its methods can raise a source's visibility in generative engine responses by up to 40%60. The same paper reports that the effectiveness of its strategies varies across domains, "underscoring the need for domain-specific optimization methods"60. Its authors describe generative engines as black boxes that leave content creators with "little to no control over" when and how their content is displayed60.
The result is a best case under the authors' own test conditions ("up to" 40%60) and not an expected gain, and this paper has not tested the method on accounting firms or on local queries. A firm should treat GEO as a set of hypotheses to test on its own pages. A vendor that promises a result from it promises more than the research does, and we promise none.
Separate research asks how far the answers can be trusted. In a 2023 audit of commercial generative search engines, only 51.5% of generated sentences were fully supported by citations61, and only 74.5% of citations supported their associated sentence61. The audit covers the systems as they stood in 2023, and the engines may have changed since. For a firm the implication is a habit: read what the assistants say about the firm and check it against the firm's own pages, because a cited page may not say what the answer says it does.
What an accounting firm controls
A firm controls its inputs, not its outputs. Google says a better local ranking cannot be requested or paid for43, and the firm does not control which firms an assistant names. What it controls are the facts that exist about it and the pages on which those facts sit.
The first input is the business profile. Google's description of local results rests on relevance, distance and popularity43, and we think the profile is where relevance is stated: the firm's name, address, hours, categories and services, written the same way everywhere. The second is the listings on the directories and review platforms that the assistants cite most. We cannot say which of them matter for accountants, but the same name, address, phone number and credentials on each of them may remove one reason for an assistant to treat one firm as two.
A third input is a page for each service and each kind of client. A page for individual returns, one for small-business bookkeeping and one for a nonprofit are examples, and each page says who the work is for, what is included and what is not, who does it, and how a prospect starts. In our approach each page opens with a short answer to the question it targets, in a form an assistant could quote. The fourth input is credentials, stated exactly: who holds which credential, and what the firm does not offer. Chapter 3 describes the texts on advertising and misleading claims.
Reviews are the fifth. Google says that more reviews and positive ratings can help a business's local ranking43. How clients rate accountants online, and what a review is worth to an accounting firm, are NF. The nearest evidence is general or from medicine. BrightLocal† (local SEO software) reports that 97% of 1,002 US adults on a survey panel read reviews for local businesses, across all kinds of business62. In a study of one large single-specialty physician practice, mean new-patient volume fell 2.3% to 2.6% after a 1-star review63. We take the second as an indication of the direction of the effect and not of its size for an accounting firm. A firm asking for reviews works inside rules, and a public reply to a review can touch client information. Chapter 3 covers both. Our pages for accountants and bookkeepers describe our service for each trade.
A monthly answer check, as a design
A monthly answer check is a design and not a result. We have not run it on accounting firms, and this paper does not test whether it changes what any assistant says. It borrows the structure of our study at a size a firm can manage.
- Fix a list of questions a prospective client would ask, such as who is a good accountant for a small business in the firm's city, and write each question the same way every month.
- Run each question in each assistant once on the same day each month, and save the full answers.
- Record which firms were named and in what order, and which sites were cited.
- Compare the month with the last, and mark any statement about the firm that is wrong, such as a credential, a service or an address.
- Correct what the firm controls, such as the profile, the listings and its own pages, and note what it was.
The check carries our study's limits: each question was asked once, and run-to-run variation is unmeasured2. A single month's change in one assistant is therefore weak evidence, and a pattern over several months across all three is stronger. Chapter 6 treats the check as one measurement among others, and measurement in depth is argued in the professions paper.
Chapter 3
Advertising rules for CPAs and tax preparers: what the texts say
No single text covers how an accounting firm markets itself. This chapter reads the texts that bear on a firm's advertising, its solicitation of clients, its handling of the information that prospects and clients give it, and its reviews, and turns each into a requirement for a page, a marketing email, an intake form or a vendor contract. Each text is described as the claim in our evidence file prints it, and how it applies to a particular firm is left open.
Reading CPA advertising rules without giving advice
The texts that touch an accounting firm's marketing have different authors and different reach, and none of the claims says which of them binds a given firm. The chapter reads one state board's rule, the Texas State Board of Public Accountancy's, and no other board's, and it does not cover statutes on marketing email or text messages, or enforcement cases against accounting firms over their advertising. Texts change, and a description in a paper does not replace the current wording.
Dulleck and Kerschbamer survey the economics of credence goods, asking under which conditions experts such as doctors, mechanics and computer specialists have an incentive to exploit the informational problems of diagnosis and treatment24. We think an accountant's work often has that shape, because many clients cannot readily check a return or a set of books. Akerlof named guarantees as one obvious institution among the numerous that arise to counteract the effects of quality uncertainty23. A firm therefore wants to reassure a buyer who cannot check the work, and the professional texts limit the promises it may make. The long paper argues the theory of choosing a professional at length.
The AICPA advertising rule and promotion done through others
The AICPA Code's Advertising and Other Forms of Solicitation Rule (1.600.001) says a member in public practice shall not seek to obtain clients by advertising or other forms of solicitation in a manner that is "false, misleading, or deceptive", and it prohibits solicitation by coercion, over-reaching or harassing conduct64. The interpretation numbered 1.600.010 says promotional efforts would be considered false, misleading or deceptive if they "create false or unjustified expectations of favorable results"64.
The first sentence is the general test of falsity, misleading statements and deception. The second names one way a statement can mislead, by creating an expectation of results, and we think a headline can do that while being accurate word for word. For a tax practice a reviewer should look first at a refund, a saving or an outcome with a tax authority; for a bookkeeping or advisory firm, at growth and lower costs. These are our examples. We think a page that says what the firm does, for whom and how, and leaves the result to the engagement, gives the second sentence little to catch.
Interpretation 1.600.010 adds a point about promotion by other people. It says a member who performs services for a client or customer of a third party should determine that the third party's promotional efforts comply with the advertising rule, because "members must not do through others what they are prohibited from doing themselves"64. The claim addresses that member and does not mention an agency that the firm hires to write its own copy, or a directory or a platform. The reasoning travels, though. We think a firm should treat everything a vendor publishes in its name as its own advertising and approve each claim before the claim goes live. Whether the Code reaches a particular arrangement is for counsel.
A related risk is a vendor that promises rankings, leads or revenue. Google's Business Profile help page says there is no way to request or pay for a better local ranking on Google43, so a promise of one is a promise about something the firm does not control. We promise no ranking, no fixed number of leads and no revenue, and we think a vendor that does is making a claim to be tested against the unjustified-expectations sentence. The two guarantees we offer are of another kind: a measure agreed after an audit and set against the firm's own baseline, with more work as the remedy if it is missed34.
A state board's advertising rule: Texas on praise, comparisons and testimonials
A state board of accountancy is another author of advertising rules. This chapter reads the Texas State Board of Public Accountancy's rule at 22 TAC 501.82. Subsection (b)(1) treats as a false, fraudulent, misleading or deceptive advertising claim, among others, self-laudatory statements not based on verifiable facts, untrue comparisons with other accountants, and testimonials or endorsements not based upon verifiable facts65. We read the rule on a reproduction that its publisher updates quarterly.
The AICPA text speaks of advertising that is false, misleading or deceptive64. The Texas rule names three kinds of claim: praise that cannot be verified, comparisons that are untrue, and testimonials that cannot be verified. We think the design consequence is the same as before and sharper for a firm licensed in Texas. A line such as "the best CPA in town" is self-praise, and a line that the firm beats other accountants is a comparison, so each needs a fact behind it that a reader could check. A testimonial needs the same. Whether a given claim is covered is open.
Circular 230 section 10.30 and solicitation before the IRS
Circular 230 section 10.30(a)(1), a Treasury Department regulation at 31 CFR 10.30, bars a practitioner from using or participating in "any form of public communication or private solicitation" that contains a false, fraudulent or coercive statement or claim, or a misleading or deceptive one, with respect to any IRS matter66.
We think four features of that sentence matter for design. First, it is tied to an IRS matter, so pages about notices, audits, collections, penalties or representation are the ones to read against it first; whether a page is with respect to an IRS matter is open. Second, it names a private solicitation as well as a public communication, so a cold email or a letter to a named prospect sits beside a web page. Third, the words "use or participate in the use of" put copy that a vendor writes and the firm publishes within reach of the sentence66. Fourth, it names coercive statements. We expect pages about notices and penalties to invite fear-based copy, because fear is the obvious lever for them, and a firm that leads with what it does and what the reader may ask, and not with a threat or deadline pressure, gives the word little to attach to.
The AICPA rule and section 10.30 share some wording and attach to different things: the AICPA text to seeking clients, and section 10.30 to IRS matters. One piece of copy, such as a landing page for people who have received an IRS notice, may need to be read against both. The cited claim for section 10.30 names a practitioner and does not print the definition of that word, so who is covered is a question for counsel.
Confidential client information and the vendors a marketing system uses
The AICPA Code's Confidential Client Information Rule (1.700.001) says a member in public practice shall not disclose any confidential client information without the specific consent of the client64. Interpretation 1.700.040 says that before disclosing confidential client information to a third-party service provider a member should either enter into a contractual confidentiality agreement with the provider and obtain reasonable assurance that it has procedures to prevent unauthorized release, or obtain the client's specific consent64.
A marketing system touches client and prospect information in the contact form, a chat widget, scheduling, call handling, the email platform, a customer database, call tracking and analytics. Someone who writes into a form that a return was rejected and asks the firm to look at it has put something there that the firm now holds. Whether that is confidential client information, and whether the tool's maker is a third-party service provider within interpretation 1.700.040, are questions for counsel. A design that limits what the form collects leaves fewer of those questions to ask.
The two options in the interpretation suggest five questions for any vendor, ours included:
- Will it sign a confidentiality agreement, and what procedures does it describe for preventing unauthorized release?
- Which of its staff and subcontractors can reach the information, and where are they located?
- What does it keep, for how long, and how is the information deleted?
- If information is released, whom does it tell and how fast?
- Does it use what the firm sends it for any purpose beyond serving the firm?
A marketing system now often includes an AI feature, such as a chat widget or a voice agent. An August 5, 2026 article in the Journal of Accountancy by the communications manager of the AICPA Professional Ethics Division says the Code still provides clear, enduring guidance despite AI, and that members' core responsibilities of competence, due care, confidentiality and independence do not disappear because new technology is involved67. It adds that, before pasting client data into any AI system, a member should ask whether it is confidential client information, whether consent is required and has been obtained, and whether the use could result in disclosure to third parties67. An AICPA Code interpretation on generative AI is NF; the nearest evidence is this article, which is staff-written and non-authoritative. Chapter 5 returns to AI in the firm's own work.
Tax preparer marketing and tax return information: Sections 7216 and 6713
Table 1 sets out the provisions on tax return information, with the amounts and conditions the claims print. Two are statutes and three are Treasury regulations.
| Provision | What it addresses | What the claim prints |
|---|---|---|
| 26 U.S.C. 7216 | Knowing or reckless disclosure or use of information furnished for a return, other than to prepare the return | Misdemeanor: fine of not more than $1,000 ($100,000 for disclosure or use covered by section 6713(b)), imprisonment of up to 1 year, or both68 |
| 26 U.S.C. 6713 | Disclosure or use of return information other than to prepare the return | Civil penalty of $250 per disclosure or use, capped at $10,000 for any calendar year69 |
| 26 CFR 301.7216-3 | Disclosure or use of a taxpayer's tax return information | Written consent from the taxpayer first, unless section 7216 or 301.7216-2 authorizes it4 |
| 26 CFR 301.7216-1(b)(2) | Who counts as a tax return preparer | Includes a person engaged in the business of providing auxiliary services in connection with preparing returns, including one who develops software used to prepare or file a return70 |
| 26 CFR 301.7216-2 | Disclosure inside the firm | Allowed within the United States without consent; consent under 301.7216-3 required before any disclosure to a person located outside the United States71 |
Table 1. The tax return information provisions as the evidence prints them. Amounts and conditions are the claims' own, not the full provisions.
The fourth row defines a tax return preparer to include any person engaged in the business of providing auxiliary services in connection with the preparation of tax returns, and any Authorized IRS e-file Provider70, so the text reaches beyond the firm's own staff. Whether a marketing agency, a call-answering service or a scheduling tool provides such services is for counsel, and the question matters less to a marketing system that never receives return information. The fifth row ties consent to a recipient located outside the United States71, which is one reason to ask a vendor where its staff sit.
None of these claims says anything about a prospect's first inquiry; they concern information furnished for a return. Whether a given inquiry contains any, and whether a firm uses return information by adding a person to a mailing list, are for counsel. We think an inquiry form that asks for none of it, and a marketing list kept apart from return data, keep both questions from arising. Chapter 4 applies this to the intake form.
The FTC Safeguards Rule and the written information security plan
The FTC says section 314.2(h) of the Safeguards Rule lists examples of financial institutions covered by the Rule, among them "tax preparation firms"72. IRS Publication 5708 (Rev. 8-2024) says that under the GLBA and the Safeguards Rule tax and accounting professionals are considered financial institutions, "regardless of size", and that the Rule requires a written, accessible information security plan, which it calls a WISP73. Under 16 CFR 314.4(j), a financial institution must notify the FTC as soon as possible, and no later than 30 days after discovery, of a notification event involving the information of at least 500 consumers74. Section 314.5 states that this paragraph is effective as of May 13, 202475. The cited claim does not print the definition of a notification event.
A marketing site is a data system. A form that collects a name, a phone number, an email address and a sentence about the person's situation creates records, and a call-handling tool creates more. The written plan is the firm's own document, but the marketing system can hand its author a list of every tool that receives personal information, the fields each collects, where the data is stored, how long it is kept, who at the vendor can see it, and how the vendor will tell the firm if something is released. Whether the information a prospect gives is covered by the Rule is open.
The less a form collects, the shorter that list. We think the confidentiality interpretation, Section 7216 and the Safeguards Rule point the same way: collect less, and name every party that touches what is collected. That is a design principle, not a statement of what any of them requires.
Reviews and testimonials: the FTC rule on fake and bought reviews
Google's help page says more reviews and positive ratings can help a business's local ranking43, and Chapter 2 sets out what Google says. A firm may be tempted to shape its reviews, and the FTC's rule on consumer reviews and testimonials, 16 CFR part 465, speaks to that. It was published in the Federal Register on August 22, 2024 and took effect October 21, 202476.
Under 16 CFR 465.2 it is an unfair or deceptive act for a business to write, create or sell a consumer review or testimonial that materially misrepresents that the reviewer exists, that the reviewer used the product or service, or the reviewer's experience77. The FTC said the rule addresses reviews that misrepresent that they are by someone who does not exist, "such as AI-generated fake reviews"78. Under 16 CFR 465.4 it is an unfair or deceptive act for a business to provide compensation or other incentives in exchange for consumer reviews expressing a particular sentiment, positive or negative79. Under 16 CFR 465.7 it is 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 removed80.
On penalties, the FTC said in its December 22, 2025 warning letters under the rule that violations can bring a lawsuit or other legal action and civil penalties of up to $53,088 per violation81. A notice the FTC published in September 2026 said its civil penalty amounts will remain unchanged during 202682. The cited claims do not print the rule's definitions of a consumer review or a testimonial, so whether a given review of an accounting firm is covered is for counsel.
We think four habits follow from the text:
- Ask every client, in the same words, at the same point in the engagement.
- Offer nothing in return for a review.
- Never write, edit or buy a review, and keep an agency from doing so.
- Answer a negative review without a threat, and without confirming or discussing any client's affairs.
The last habit rests on the Code's confidentiality sentence, described earlier, because a public reply is a disclosure channel. In dentistry, HHS's Office for Civil Rights said in 2019 that a Dallas dental practice agreed to pay $10,000 and adopt a corrective action plan to settle potential violations of the HIPAA Privacy Rule83, and the resolution agreement recites a patient's complaint that the practice, replying to her Yelp review, disclosed details of her treatment plan, insurance and cost information84. That is a different rulebook and profession, and it shows only how a reply to a review can carry detail it should not. We think a conservative reply confirms nothing about whether the reviewer is a client and gives a telephone number.
A testimonial that a firm publishes about itself is the firm's own promotion. We think it should be read against the unjustified-expectations sentence if it reports a result; against the consent sentence if it names a client, with the client's specific consent on file; and, for a firm licensed in Texas, against that board's treatment of testimonials and endorsements that are not based upon verifiable facts65.
Investment advice, the SEC marketing rule and FINRA's content standard
A firm that also advises on investments or sells securities carries more texts, and the rest belongs to a securities lawyer. The SEC's small entity compliance guide says the investment adviser marketing rule (Rule 206(4)-1) became effective May 4, 2021, with a compliance date of November 4, 202285. The guide says the rule permits testimonials and endorsements in an advertisement if the adviser satisfies disclosure, oversight and disqualification provisions, including clear and prominent disclosure of whether the promoter is a client and whether the promoter is compensated85. FINRA Rule 2210(d)(1)(A) requires member communications to be based on principles of fair dealing and good faith, to be fair and balanced, and to provide a sound basis for evaluating the facts86. Whether either text reaches a particular firm is for its compliance lead. For a marketing design the consequence is only that testimonial and endorsement features, and any paid promoter, get a second review before launch.
Design requirements for a site, a marketing email, an intake form and a vendor
Table 2 reduces the texts above to design requirements for the four places a marketing system meets them. Each row is what we think a conservative build would do.
| Text | Site page | Marketing email | Intake form | Vendor |
|---|---|---|---|---|
| AICPA 1.600.001 and 1.600.01064 | No promised results, savings or refund amounts. Credentials, fees and services stated as they are. Every claim is checked against the texts before it publishes. | Same review as a page. No repeated follow-up after a prospect declines. | Confirmation text says what happens next, not what the prospect will get. | Firm approves each claim before it goes live, including copy a directory or agency writes. No vendor promise of a ranking, a fixed number of leads or revenue. |
| Texas State Board of Public Accountancy, 22 TAC 501.8265 | For a firm licensed in Texas: no self-praise without a verifiable fact behind it, no untrue comparison with other accountants, no testimonial or endorsement without verifiable facts. | Same review as a page. | Not applicable. | Copy that praises the firm, compares it with others or quotes a client is checked against verifiable facts before it goes live. |
| Circular 230 section 10.3066 | Pages on notices, audits and collections describe the service, not an outcome, and avoid fear-based urgency. | Letters and emails to a named prospect get the same review as public pages. | Text shown beside the form on an IRS-matter page is read against the section like the page itself. | Copy the vendor writes for these pages is reviewed before it publishes. |
| AICPA 1.700.001 and 1.700.04064 | No client named or quoted without that client's specific consent on file. | No client details in marketing email. Marketing lists built apart from client files. | Collect the least that lets the firm answer. Say where the answer goes. | Confidentiality agreement and reasonable assurance of the vendor's procedures, or the client's specific consent. Ask who can reach the data. |
| Sections 7216 and 6713 and the regulations4,68,69,71 | No tax return information published, or invited in comments or chat. | Marketing lists not built from return data without an approved consent design. | No taxpayer identification number, income figure, bank detail or return document at inquiry. | Ask where the vendor's staff are located. Ask whether the vendor falls within the preparer definition. Keep return information out of marketing systems. |
| Safeguards Rule and Publication 570873,74 | The site and its tools are listed in the written plan. | The email platform is listed with the fields it holds. | The field list matches the plan, and a retention period is stated. | The vendor is named in the plan, with how it will tell the firm of a release. |
| FTC reviews rule, part 46577,79,80 | Reviews shown are real. None bought, written or edited by the firm or an agency. | Review request goes to every client alike, with nothing offered for it. | No review prompt at intake. | Agency never writes, buys or filters reviews and never threatens a reviewer. |
| SEC marketing rule and FINRA 221085,86 | If the firm advises on investments, testimonial and endorsement features get a second review by its compliance lead. | Same second review. | Not applicable. | Compliance lead approves any paid promoter or endorsement arrangement before launch. |
Table 2. Design requirements that follow from each text. They are not a statement of what any text requires of a particular firm.
The long paper compares how one design meets the dental, medical, legal and accounting rulebooks in its chapter on one system and four rulebooks.
What stays with the firm's counsel, and where the evidence is thin
This chapter leaves five questions for counsel. What do the advertising rules of the board that licenses the firm's CPAs say, and how do they differ from the AICPA text and from the Texas rule read above? Which laws govern marketing email and text messages? Does a prospect's first inquiry contain information that the tax return provisions cover? Is a given vendor a third-party service provider, or a tax return preparer within the regulation's definition? Which of the securities texts, if any, reaches the firm? Each turns on the current wording of the texts. This paper cannot settle them, and a vendor should not.
We sell marketing systems, so we gain if a firm believes that sound design is possible inside these rules. That is a reason to check this chapter's sentences against the texts they cite instead of taking our reading of them. Chapter 4 turns the requirements into an intake form and a phone line, and reads the texts on AI voices and call recording.
Chapter 4
Accounting firm website design, intake and busy season
An accounting firm's website has two jobs before anyone picks up the phone: to let the right prospect find what is needed to decide, and to let the wrong prospect decide not to call. The intake form and the phone line add a third, which is to protect the firm's hours and its clients' information. This chapter treats first contact as the stage where a firm chooses its clients.
Accounting firm website design as a filter
A firm short of hours still gains from demand, because it can choose among the people who ask: it takes the prospect who fits and pays well, raises its minimum fee and fills the months outside busy season, as Chapter 1 argued. The website is where that choice begins. Stigler's 1961 paper offers a way to see why. He wrote that the identification of buyers and sellers reduces drastically the cost of search, and he called advertising the obvious modern method of identifying them87. He held that the optimum amount of search is found when its cost is equated to its expected marginal return87. A person choosing an accountant is searching, and each page that answers a question lowers the cost of the next step. We think the same logic holds on the other side: every inquiry from a person the firm cannot serve costs staff time, so a page that tells that person early lowers the cost for both.
Search interest in this subject is small in our data. In Google Ads keyword data for November 2026, accounting firm website design was searched about 70 times a month in the United States with a cost per click of $13.223, and accounting lead generation about 70 times a month with no cost per click reported3. These are phrases a firm owner or a vendor would type, not a prospective client. Volume shows interest, not purchase, and a cost per click is what advertisers pay, not what a client is worth.
A page that filters well lets a prospect learn seven things before calling:
- Who the firm serves, and who it does not, such as individuals, households, owner-run businesses or particular industries.
- Which services it offers and which it does not.
- How an engagement begins and what the first conversation covers.
- How fees are set: a published price, a range, or a quote after a defined step. The firm chooses which, and the page says which.
- Who does the work and who reviews it.
- Whether the firm is taking new clients now, and what happens if it is not.
- How to reach the firm, how soon a reply comes, and what the firm will not discuss on a first call.
We think specific answers matter more than warm ones. Spence distinguished observable, unalterable attributes, which he called indices, from observable characteristics the individual can manipulate, which he called signals88. Words on a website are signals in his sense, because any firm can write them. A sentence every firm could write, such as a promise of personal service, tells a prospect little. A sentence a firm can be held to, such as the day by which it answers an inquiry or the kinds of client it declines, tells more because it costs something to say. The 1986 working paper that introduced the SERVQUAL scale measured five dimensions of service quality: tangibles, reliability, responsiveness, assurance and empathy89. We expect a prospect to meet assurance first in the page and responsiveness in the reply. Our page for accountants describes what we build.
Accounting lead generation and the intake form
The intake form is the first filter that asks a question, and a question costs the prospect effort. It should ask what decides fit and nothing else. Table 1 sets out each field of an inquiry form and what it is for.
| Question or field | What it settles | At inquiry? |
|---|---|---|
| Who the work is for: an individual, a household, an owner-run business or another entity | Whether the firm serves this kind of client at all | Yes |
| The service wanted: preparation, bookkeeping, payroll, advisory or a response to a tax authority | Which staff member replies, and whether the firm offers it | Yes |
| Rough size, in plain categories such as one employer or several | Whether the work fits the hours the firm has | Yes, as choices and not as figures |
| The date the person needs the work done by | Whether the firm can meet it | Yes |
| How the person found the firm | The count of inquiries by source that Chapter 6 describes | Yes, one optional field |
| Preferred way and time to be reached | The speed and fit of the reply | Yes |
| A free-text box | Anything the choices missed | Yes, with a label saying what not to type |
| Taxpayer identification number, income figures, bank details, return documents | Nothing needed to decide fit | No: after engagement, by the channel the firm's written security plan names |
Table 1. An inquiry form that decides fit and collects no tax return information. Design, not evidence.
Chapter 3 explains the last row, setting out the texts on tax return information, confidential client information and safeguarding customer data. We infer a design consequence, not a reading of what any text requires: the inquiry collects none of it. A free-text box is the field most likely to receive what the form was built not to collect. A label that names what not to type, such as identification numbers, account numbers and the contents of a notice, and says what happens instead, costs one line and spares the firm a pile of records.
A form must also be able to say no. It needs at least three outcomes: accept, which books a consultation; redirect, which offers a referral, a later date or a wait list; and route to staff, for the inquiry the rules cannot classify. The confirmation message says what happens next and by when, in a commitment the firm will keep, and says nothing about the person's tax position.
Response speed: what the lead research says
The general lead research on response speed cited here is Oldroyd, McElheran and Elkington's 2011 article in Harvard Business Review, in which they write that their research shows most companies are not responding nearly fast enough90. Its detailed figures below are secondary: they reach us only through a 2011 SmartCompany report that quotes the article, and none is a figure we read in the journal.
The authors audited 2,241 US companies, measuring how long each took to respond to a web-generated test lead91. In that audit, 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 all91. Among companies that responded within 30 days, the average response time was 42 hours91. A separate study cited in the same article involved 1.25 million sales leads received by 29 B2C and 13 B2B companies in the US91. In that study, firms that tried to contact potential customers within an hour were nearly seven times as likely to qualify the lead, defined as having a meaningful conversation with a key decision maker, as those that tried even an hour later91, and more than 60 times as likely as companies that waited 24 hours or longer91.
Four limits apply. The audit covered 2,241 US companies91 and was not a study of accounting firms, so it cannot say whether an accounting prospect behaves like a sales lead. A measured effect of response time on booking in any of the professions this series covers is NF; the nearest evidence is the general lead research above, and we do not say what a slow reply costs an accounting firm. The lead comparisons group outcomes by delay, so we read them as associations: a lead that gets a fast call may differ in other ways from one that does not. The outcome counted is a meaningful conversation with a decision maker, not a signed engagement.
A firm short of hours may still want a fast reply, because the reply is where it starts to choose. We think search cost supplies three reasons. First, in the terms of Stigler's point that identification reduces the cost of search87, a prompt message saying the firm is not taking new returns, or can see the prospect after a date, spares that person more searching. Second, the reply is the prospect's first observation of responsiveness, one of the five dimensions above. Third, the earlier the firm learns who fits, the more room it has to say yes to the best of them.
A stated preference points the same way, though it is not behavior. In a 2025 survey run by TaxDome† (practice management software) of more than 350 US business decision-makers, 85% said responsiveness is very or extremely influential when they select an accountant28. The respondents are business clients, not individual taxpayers. The speed that matters is that of a useful reply, which may be a polite no, not that of acceptance.
Busy season and accounting practice management
In this chapter busy season means the stretch each year when filing deadlines bunch a tax firm's work. The phrase accounting practice management is searched about 90 times a month in the United States3.
A longitudinal study of public accountants at one national firm, published in 2002, found that subjects reported an average workload of 49 hours a week before the busy season and approximately 63 hours during it18. That is 14 more hours a week (63 - 49), or 29% more (14 / 49). It is one firm and the data predate 2002, so it shows a pattern, not today's hours. A survey of over 700 auditors, published in 2019, found that they work on average five hours a week above the threshold at which they believe audit quality begins to deteriorate, and often 20 hours above it at the peak of busy season19. The threshold is the auditors' own belief, and the survey covers auditors, not tax-only staff. A 2025 survey by the accounting staffing firm Distinct† (secondary, as reported by CPA Practice Advisor) found that nearly 80% of respondents worked more than 51 hours a week during busy season and 31.4% worked more than 6192. We treat it as indicative only.
The IRS received 99,802,000 returns through April 3, 202620, 114,333,000 through April 1021 and 144,992,000 through May 822. Of the returns received by May 8, 68.8% were in by April 3 (99,802,000 / 144,992,000) and 78.9% by April 10 (114,333,000 / 144,992,000), shares of a count cumulative to May 8, not of a final season total. Of the 141,046,000 returns e-filed through May 8, tax professionals filed 75,316,000 and taxpayers who prepared their own filed 65,730,00022, so tax professionals filed 53.4% of e-filed returns (75,316,000 / 141,046,000). The IRS table splits only e-filed returns by preparer, so 53.4% is not a share of all returns. The IRS's 2021 taxpayer attitude survey found that around half of taxpayers used a paid tax professional31; the two sources agree on that order although they count different things and years.
The argument is about hours and runs two ways. First, the early weeks carry most of the volume: about one individual return in five arrives after April 10 (calc., above). We think a firm working 63-hour weeks has little room to take on more, and each hour spent with a person it cannot serve, or on a question the website could have answered, is an hour not spent on work already agreed. Second, the pattern is a reason to want work that falls outside the season. A monthly bookkeeping or advisory client pays every month, including the months when tax hours are free, and the fit questions on the intake form are how a firm tells which prospects those are. This paper does not measure a firm's inquiries by month; the IRS counts are of returns. The BLS annual average of 99,216 employees in tax preparation services in 20248 also understates the filing-season workforce, because the work is seasonal.
The hiring data bear on whether relief is coming. The Bureau of Labor Statistics counts 1,595,200 jobs for accountants and auditors in 2025 and projects 5 percent growth from 2025 to 203515. That is a count of jobs in an occupation, not of firms or the staff of any one kind of employer. On the supply side, the AICPA reports that graduates who earned a bachelor's or master's degree in accounting fell to 55,152 in the 2023-24 academic year, a drop of 6.6% from the year before1. Provisional federal data for 2023-24 show 40,817 accounting bachelor's degrees, down 3.3%, and accounting or taxation master's degrees down approximately 15% to 14,3351.
New CPA candidates tell a similar story, with a caution. In 2023, 42,626 new candidates entered the exam pipeline, the highest tally since 2016, a surge the AICPA says typically comes the year before a major exam change1. In 2024 the figure was 28,0821. The fall from 42,626 to 28,082 is 14,544 (42,626 - 28,082), or about 34% (14,544 / 42,626)1. Because 2023 was the high, we think that fall overstates any trend. Public accounting firms that responded to the AICPA survey reported hiring 11,985 new graduates in 2024, but the AICPA says low firm response meant the national total could not be projected with confidence1, so that number is not a national count.
Read together, these figures describe the supply of new accountants, not any firm's ability to hire. We think projected growth in jobs beside falling degree counts points to a tight market, which is a reason not to plan on hiring as the only relief. The pipeline figures are not a forecast: they cover 2023 and 2024. A firm that cannot easily add hours has two moves. It can protect the hours it has, which is what the page, the form, the reply and the phone line do, and it can decide what those hours earn, through the choice of client and the minimum fee that goes with it. The AICPA's 2025 MAP survey reported a median 6.7% increase in total net client fees over the prior year33; this paper does not test why.
An AI receptionist on an accounting firm's line
This section is a vendor's description of our own product. Our AI receptionist is a voice agent that answers the business phone when staff cannot. It picks up day and night and books appointments from real openings on the calendar. For an accounting firm we add two rules: it does not discuss a client's account, and it does not give tax advice. It is priced at $299 a month with 250 call minutes included, then $0.39 a minute34, added to one of the plans34. The AI receptionist page describes it further.
What the agent may do on an accounting line:
- Give hours, location, and the services the firm's website says it offers and does not offer.
- Say whether the firm's website says it is taking new clients, and offer the consultation openings on the calendar.
- Take a name, a number and the reason for the call, and ask the same fit questions as the intake form.
- Text a summary to the staff member who will return the call.
- Pass a call to a person at once when the firm has marked its type as urgent or the caller asks.
What stays with staff:
- Anything about a client's account, return, balance or status.
- Tax advice of any kind, including whether an item is deductible or whether to file.
- Fees beyond what the firm has published. The agent answers only from the firm's website and the facts the firm gave it, so a fee the firm has not published goes to staff.
- Whether the firm accepts an engagement.
- Documents and sensitive details, which go by the channel the firm names and not by phone.
The agent and the form can work as one filter: the agent can be set up to ask the form's questions, and its summary tells staff who called, what they need and what was booked. It does not replace the people who hold the client relationships.
Recording and disclosure texts for an AI voice agent
Table 2 sets out the legal texts on AI voices and call recording as the claims print them, with the limit each one states. The firm's counsel decides which applies, and a caller may be in any state.
| Text | What the claims print | Limit stated in the evidence |
|---|---|---|
| FCC Declaratory Ruling 24-17 | The TCPA's restrictions on an artificial or prerecorded voice encompass current AI technologies that generate human voices, so calls using them require the prior express consent of the called party to initiate, absent an emergency purpose or exemption93 | Concerns initiating calls. In a September 2024 proposed rule (FCC 24-84) the FCC said the TCPA's requirements do not extend to technologies used to answer inbound calls94; that is a proposal, not a holding. |
| 18 U.S.C. 2511(2)(d) | A private person may intercept a wire, oral or electronic communication where the person is a party or one of the parties has given prior consent95 | Does not apply if the interception is for the purpose of committing a criminal or tortious act. |
| RCW 9.73.030(1), Washington | Unlawful to intercept or record a private communication transmitted by telephone or other device without first obtaining the consent of all the participants96 | Compare the federal text: consent of one party there, of all participants here. |
| 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; fine not exceeding $2,500 per violation, and not exceeding $10,000 per violation for a person previously convicted under certain sections97 | The claims do not print the definition of a confidential communication. |
| California Business and Professions Code, sections 17940 and 17941 | A bot is an automated online account whose actions or posts are not the result of a person; it is unlawful to use one to interact with a person in California online with intent to mislead about its artificial identity, in order 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 bot, and the disclosure must be clear and conspicuous97 | The text speaks of online. Whether a telephone voice agent is covered is a legal question the source does not answer. |
| Utah Code 13-77-103(1) | A supplier using generative AI with an individual in a consumer transaction must disclose that it is generative AI and not a human if the individual asks or prompts, and the question must be a clear and unambiguous request98 | Triggered only when the individual asks. |
| Utah Code 13-77-103(2)(a) and (3), and 13-77-101 | An individual providing services in a regulated occupation must prominently disclose generative AI use in regulated services when the use is high-risk, verbally at the start of a verbal interaction; high-risk includes collecting sensitive personal information such as financial data and giving personalized financial advice98,99 | A regulated occupation is one regulated by the Department of Commerce that requires a license or state certification99. Whether accounting is one is a legal question the source does not answer. |
Table 2. Recording and AI disclosure texts as the evidence prints them, with the limit stated for each.
We infer four design consequences. First, the agent answers calls and does not place them. The FCC ruling concerns calls initiated with AI voices, and under 47 U.S.C. 227(b)(3) a person may bring a state-court action to recover actual monetary loss or $500 in damages for each violation, whichever is greater100. An agent that takes a message while a person returns the call stays on the answering side, and we do not say the ruling reaches inbound calls. Second, recording is a choice. A summary of a call is not an audio recording, and a firm that keeps audio or transcripts has made a further choice. Whether the recording texts in Table 2 apply to such a firm, and whether a transcript counts, is for counsel. A spoken notice at the start of any recorded call is the conservative design, given that the federal text and the Washington and California texts differ.
Third, the disclosure comes first: our agent tells every caller at the start that it is an AI assistant. Utah gives a safe harbor from enforcement for violating section 13-77-103 if the person's generative AI clearly and conspicuously discloses, at the outset of the interaction and throughout it, that it is generative AI, is not human, or is an AI assistant101, and the division director may impose an administrative fine of up to $2,500 for each violation102. We do not say that any of these texts covers a telephone agent, or that our disclosure satisfies any of them.
Fourth, two texts in our evidence reject the argument that an AI tool, and not the business, is responsible for what it said. In Moffatt v. Air Canada the tribunal called the suggestion that a chatbot was a separate legal entity responsible for its own actions a remarkable submission, and wrote that Air Canada is responsible for all the information on its website, whether from a static page or a chatbot103. That is an airline case, not an accounting one. Utah's text says it is not a defense to the violation of any statute administered and enforced by its Division of Consumer Protection that generative AI made the violative statement104. Both point to the same design: the agent says only what the firm has published.
The wage, the price and the clients that cover it
Plans start at $999 a month and the voice agent is $299 a month with 250 call minutes included, then $0.39 a minute34. The included minutes are about 62 four-minute calls (250 / 4), and the cases in this paper stay inside them. The agent added to a plan the firm already has therefore costs $299 a month, and the lowest plan plus the agent costs $1,298 a month ($999 + $299). Over a year these are $3,588 and $15,576 (12 x $299 and 12 x $1,298).
Two scopes keep the comparison matched. The agent answers the phone and the form, so it is set against hours and answered inquiries. The plan produces the demand the agent answers, so it is set against new clients. A ledger for the agent alone would leave the plan out of both sides. Against hours, the Bureau of Labor Statistics reports a median hourly wage for receptionists of $18.27 in May 2025105, so $299 a month is 16.4 hours of that wage ($299 / $18.27). A wage is not the full cost of an employee, and the comparison does not claim that the agent replaces a receptionist: it covers the calls staff cannot take. Chapter 5 sets the agent against hours at three values of an hour.
How often an accounting firm misses a call, how many calls it receives and how many inquiries become clients are NF; the nearest evidence for the first is two neighbors that bound it, each in its own units, and Chapter 1 estimated it from them. Among the health care clients of Invoca† (call tracking) in 2026, 46% of calls were not answered by a person (calc., the remainder after the 54% answered that Invoca† reports)42. In a 2024 secret-shopper study run by Clio† (legal practice software), 60% of 500 law firms did not answer a test call (calc., the remainder after the 40% who answered)40. One counts calls answered by a person and the other is a single test call to each firm, so we do not average them.
We estimate that an accounting firm without a full-time receptionist leaves a large minority of first calls unanswered by a person (from the two neighbors above; it errs high for a firm with a full-time receptionist, which will sit below it). A sole practitioner who is with a client when the phone rings will sit nearer the top. A third neighbor, 12.0% of calls abandoned before a staff member answered across Veterans Health Administration medical centers in October 2014106 and 8.3% by September 2016106, is a government call center and a different measure, and we use it only to show how far a staffed line can sit below the other two.
Chapter 1 does not carry lead conversion over as an estimate. The nearest measure is Invoca†'s, again from health care: 43% of calls answered by a person were leads and 45% of those leads converted on the call42. Whether a call is a lead is Invoca†'s own classification, across six health care fields. We use it only as the top of a range of shares in Chapter 5, where it likely errs high, because an engagement usually needs a proposal or a letter after the call. We do not estimate call volume either. The neighbors are 299 to 505 calls a month per office among more than 1,700 health care practices in Call Box†'s 2020 data107, which does not describe an office of fewer than five employees, as 71.3% of CPA offices are (44,800 / 62,828 establishments in the first quarter of 2024)10,11.
Because those links are estimates, Table 3 does not use them. It solves instead for the new clients a year that each purchase needs, by kind of client, from five published fees. Median revenue per advisory client was $17,867 a year in the 2024 CPA.com and AICPA benchmark of 206 self-selected practices35; the benchmark covers the client advisory line, and its median respondent firm revenue was $7,050,00035. A 2025 benchmark by Ignition† (billing software) of 219 of its customers found that monthly bookkeeping and accounting most commonly ran $250 to $499 a month36, or $3,000 to $5,988 a year (calc.), that business tax returns most commonly cost $1,000 to $1,49936, and that individual returns most commonly cost $400 to $59936. Those are the most common bands among Ignition†'s customers, solo bookkeepers included, not averages. The National Association of Tax Professionals reported an average of $236 for a base Form 1040 in its 2025 fee study37.
| Kind of client | Fee a year | New clients a year to cover the agent on a plan ($299 a month) | New clients a year to cover the lowest plan plus the agent ($1,298 a month) |
|---|---|---|---|
| Advisory client at a CPA firm that sells advisory services (AICPA benchmark of 206 self-selected practices) | $17,867 | 0.2 | 0.9 |
| Monthly bookkeeping client, most common band of Ignition† (solo bookkeepers included) | $3,000 to $5,988 | 0.6 to 1.2 | 2.6 to 5.2 |
| Business tax return, most common band of Ignition† | $1,000 to $1,499 | 2.4 to 3.6 | 10.4 to 15.6 |
| Individual return, most common band of Ignition† | $400 to $599 | 6.0 to 9.0 | 26.0 to 38.9 |
| Individual Form 1040, NATP average | $236 | 15.2 | 66.0 |
Table 3. New clients a year needed to cover each purchase, on fees alone. Each cell is the yearly cost ($3,588 or $15,576) divided by the fee (calc.).
At a CPA firm that sells advisory services, 0.9 new advisory clients a year cover the lowest plan plus the agent ($15,576 / $17,867), so one such client pays for more than a year of it. That figure comes from 206 practices that chose to answer a survey about a service they sell well35, so it is the high end for a small firm that does mostly tax returns, which should read the lower rows.
We infer that an advisory client stays about ten years, from the survey's median turnover of 10% a year35 and the rule that a constant yearly chance of leaving gives an expected stay of 1 divided by that chance. The same survey's median practice served 67 clients and reported 14 new ones, with no period printed for the 1435, so if the 14 is a year's count, 0.9 is about 6% of it (0.9 / 14). A firm that sells only base 1040s at the NATP average needs about 66, so the count depends on which clients the firm markets for.
The table counts fees, not what a firm keeps after the hours a client takes, so if a firm keeps half of each fee, every count doubles. Released hours repay nothing unless they are used on work or kept as rest, and the filter can turn away a client who would have been a good one. Chapter 5 turns the same costs into hours and rescued inquiries, at a range of values for the one link nobody has measured, and gives the margin that errors, which are not priced here, would have to consume. The AI-profit paper builds the full accounting ledger, and our pricing page lists the plans.
What this chapter cannot settle
Four things stay open. The busy-season sources measure hours for accountants at one national firm and for auditors, not for a small tax firm. Missed calls at accounting firms and lead conversion are NF; the nearest evidence is the law firm test and the health care call data named above. Call volume is not estimated, and how many calls an agent resolves for a front desk has no measure here either. Whether accounting is a regulated occupation under the Utah text, or whether a telephone voice agent falls under the California bot text, is for the firm's counsel. The firm's own records can settle the first three: a month of call summaries and a count of inquiries by source and by fit would give a small firm its own figures, and Chapter 6 shows how to count them and what small numbers cannot show.
Chapter 5
AI for accountants: what the evidence says and where it pays
AI for accountants, AI in accounting and AI for CPAs: what owners search for
Interest in AI for accounting is high, but the evidence on accounting work is thin. This chapter covers what that evidence shows, what the regulator and the profession have said, and which use is likeliest to pay in a small firm. The use we can price most firmly is the answering agent, set against hours and answered inquiries, with the plan that feeds it set against new clients.
In Google Ads keyword data for November 2026, "ai accounting", "ai for accountants" and "ai in accounting" carry identical US search volumes and identical costs per click3. That suggests one pool of searches counted three times, so the three should not be added together. The three unambiguous marketing variants, "accountant marketing", "accounting marketing" and "marketing for accountants", also carry identical figures and may be one pool3.
The phrase "cpa marketing" is ambiguous, because CPA also abbreviates cost per action in advertising. In our data "what is cpa marketing" and "cpa meaning marketing" average about 720 searches a month each, and "cpa marketing network" about 4803. Those read like searches about the advertising model, so an unknown share of the 1,300 monthly searches for "cpa marketing"3 is not about accountants. Chapter 1 explains the point.
One AI phrase, at about 6,600 searches a month, is searched about 25 times as often as one of the marketing variants, at about 260 (6,600 / 260). It is searched at least about five times as often as "cpa marketing" (6,600 / 1,300), and five is a floor because part of the 1,300 is not about accountants3. Table 1 lists the eight phrases.
| Search phrase | US searches a month | Cost per click | Citation |
|---|---|---|---|
| ai accounting | 6,600 | $38.95 | 3 |
| ai for accountants | 6,600 | $38.95 | 3 |
| ai in accounting | 6,600 | $38.95 | 3 |
| ai for cpas | 90 | $49.33 | 3 |
| cpa marketing | 1,300 | $12.31 | 3 |
| accountant marketing | 260 | $24.97 | 3 |
| accounting marketing | 260 | $24.97 | 3 |
| marketing for accountants | 260 | $24.97 | 3 |
Table 1. Searches for AI phrases about accounting and for marketing phrases, Google Ads keyword data, November 2026. The first three rows show the same volume and may be one pool counted three times, as may the last three. The volume for cpa marketing includes an unknown share of searches about the advertising model.
The phrases are ones a partner, an administrator or a software vendor would type, and a phrase such as "ai accounting" does not say who typed it. Volume shows interest, not purchase, and cost per click is what advertisers pay, not what a client is worth.
Vendor and publisher surveys of accounting professionals report heavy use and large expected gains. Each is a secondary source or one we could not read in full, so none carries a conclusion here. In an online survey that Intuit† (accounting software) commissioned in April 2025 of 700 US accounting professionals, 81% of respondents reported that AI has positively impacted their productivity108. That is a perception, not a measured change in output. Intuit†'s 2026 survey of 725 US accounting and bookkeeping professionals, published as sponsored content, found that 88% used AI for at least one client service in the last 12 months, that 30% said AI is embedded as the default in their day-to-day work, and that the largest group, 54%, used it situationally109. Using AI for at least one service is a low bar.
The 2025 State of AI in Accounting report from Karbon† (practice management software), a survey that includes input from more than 500 accounting professionals across six continents, found that 70% of respondents expressed concerns about data security110. Thomson Reuters† (publisher of AI products) reports two figures from its 2025 Future of Professionals study of 2,275 professionals worldwide. Respondents predicted that AI would save professionals an average of 5 hours a week over the next year111, and 53% said their organizations were already seeing a return as a direct or indirect result of AI adoption111. The first is a prediction, not a measured saving, the second is self-reported, and the study is not one of small accounting firms.
The Intuit† and Thomson Reuters† figures count professionals who answered a survey, not firms. The Census Bureau's BTOS counts businesses: its data for December 2025 to May 2026 show overall business AI usage hovering between 17% and 20%112, a figure for businesses overall, not for accounting firms. A Thomson Reuters† Institute survey of 1,702 respondents across legal, tax, accounting, audit, corporate risk, fraud and government sectors, collected in January and February 2025, is reported by LawSites113. The use figure LawSites reports from it is for legal organizations only: 26% actively using generative AI, up from 14% in 2024113. These instruments cannot be set on one line as a trend, so we state no share of accounting firms that use AI.
How well AI does accounting work: assessments, exams and confabulation
NIST's generative AI profile defines confabulation as the production of confidently stated but erroneous or false content, known colloquially as hallucinations or fabrications114. This chapter uses "hallucination" for that failure. For an accounting firm the question is how often it happens on the firm's own work. The direct evidence is two sets of tests built from accounting questions.
Wood and co-authors tested ChatGPT, as of January 2023, on 28,085 accounting assessment and test-bank questions from 14 countries and 186 institutions. They report that students "significantly outperform ChatGPT" on those assessments5. Eulerich, Sanatizadeh, Vakilzadeh and Wood report that ChatGPT 3.5 could not pass any of the CPA, CMA, CIA or EA exams in their test6. After moving to ChatGPT 4, adding few-shot training and allowing reasoning and acting, such as using a calculator, they report that it passed all the exams tested6. Table 2 gives the scores.
| Study | What was scored | Result | Citation |
|---|---|---|---|
| Wood and co-authors, ChatGPT as of January 2023 | 28,085 accounting assessment and test-bank questions | Students 76.7% on assessments; ChatGPT 47.5% without partial credit, 56.5% with | 5 |
| Eulerich and co-authors, ChatGPT 3.5 | CPA, CMA, CIA and EA exams | Average 53.1% across all assessments; passed none | 6 |
| Eulerich and co-authors, after the improvements named in the text | The exams tested | Average 85.1%; passed all | 6 |
Table 2. Two studies of AI on accounting questions. The rows count different things and are not a trend.
Within the Eulerich study the score depended on the model and on how it was set up: the same authors moved from failing every exam to passing every exam by changing the model, adding few-shot training and allowing reasoning and acting. A score in a paper is a property of a configuration, not of AI in general, and a firm's tool may not match it. The two studies also count different things, a comparison with students across many questions and a pass or fail on named professional exams, so Table 2 is not a rise from 47.5% to 85.1%.
We think an exam item hands the model a self-contained question with a known answer, while an engagement hands an accountant incomplete records, a client who misremembers and a rule that changed since last year. None of the tests above uses a firm's own client work, so the exam results say what a model can do on questions and little about what it does in a file.
A tool built to cite sources is not thereby correct. A 2023 audit of commercial generative search engines found that on average only 51.5% of generated sentences were fully supported by citations and only 74.5% of citations supported their associated sentence61. In law, a different profession, Magesh and co-authors found in 2024 that the AI legal research tools of two providers each hallucinate between 17% and 33% of the time115. The closest test of models on tax questions that we found used IRS-endorsed Volunteer Income Tax Assistance certification questions. In a 2025 paper with 390 questions including perturbed versions, top models such as GPT-4 and Claude 3.5 Haiku reached 83.33% accuracy on basic multiple-choice questions, against 54.17% for GPT-3.554. Those are certification questions, not client returns, so neither this result nor the legal one is an error rate for a research tool inside a firm. The practical consequence is the one the IRS guidance below draws: a person verifies what the tool produced.
What studies of AI at work show, and what they leave out
None of the studies of AI at work in our files is a study of accountants. Noy and Zhang found that for mid-level professional writing tasks, access to ChatGPT reduced completion time by 37% (10 minutes) relative to the control group116, and that it compressed the productivity distribution by benefiting low-ability workers more116. Brynjolfsson, Li and Raymond found that generative AI assistance raised customer support productivity, measured as issues resolved per hour, by 14% on average, including a 34% improvement for novice and low-skilled workers and minimal impact on experienced and highly skilled workers117.
Dell'Acqua and co-authors separated tasks by whether they lay inside or outside the technological frontier. For tasks inside it, consultants using AI completed tasks 25.1% more quickly and produced over 40% higher quality results118. For a task selected to be outside the frontier, consultants using AI were 19 percentage points less likely to produce correct solutions118. Consultants below the average performance threshold improved by 43% and those above by 17%, compared with their own scores118.
A different kind of study asks whether pay and hours changed. In the revised NBER version of a study by Humlum and Vestergaard, the only version we cite, the authors estimate precise null effects on earnings and recorded hours at both the worker and workplace levels. They rule out effects larger than 2% two years after the launch of ChatGPT119. A task can get faster without hours or earnings moving, so this result does not contradict the task experiments. Neither kind of study says what a tool does inside a small accounting firm.
We think three cautions follow. First, in three of the studies the largest gains went to less experienced or lower-scoring workers. In a firm those are the staff least able to spot a wrong answer, so review has to be designed in. Second, the outside-the-frontier result means a tool can lower correctness on a task that looks like one where it helps, and a firm cannot know in advance where the line falls. Third, a faster task does not by itself change what a firm earns. Our paper on AI profit weighs the wider evidence on AI returns.
AI for CPAs and the IRS Office of Professional Responsibility
On June 24, 2026 the IRS Office of Professional Responsibility issued an alert titled Introductory Guidelines for Responsible AI Use in Federal Tax Practice. It says that technology is a powerful tool, not a substitute for professional judgment, and that practitioners must review AI output, validate its facts and citations, and handle sensitive client data safely and securely120. In its words, "final decisions must always rest with qualified professionals"120. We think the alert is guidance applying existing duties to AI, not a new rule.
The alert says that due diligence requires verifying "the accuracy of facts, citations, and calculations produced by AI", and that practitioners "cannot rely solely on AI"120. On billing, it says that billing clients for manual labor or time that was not actually spent, or double billing for AI-assisted tasks, may violate section 10.27, depending on the facts120. On firm procedures, it names protocols for secure data handling and AI accuracy monitoring, and says that outsourced or third-party AI tools should be vetted120.
Two Circular 230 sections sit beside the alert. Section 10.22(b) presumes a practitioner exercised due diligence when relying on another person's work product if the practitioner used reasonable care in engaging, supervising, training and evaluating that person121. Section 10.35 requires a practitioner to possess the necessary competence to practice before the IRS, meaning the appropriate knowledge, skill, thoroughness and preparation for the matter122.
We think a firm that wants to be able to show its work needs a short written procedure: which tasks may use a tool, who reviews the output, what the reviewer checks against a source, and where the result is logged. How either Circular 230 text bears on a software tool, and whether a particular procedure meets it, is for the firm's counsel.
Whether clients must be told is a separate question. A Journal of Accountancy column dated April 1, 2025 states that "no specific federal law or professional standard applicable to CPAs mandates disclosure when generative AI is used or if client data is used within it"123. It is one column, a secondary source and commentary, not a rule, and it predates the alert, so it says nothing about the alert.
On confidentiality, an August 5, 2026 Journal of Accountancy article by the communications manager of the AICPA Professional Ethics Division says the Code of Professional Conduct still provides clear guidance despite AI. Members' core responsibilities, confidentiality among them, do not disappear because new technology is involved, it says67. It adds that, before pasting client data into any AI system, a member in public practice should ask whether it is confidential client information, whether consent is required and obtained, and whether the use could result in disclosure to third parties67. It is a staff article. An AICPA Code interpretation on generative AI is NF; the nearest evidence is that article, so disclosure to clients, and consent where client data enters a tool, are for the firm's counsel.
AI and client information: Section 7216 and the confidentiality rule
Chapter 3 sets out the texts on client information; this section adds only what they mean for the choice of an AI tool. The AICPA Code's Confidential Client Information Rule says a member in public practice shall not disclose any confidential client information without the specific consent of the client64. Its interpretation on third-party service providers says that before disclosing such information to a provider, the member should either enter into a contractual confidentiality agreement with the provider and obtain reasonable assurance that it has procedures to prevent unauthorized release, or obtain the client's specific consent64.
Tax return information has its own statute and regulations. Under 26 U.S.C. 7216, a tax return preparer who knowingly or recklessly discloses or uses information furnished for a return, other than to prepare the return, is guilty of a misdemeanor68. Under section 6713 a preparer who discloses or uses return information other than to prepare the return pays a civil penalty of $250 per disclosure or use, capped at $10,000 for any calendar year69. Treasury Regulation 301.7216-3(a)(1) says that unless section 7216 or 301.7216-2 authorizes it, a preparer may not disclose or use a taxpayer's tax return information before obtaining the taxpayer's written consent4.
Two more provisions bear on choosing software. The definition of tax return preparer in 301.7216-1(b)(2) includes any person engaged in the business of providing auxiliary services in connection with the preparation of tax returns, including a person who develops software that is used to prepare or file a tax return70. Section 301.7216-2 allows a preparer to share tax return information inside the firm within the United States without consent, but requires the taxpayer's consent before any disclosure to a person located outside the United States71.
The FTC says its Safeguards Rule lists tax preparation firms among the financial institutions it covers72. IRS Publication 5708 says that tax and accounting professionals are considered financial institutions regardless of size and that the Rule requires a written, accessible information security plan73. As a design matter, a written plan is a natural place to keep the list of tools that touch client data. None of this is advice on what a given firm must do. Whether these texts reach a given AI tool, including one that reads a return, is for counsel.
Where AI pays in a small accounting firm: capacity and the cost of review
Our paper on AI profit adds up four possible gains from an answering agent: capture, conversion, capacity and retention. Its system ledger adds demand above them: the contacts a month the plan adds, times the share that become clients, times what a client is worth. It subtracts the run cost and the fixed cost, and that sum is its profit identity. For a firm short of hours, capacity is the first agent term to look at: the hours a tool releases in a month, times what an hour is worth in its next best use. A released hour is worth something only if the firm fills it.
The supply of hours is part of why capacity comes first. The AICPA reports that graduates who earned a bachelor's or master's degree in accounting fell to 55,152 in the 2023-24 academic year, down 6.6% from the year before1; Chapter 1 sets out the pipeline. We think a firm that cannot easily add staff values a released hour at more than a wage. The wage itself is known: the Bureau of Labor Statistics reports a 2025 median pay of $83,680 per year, or $40.23 per hour, for accountants and auditors15. An occupation's median wage is neither what a firm bills for an hour nor what an hour is worth when the alternative is turning work away. The billing rate is nearer: the AICPA and CIMA's 2023 MAP survey of 1,117 firms reported a median hourly billing rate of $159 in fiscal year 2022, across all positions32.
The two purchases have two scopes. The agent answers the phone and the form, so it is set against hours and answered inquiries. The plan produces the demand the agent answers, so it is set against new clients. A ledger for the agent alone would leave the plan out of both sides. No measured saving of hours in an accounting firm is cited here; the nearest figure is the prediction of 5 hours a week from Thomson Reuters†111. The question is therefore how many hours would have to be released.
The agent is $299 a month with 250 call minutes included, about 62 four-minute calls (250 / 4), and $0.39 a minute after that34. The cases in this paper stay inside the included minutes, so the agent added to a plan the firm already has costs $299 a month. Table 3 values an hour three ways: at the Bureau of Labor Statistics median wage for receptionists, the person who would otherwise answer105; at the median pay of accountants and auditors15; and at the AICPA median billing rate32.
| Value of one hour | What it is | Hours a month to cover $299 |
|---|---|---|
| $18.27 | BLS median wage, receptionists | 16.4 |
| $40.23 | BLS median pay, accountants and auditors | 7.4 |
| $159 | AICPA median billing rate, all positions | 1.9 |
Table 3. Net hours released a month needed to cover the agent at $299 a month, its published price34, by the value of one hour. Each cell is $299 divided by the hourly value (calc.).
The wage rows ask what the agent costs against the pay it can stand in for. The billing-rate row asks what the same hours earn if the firm fills them at its median rate, which holds only when demand is there to fill them. In busy season it is; outside it, that depends on the plan. The agent's $3,588 a year (12 x $299) is 1.9% of the $189,695 in net client fees that the AICPA's median firm earned per full-time professional in fiscal year 202232 ($3,588 / $189,695).
Net hours are smaller than gross hours. Suppose a share of every hour a tool saves goes back into reading, checking and correcting its output, because the IRS guidance says practitioners cannot rely solely on AI. Table 4 shows how many gross hours must be saved to reach a given number of net hours across a range of review shares, since nobody has measured the share. At a 50% share, the 7.4 net hours that cover the agent at the accountant wage need 14.9 gross hours, and the 1.9 hours at the billing rate need 3.8 (calc., Table 4).
| Net hours wanted a month | Review share 25% | Review share 50% | Review share 75% |
|---|---|---|---|
| 4 | 5.3 gross hours | 8.0 gross hours | 16.0 gross hours |
| 8 | 10.7 gross hours | 16.0 gross hours | 32.0 gross hours |
| 16 | 21.3 gross hours | 32.0 gross hours | 64.0 gross hours |
Table 4. Gross hours saved a month needed to reach a net figure, on a range of shares of saved hours returned to review. Each cell is the net hours divided by one minus the review share.
The agent is also set against answered inquiries. A call or form that would have gone unanswered, and becomes a client, is worth that client's fee. Chapter 1 estimates how many first calls reach no one, and both it and Chapter 4 say the estimate errs high for a firm with a full-time receptionist. It is a share and cannot supply a count of inquiries, so Table 5 solves for the count. The agent on a plan costs $3,588 a year (calc., above). Nobody has measured the share of rescued inquiries that become clients at accounting firms, so the table shows three values for it. The top value is close to the 45% of leads that Invoca† (call tracking) reports converting on the call in health care42, a figure that likely errs high for an engagement that needs a proposal first.
| Kind of client | Fee a year | Clients a year to cover $3,588 | Inquiries a year if 50% become clients | If 25% do | If 10% do |
|---|---|---|---|---|---|
| Advisory client at a CPA firm that sells advisory services (AICPA benchmark of 206 self-selected practices) | $17,867 | 0.2 | 0.4 | 0.8 | 2.0 |
| Monthly bookkeeping client (low end of the most common band of Ignition†, billing software) | $3,000 | 1.2 | 2.4 | 4.8 | 12.0 |
| Business tax return (low end of the most common Ignition† band) | $1,000 | 3.6 | 7.2 | 14.4 | 35.9 |
| Individual return (low end of the most common Ignition† band) | $400 | 9.0 | 17.9 | 35.9 | 89.7 |
Table 5. Inquiries a year the agent must rescue to cover $3,588 (calc., above), by kind of client and by the share of rescued inquiries that become clients. Each cell is the clients needed divided by that share (calc.). The shares are a range to test, and the top value is Invoca†'s health care figure above, used as a ceiling, not as an estimate. Fees are those of Chapter 4's Table 3.
Beside a base rate, the median advisory practice serves 67 clients and reports 14 new ones, with no period printed for the 1435. Rescuing 0.4 to 2.0 inquiries a year covers the agent for a firm that sells advisory services at that fee, which is small beside 14 new clients, each of which took at least one inquiry. That fee is the high end for a small firm that does mostly tax returns, because it comes from practices that chose to answer a survey about a service they sell well. A firm that sells base individual returns at the low end of the band needs 18 to 90 (calc., Table 5). The fee a firm charges therefore decides whether an agent pays, which is the case for choosing clients and raising the minimum fee.
The plan is set against new clients, not hours. The lowest plan plus the agent costs $1,298 a month, or $15,576 a year, which is 8.2% of the $189,695 in median net client fees per professional32 ($15,576 / $189,695). Chapter 4's Table 3 gives the clients that cover it: 0.9 new advisory clients a year at a firm that sells advisory services, 2.6 to 5.2 monthly bookkeeping clients, 10.4 to 15.6 business tax returns, or about 66 base Form 1040s at the NATP average. One new advisory client brings $17,867 a year35 against the $15,576, which leaves $2,291 a year ($17,867 - $15,576) before the cost of serving that client. The test of the plan is therefore a count of new clients by kind, which Chapter 6 shows how to keep. A firm short of hours uses the demand it brings to take better-paying clients, raise its minimum fee and fill the months outside busy season.
Chapter 1 states the one corner in which the lowest plan and the agent lose money and gives the base rates that put the median firm away from it. The tables here do not move that corner. They show what a firm outside it needs: hours released, inquiries rescued and new clients won, each set against a cost.
We draw three readings from the tables. First, the break-even moves by about a factor of nine on a number the firm knows and we do not, the value of its next best hour: 16.4 hours at the receptionist wage against 1.9 at the billing rate. Second, review can multiply the gross requirement, to four times the net figure at a 75% share in Table 4.
Third, the tables do not price errors. An error that must be corrected, or one that reaches a client, consumes hours of its own, and this paper does not measure how often that happens in accounting work. The room for errors is the gain left after the cost: $2,291 a year for one advisory client (calc., above), which a client lost to an error would erase. Our paper on AI profit works a full accounting ledger with every input marked; these tables are smaller on purpose.
How a firm bills changes what a released hour is worth. Under hourly billing, an hour not worked is an hour not billed, and the alert says that billing for time not actually spent may violate section 10.27 depending on the facts120. A released hour therefore has value only if the firm fills it with other work, and AI moves the constraint from hours to demand, where the plan's new clients and the firm's choice among them decide what the hours earn. Under a fixed fee the saved hour is margin from the first day. That is argument, and the choice of fee model belongs to the firm. Chapter 6 describes how to count hours and fee per engagement.
The objection that thin evidence argues for waiting
The strongest objection to the chapter so far is that the evidence is too thin to act on, so a firm should wait. We think waiting is not free in a profession where hours are scarce, and a narrow, reviewed trial costs little to run. That holds only if the trial is counted: a firm that adopts a tool and does not record the hours it saved, the hours it spent on review and the errors it caught cannot learn which side of the break-even it landed on.
The studies of other occupations make a gain plausible, and the accounting tests show that models can score well on questions in some configurations. But the surveys above report use and predicted savings, and none is a measured net result in a small accounting firm. A randomized or field study of an AI agent answering inbound calls or scheduling in accounting settings is NF; the nearest evidence is the outbound-call study of one hospital cath lab, in which an AI voice assistant completed 86.4% of its pre-procedure calls in the first phase124. That shows what a voice agent can finish, not how an accounting front desk performs. A firm's own count is the only measurement of its case that will exist.
We sell an AI voice agent and have an interest in the answer. We do not recommend a tool for work inside the firm, such as drafting, summarizing or reading documents. A firm choosing one should ask the vendor where client information goes, who can see it, what the tool does when it is wrong, and what the firm can log. Our page on AI for accountants is a practical guide to the same subject.
What stays with an accountant
According to the texts above, several decisions stay with a person. Each item names its text. None is an instruction.
- The final decision on any figure, position or advice that leaves the firm120.
- Verifying the accuracy of facts, citations and calculations an AI tool produced120.
- Whether client information may enter a tool at all, under the confidentiality rule and its interpretation on third-party providers64 and, for tax return information, the regulation's written-consent requirement4.
- Vetting any outside or third-party AI tool120, and the care in engaging, supervising, training and evaluating another person whose work a practitioner relies on, under section 10.22(b)121.
- What is billed, set against time actually spent120.
- The competence that section 10.35 describes122.
- Whether and how clients are told; the sources above are a column and an ethics staff article, not a rule67,123.
Chapter 6
Measuring accounting firm marketing, and a first 90 days
Why accounting firm marketing is hard to measure, even for large advertisers
A firm wants to know which marketing brought which clients. Research on large advertisers shows why that question is harder to answer than a report suggests.
Lewis and Rao report on twenty-five large field experiments with major US retailers and brokerages, most reaching millions of customers, which show that measuring the returns to advertising is difficult125. The median confidence interval on return on investment across their experiments is over 100 percentage points wide125. 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-weeks, "making experiments costly and potentially infeasible for many firms"125. They also conclude that selection bias, due to the targeted nature of advertising, is a "crippling concern" for widely employed observational methods125.
Advertising is often aimed at people who are already likely to buy, so a comparison of people who saw it with people who did not mixes the effect of the advertisement with the effect of the aim. Gordon, Zettelmeyer, Bhargava and Chapsky used data from 15 US advertising experiments at Facebook, comprising 500 million user-experiment observations and 1.6 billion ad impressions, to compare experimental results with observational ones126. 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 variables126.
One stark case is paid search at eBay. In large-scale field experiments there, Blake, Nosko and Tadelis found that returns from paid search are a fraction of non-experimental estimates and that brand-keyword ads have no measurable short-term benefits127. Typical OLS methods gave a return on investment of over 4,100% without time and geographic controls and over 1,400% with them127. The experimental estimate was -63%, with a 95% confidence interval of -124% to -3%127. Berman, in a game-theoretic model of multi-publisher advertising, shows that the popular last-touch attribution method over-incentivizes ad exposures, often resulting in lower advertiser profits128. That is an analytical result under the paper's assumptions. An estimate of how far last-click or platform-reported attribution misstates results, as such, is NF; the nearest evidence, the comparison of observational and experimental estimates above, does not support a statement that last-click reports are wrong by a given amount.
Every empirical result above concerns a large advertiser in a digital channel (retailers and brokerages, Facebook, eBay), and none is a study of a professional practice. We think what transfers is the logic: a person who types the firm's name into a search box was already looking for the firm, so a report that credits the click that came last may credit the wrong thing. The size of the error may not transfer, and we do not know it for accounting firms. We expect experiments of that scale to be out of reach for a small firm, and the authors call experiments potentially infeasible for many firms125.
For a small firm the lesson is a posture: treat each source figure as a claim to be checked, count outcomes that do not depend on attribution, and keep the record long enough to see a pattern. The measurement chapter of our companion paper treats attribution for a small practice at greater length.
Accounting lead generation: what a firm that sells hours can count
A firm that sells hours has five counts worth keeping, and Table 1 defines them.
| Count | What it is | What it cannot tell you |
|---|---|---|
| Inquiries by source | Each call, form or email from a person who is not yet a client, tagged with where it came from as far as is known | Why the person chose the firm; sources overlap |
| Inquiries that fit | Inquiries the firm would accept, judged against written criteria set before the inquiry arrives | Whether the criteria are right |
| Engagements signed | Inquiries that became paid work | Which marketing step earned the signature |
| Hours and fee per engagement | The hours the firm spends on an engagement set against the fee, giving an effective rate | The value of the work turned down; needs timekeeping even for fixed-fee clients |
| Clients kept | Clients active in one year who are still active the next, as a share of those active in the first | Why a client left |
Table 1. The five counts. None needs special software; all need the same record kept the same way.
The rows follow the funnel, but the second carries the most weight in a firm short of hours. An inquiry that fits is one the firm would accept at the fee on terms it can staff. The criteria might cover the service, the kind of entity, the volume of records, the timing against busy season and the client's willingness to supply them. They should be written before the call, because a criterion invented after an inquiry arrives is a rationalization. Chapter 4 describes how the website and the intake form apply them. The count of inquiries says little about the firm's choice until it is split into inquiries that fit and inquiries that do not.
Many firms also want a figure for what a typical engagement is worth. Table 2 gives the published figures for what engagements bring in and how long clients stay, each with its owner and its limit, so that a firm's own records can be read against them. The rows count different things, so none is a typical firm.
| Figure | Source | Limit |
|---|---|---|
| Median revenue per advisory client, $17,867 a year; typical monthly fee for one, $3,000; client turnover, 10% a year35 | CPA.com and AICPA benchmark, 2024 | 206 self-selected practices, with a median firm revenue of $7,050,000; the advisory line only, not tax clients, so the high end for a small firm that does mostly tax returns35 |
| Median practice: 67 clients served, 14 new35 | Same survey | No period is printed for the 14 |
| Monthly bookkeeping and accounting most commonly $250 to $499 a month; business tax return $1,000 to $1,499; individual return $400 to $59936 | Ignition† (billing software), 2025, 219 customers | Most common bands and not averages; solo bookkeepers are in the sample; a price list and not what clients paid |
| Base Form 1040, $236 average; returning clients, 87%37,38 | National Association of Tax Professionals, 2025 fee study | Preparers of every credential; the 87% is read from a chart and does not say whether it counts clients or revenue |
| Form 1120-S, $721 average129 | Canopy† (practice management software), 2020 survey of hundreds of tax professionals | Six years old |
| 56% of firms' clients with the firm eight years or more; 34% a decade or more29 | Accounting Today and ADP survey, December 2019 | Self-reported by firms; no sample size on the page |
| Median hourly billing rate, $159; net client fees per full-time professional, $189,69532 | AICPA MAP survey, fiscal year 2022 | Self-selected participating firms, most of them small; the rate is across all positions, not by title |
Table 2. Published figures for fees, billing rates and client tenure at accounting firms, with each source's limit.
The survey's typical monthly fee of $3,000 would be $36,000 a year (12 x $3,000), twice the $17,867 median revenue per client35. The two medians answer different questions, a typical fee and an average revenue per client, and we do not reconcile them; the worked numbers here use the lower. The client tenure in the table is a trade panel's report and the 10% turnover is a survey median for one service line35, so we do not line them up as one measure.
Row four of Table 1 turns a client list into a view of capacity. A fee divided by the hours an engagement takes is its effective rate, and the median billing rate of $159 is the reference32. An engagement earns that rate if its hours do not exceed its fee divided by 159. At the published fees that is 112 hours for the median advisory client at $17,867 a year, 19 to 38 hours for a bookkeeping client at $3,000 to $5,988 a year, and 1.5 hours for a base Form 1040 at $236 (calc., Table 2). An engagement that takes more hours earns less than the median rate.
In a firm with idle hours, such a client may still be worth taking. In a firm short of hours, the extra hours come from other clients or from the partner's evenings, and the firm does better to raise its minimum fee or to choose the client who pays more for the same hours. More inquiries let a firm choose, and the counts of inquiries that fit and of effective rate show whether it chose well.
A firm's own records supply comparisons from the first month. Invoices give the fee per engagement and the number signed. The client list at the start of last year, set against the list a year later, gives the clients kept and lost, and so a turnover rate to read against the 10% median35. The same lists give the clients added, to read against the 14 new clients of the median advisory practice35. One month of timekeeping gives the first effective rates to set beside the $159 median32.
The phrases an owner might type for help with this work draw few searches. In Google Ads keyword data for November 2026, "accounting lead generation" averages about 70 US searches a month3, "how to get accounting clients" about 503 and "accounting practice management" about 903. Volume shows interest, not purchase, so these figures show only that the topic draws little search. Chapter 1 gives the full table of what owners search for.
What small numbers can and cannot tell you
A small count gives a rate with wide error, and most CPA offices are small. Of the 62,828 CPA establishments counted in the first quarter of 2024, 44,800 had fewer than 5 employees10,11, which is 71.3% (44,800 / 62,828); the count is of offices, not firms. The median advisory practice serves 67 clients and loses 10% of them a year35, which is 6.7 clients (10% x 67), and a quarter's records would be smaller still.
Table 3 gives a 95% range around that record and four others, computed with the Wilson score method, a standard way to put a range around a share. The shares are the survey's 10% turnover35 and the 87% returning share of the National Association of Tax Professionals38; the client counts are the survey's median of 6735 and sizes chosen for the example. The range describes sampling noise only and assumes the records resemble a random sample of what could have happened. Records that follow the tax calendar are unlikely to, so we think the true uncertainty is larger than the table shows.
| Record | Share | 95% range |
|---|---|---|
| 7 of 67 advisory clients lost in a year (10% turnover at the median client count) | 10% | 5% to 20% |
| 70 of 670 clients lost (the same share, ten times the clients) | 10% | 8% to 13% |
| 4 of 67 clients lost | 6% | 2% to 14% |
| 87 of 100 clients returning (the NATP share, for a firm of 100 clients) | 87% | 79% to 92% |
| 870 of 1,000 clients returning | 87% | 85% to 89% |
Table 3. How wide the range is around a share, for records built on the survey medians and on sizes chosen for the example. Each range is a Wilson score interval at 95% confidence computed by us from the counts shown (calc.).
Read the table in three ways. First, ten times the records (70 of 670 against 7 of 67) give the same 10% share with a range about a third as wide (Table 3: 8% to 13% against 5% to 20%). Second, a firm that goes from 7 leavers to 4 among the same 67 clients, from 10% to 6%, cannot say whether the change was real, because 6% lies inside the first range in Table 3 and the two ranges overlap. Third, one year of records from 100 clients leaves a range from 79% to 92% around 87% returning (calc., Table 3), which cannot separate a true 80% from a true 90%.
The practical rule is ours and conservative on purpose. Decide in advance what a change has to look like before you act on it, and do not treat a change in a rate as real unless the new range does not overlap the old one. Compare the same months in different years, because we expect inquiries for a firm that does tax work to follow the filing-season calendar, as the IRS return counts in Chapter 4 do; this paper does not measure inquiries by month. A firm with few inquiries will often learn nothing from a quarter of records about whether one step worked. The response is to count for longer and to read the process facts (what was built, when) beside the outcome counts.
Client value and retention in a business built on recurring work
A firm whose engagements recur, such as annual returns, monthly books and payroll, has a client list that works like a stream of future fees. Gupta, Lehmann and Stuart define the value of a customer as the expected sum of discounted future earnings130. For a firm, that sum is the fees of the years a client stays, less the cost of serving them, discounted, so the retention rate sets how many years the sum runs.
Gupta, Lehmann and Stuart, valuing five firms from public data, found that a 1% improvement in retention, margin or acquisition cost improves firm value by 5%, 1% and 0.1% respectively130. 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 capital130. The result comes from a model fitted to five firms, not from an experiment, and we do not know that it holds for a firm of accountants. In the Idea in Brief of their 1990 Harvard Business Review article, as republished by Bain & Company, Reichheld and Sasser state that companies can boost profits by almost 100% by retaining just 5% more of their customers131. We cite that statement only to show the idea is old.
Table 4 shows why retention matters, on the survey's own figures. Suppose a client leaves with the same chance each year. In that simple geometric model, the expected number of years a client stays is 1 divided by the yearly chance of leaving. The 2024 benchmark's median turnover of 10% a year35 therefore gives a stay of about 10 years, and its median revenue of $17,867 per advisory client a year35 gives about $178,670 over that time (10 x $17,867). We infer this from two medians, and it describes a CPA firm that sells advisory services, the high end for a small firm that does mostly tax returns. It is revenue, not margin, and it ignores the discounting that the definition above includes, which would lower every value. The table varies the retention rate around the survey's 90%, the complement of the 10% turnover35.
| Yearly retention | Expected years a client stays | Revenue over the stay at $17,867 a year |
|---|---|---|
| 80% | 5.0 | $89,335 |
| 85% | 6.7 | $119,113 |
| 90% (the survey's median) | 10.0 | $178,670 |
| 95% | 20.0 | $357,340 |
Table 4. Expected tenure and undiscounted revenue of an advisory client under a constant yearly retention rate. Each tenure is 1 divided by one minus the retention rate (calc.), and each revenue figure is that tenure times the median $17,867 a year35. Only the 90% row is the survey's figure35; the other rows are a range around it.
In the model of Table 4, moving from 90% to 95% doubles the expected tenure. Table 3 showed that a year of records from 67 clients leaves a range of 5% to 20% around a 10% turnover, and the lower end is the 5% turnover that a 95% retention rate implies. The quantity that matters most is the one small numbers measure worst. A trade survey points to long stays without matching the model: Accounting Today and ADP reported in December 2019 that 56% of firms' clients had been with their firm eight years or more and 34% a decade or more29. That is firms reporting on their clients, so we do not line it up with the turnover figure. Two responses follow: read why each client who leaves did so, because with few leavers each one is informative, and measure retention by kind of engagement over several years, not once.
Keeping a client whose engagement takes more hours than the fee covers adds no value, and the effective rate in Table 1 is the check. Models of repeat purchase exist: Fader and co-authors calibrated one on repeat transaction data for 2,357 sampled CDNOW customers over the first half of a 78-week period132. We think a firm whose client list is far shorter has too little data to fit such a model, so we count and do not model.
The monthly AI answer check as a measurement
Chapter 2 described, as a design, a monthly check of what AI assistants say about the firm. Treated as a measurement, the check asks the questions a prospective client might ask and records which firms are named and which sites are cited.
Our study fixed the form of its questions, each of the form "Who is a good dentist in Atlanta, GA? Name specific businesses.", with the city and the kind of business changed2. A firm can use the same form with its own services and cities, and keep the wording the same every month so that only the answer changes. For accountants, asked about in 16 cities, the assistants in our study named 272 businesses, 17 per question, none of them named by all three, and 90.4% were named by only one2. With so little agreement, "is my firm named?" has three separate answers, one for each assistant, and they should be recorded separately.
That study's limits decide how the check should be read. Each question 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 sees2. Because each question was asked once, the study does not measure how much an answer changes from one run to the next, and a firm that compares two monthly checks is comparing two single runs. The first month's job is therefore to measure the spread: ask each question several times on the same day, record how often your firm appears, and treat that variation as the noise floor. The logic of Table 3 applies, so report a share and its range, not a single yes or no. Record the spellings of the firm's name that count as a match, and which interface was used.
Even a clear change would not show cause. The authors who introduced generative engine optimization describe generative engines as black boxes that leave content creators with little to no control over when and how their content is displayed60. A firm that changes a page and then sees its name in an answer has an association, not an experiment. Keep a dated log of site changes beside the log of answers, and read the pair with the caution of the first section. Our study of local AI answers describes the method in full and holds the data.
How to get accounting clients: a first 90 days
This plan is for you, the partner or administrator. It puts the count first, because a change you cannot measure teaches little, and each step names what to count. None of it is tax or legal advice; where a rule is involved, the step points to Chapter 3.
Days 1 to 14: build the count and take the baseline
Make a sheet with the columns of Table 1: date, source, service wanted, kind of client, whether the inquiry fit, outcome, hours and fee. Fill it back for the last 12 months from email, call logs and any practice management records, and expect gaps. Write your fit criteria down and say who judges. Ask staff to log time per engagement for the 90 days, even for fixed-fee clients, so that effective rates exist. Then run the AI check three times for each question and each assistant, and keep the raw answers. Buy nothing in these two weeks.
Days 8 to 30: profile and listings
Google's own help page says that local results are mainly based on relevance, distance and popularity43, that there is no way to request or pay for a better local ranking on Google43, and that more reviews and positive ratings can help a business's local ranking43. The work here is therefore accuracy and completeness. Claim and complete the profile, list each service, and make the firm's facts the same wherever a prospect, or an assistant that reads these pages, might find them. State credentials exactly as they are held. Count the listings corrected and the date each was done.
Days 15 to 45: pages for each service and each kind of client
Write one page for each service and one for each kind of client you want. Each says who the service fits, what the first conversation covers, what you need from the client and, where you can say it, who it does not fit. This is the fit filter made public: a prospect who reads it can screen themselves out before calling. The AICPA interpretation 1.600.010 says promotional efforts are false, misleading or deceptive if they create false or unjustified expectations of favorable results64. Whether a given claim about results does so is for your counsel, who should read any such claim. Do not ask visitors to upload returns or statements. Count inquiries by page.
Days 30 to 60: intake and the fit filter
Change the contact form so that it asks what you need to judge fit: the service wanted, the kind of entity, the timing and the rough volume. It should not collect tax return information at inquiry; Chapter 4 and Chapter 3 explain why. Write down who answers an inquiry, how soon, and what that person says to someone who does not fit. If you decide to use an answering agent, Chapter 4 describes what it may do and what stays with staff, and you should count the calls it handled and the calls it handed to a person. Count the time from inquiry to first reply, and the inquiries that fit.
Days 45 to 75: review requests inside the rules
The FTC said its reviews rule prohibits businesses from providing compensation or other incentives conditioned on the writing of consumer reviews expressing a particular sentiment, positive or negative78. It also said the rule addresses reviews that misrepresent that they are by someone who does not exist, such as AI-generated fake reviews, and prohibits businesses from creating or selling them78. One design that is simple to square with that text is a single request, the same wording for every client, sent when an engagement closes, with nothing offered in return; counsel decides whether it fits. A reply to a review needs a second check: the AICPA rule says a member in public practice shall not disclose any confidential client information without the specific consent of the client64. Whether a public reply could be a disclosure under that rule is for counsel, and as a design matter a reply that carries no client information does not raise the question. Count requests sent and reviews received.
Days 60 to 90: read the count
Build the monthly page from the sheet: inquiries by source, inquiries that fit, engagements signed, hours and effective rate, and the share of runs in which the AI check names your firm, each with its range from Table 3. Expect the outcome counts to be small: at day 90 you will probably know the process facts with certainty (what was built and when) and the outcomes only roughly. Keep what you can defend, drop what you cannot explain, and carry the sheet into the next quarter. A year of records will begin to show retention.
Where our prices fit
We sell plans and an AI voice agent, so weigh this section accordingly. Technical Marketing's plans are priced at $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 term34. Our AI voice agent is $299 a month, added to a plan34. A firm can run every step in this chapter with a spreadsheet and its own time. The firm's own count can answer whether a plan is worth its price, and we expect to be held to that count: new clients a year by kind of client. Chapter 4's Table 3 shows that the lowest plan plus the agent is covered by 0.9 new advisory clients a year, or by 2.6 to 5.2 monthly bookkeeping clients, at the published fees. Our pricing page and our page for accountants say what is offered.
Limits of this paper and what would settle them
Table 5 lists what the paper cannot say. Wherever a figure is estimated or not measured, the number used in its place is labeled an estimate or a range.
| What the paper cannot say | What is held | What would settle it |
|---|---|---|
| Hours per engagement at accounting firms | Fees per engagement are published (Table 2); this paper does not measure hours per engagement | A firm's own timekeeping settles it for that firm; a published survey with a described sample would settle it for firms generally |
| Client retention at accounting firms | A survey median of 10% advisory client turnover and a trade panel's report of tenure (Table 2) | The firm's own client list, kept over several years |
| How often accounting firms miss calls or reply slowly | We estimate it in Chapters 1 and 4, from health care and law firm tests; no measurement at accounting firms | A test-call audit of accounting firms with a described sample |
| How clients choose accountants | A share of actual clients by source, taken from practice records, is NF; the nearest evidence is the surveys of stated sources by TaxDome† (practice management software) and Accounting Today28,29 | The source tagged on each inquiry and each signed engagement |
| What AI assistants say about a firm, over time | One run per question | Repeated runs, and checks through signed-in interfaces |
| The net effect of AI tools in small accounting firms | A field study of an agent answering inbound calls in accounting settings is NF; the nearest evidence is surveys of use and predicted savings (Chapter 5) | A counted trial of hours saved, review time and errors |
| The effect of any marketing step on one firm's inquiries | Large-advertiser research only | A long record kept the same way |
Table 5. What the paper cannot say, what is held, and what would settle each gap.
Two further limits apply across the paper. Search volume is interest, not purchase, and the phrases are ones a firm owner or a vendor would type, not ones a client would. Our study of AI answers asked each question once and 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 examples in Chapters 4, 5 and 6 use cited fees, rates and turnover, with our own prices. The agent is $299 a month with 250 call minutes included, about 62 four-minute calls (250 / 4), and $0.39 a minute after that, and the examples stay inside the included minutes34. Two links are an estimate and a range: how many first calls reach no one, and the share of rescued inquiries that become clients. The examples show how the pieces fit and how large a quantity would have to be to matter; they do not say what any firm will find.
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.
- AICPA. Accounting Firms Say Hiring Outlook Is Strong, AICPA Report Finds (news release on 2025 Trends). AICPA & CIMA, 2023-24. aicpa-cima.com/news/article/accounting-firms-report-strong-hiring-outl
- Technical Marketing. Who AI recommends locally: a study of local answers. Technical Marketing, 2026-11. technicalmarketing.ai/reports/ai-local-recommendations/ Our own data.
- Technical Marketing. What service businesses search for: 2026 demand data. Technical Marketing, 2026-11. technicalmarketing.ai/reports/service-business-search-demand/ Our own data.
- U.S. Department of the Treasury / Internal Revenue Service. 26 CFR 301.7216-3 - Disclosure or use permitted only with the taxpayer's consent. Legal Information Institute, Cornell Law School (e-CFR text), 2026-11-10. law.cornell.edu/cfr/text/26/301.7216-3
- David A. Wood and co-authors (Issues in Accounting Education 38(4): 81-108). The ChatGPT Artificial Intelligence Chatbot: How Well Does It Answer Accounting Assessment Questions?. Issues in Accounting Education (American Accounting Association); abstract on the Penn State research portal, 2023-01. pure.psu.edu/en/publications/the-chatgpt-artificial-intelligence-chatb
- Marc Eulerich, Aida Sanatizadeh, Hamid Vakilzadeh, David A. Wood. Is it all hype? ChatGPT's performance and disruptive potential in the accounting and auditing industries. Review of Accounting Studies (Springer), 2024. link.springer.com/10.1007/s11142-024-09833-9
- U.S. Bureau of Labor Statistics. QCEW open data: 2024 annual averages, NAICS 541211 Offices of certified public accountants (all areas). U.S. Bureau of Labor Statistics, 2024. data.bls.gov/cew/data/api/2024/a/industry/541211.csv
- U.S. Bureau of Labor Statistics. QCEW open data: 2024 annual averages, NAICS 541213 Tax preparation services (all areas). U.S. Bureau of Labor Statistics, 2024. data.bls.gov/cew/data/api/2024/a/industry/541213.csv
- U.S. Bureau of Labor Statistics. QCEW open data: 2024 annual averages, NAICS 541219 Other accounting services (all areas). U.S. Bureau of Labor Statistics, 2024. data.bls.gov/cew/data/api/2024/a/industry/541219.csv
- U.S. Bureau of Labor Statistics. QCEW open data: 2024 first quarter, NAICS 541211 Offices of certified public accountants (all areas). U.S. Bureau of Labor Statistics, 2024-Q1. data.bls.gov/cew/data/api/2024/1/industry/541211.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
- National Association of State Boards of Accountancy (NASBA). How Many CPAs Are There?. NASBA, 2026-08-27. nasba.org/licensure/how-many-cpas/
- Internal Revenue Service. Tax Professional Management Office federal tax return preparer statistics. Internal Revenue Service, 2026-09-01. irs.gov/tax-professionals/tax-professional-management-office-federal-t
- U.S. Bureau of Labor Statistics. Occupational Outlook Handbook: Accountants and Auditors. U.S. Bureau of Labor Statistics, 2025. bls.gov/ooh/business-and-financial/accountants-and-auditors.htm
- AICPA. Pool of Accounting Graduates Continues to Shrink in U.S., AICPA Report Finds. AICPA & CIMA, 2021-22. aicpa-cima.com/news/article/pool-of-accounting-graduates-continues-to-
- AICPA. Diversity Hiring of Graduates into the Accounting Profession is Increasing: AICPA Trends Report. AICPA & CIMA, 2019-20. aicpa-cima.com/news/article/diversity-hiring-of-graduates-into-the-acc
- Sweeney, J. T.; Summers, S. L.. The Effect of the Busy Season Workload on Public Accountants' Job Burnout. Behavioral Research in Accounting (American Accounting Association), 2002. doi.org/10.2308/bria.2002.14.1.223
- Persellin, J. S.; Schmidt, J. J.; Vandervelde, S. D.; Wilkins, M. S.. Auditor Perceptions of Audit Workloads, Audit Quality, and Job Satisfaction. Accounting Horizons (American Accounting Association), 2019. doi.org/10.2308/acch-52488
- Internal Revenue Service. Filing season statistics for week ending April 3, 2026. Internal Revenue Service, 2026-04-03. irs.gov/newsroom/filing-season-statistics-for-week-ending-april-3-2026
- Internal Revenue Service. Filing season statistics for week ending April 10, 2026. Internal Revenue Service, 2026-04-10. irs.gov/newsroom/filing-season-statistics-for-week-ending-april-10-202
- Internal Revenue Service. Filing season statistics for week ending May 8, 2026. Internal Revenue Service, 2026-05-08. irs.gov/newsroom/filing-season-statistics-for-week-ending-may-8-2026
- 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
- 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
- Makoto Nakayama; Norma Sutcliffe; Yun Wan. Has the web transformed experience goods into search goods?. Electronic Markets (Springer), 2010. doi.org/10.1007/s12525-010-0041-z 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.
- Johannes Gluckler and Thomas Armbruster. Bridging Uncertainty in Management Consulting: The Mechanisms of Trust and Networked Reputation. Organization Studies / SAGE Publications, 2003. doi.org/10.1177/0170840603024002346 Held through its record or abstract; passage not matched.
- Jason Bramwell. What Do Businesses Actually Value When Choosing or Staying with an Accounting Firm?. CPA Practice Advisor (reporting TaxDome's 2025 Niche Business Accounting Report), 2025. cpapracticeadvisor.com/2025/10/29/what-do-businesses-actually-value-wh Secondary source.
- Daniel Hood. Half of firms' clients have been with them 8+ years. Accounting Today (Arizent), 2019. accountingtoday.com/list/half-of-firms-clients-have-been-with-them-8-y Secondary source.
- CPA.com. Not all leads are treated equally. CPA.com, 2024. cpa.com/sites/cpa/files/2024-11/Not_all_Leads_are_Treated_Equally.pdf Secondary source.
- Internal Revenue Service (survey by Pacific Consulting Group). Comprehensive Taxpayer Attitude Survey, Executive Report (January 2022), IRS Publication 5296. Internal Revenue Service, 2021. irs.gov/pub/irs-prior/p5296--2022.pdf
- AICPA and CIMA. U.S. Accounting Firms Show Strong Growth in Profit and Revenue, AICPA and CIMA. AICPA & CIMA, FY2022. aicpa-cima.com/news/article/u-s-accounting-firms-show-strong-growth-in
- American Institute of CPAs. CPA Firms Report Steady Growth in Revenue and Profit, AICPA Research Finds. AICPA, FY2024. aicpa-cima.com/news/article/cpa-firms-report-steady-growth-in-revenue-
- Technical Marketing. Pricing. Technical Marketing, 2026-11. technicalmarketing.ai/pricing/ Our own data.
- CPA.com and AICPA Private Companies Practice Section (PCPS). Findings and insights from the 2024 CPA.com & AICPA PCPS client advisory services (CAS) benchmark survey. CPA.com / AICPA, 2024. cpa.com/sites/cpa/files/2024-12/2024-CAS-Benchmark-Survey.pdf
- CPA Practice Advisor. Ignition Report Shows Shift in Pricing for Accounting Firms. CPA Practice Advisor (reporting Ignition's benchmark), 2025. cpapracticeadvisor.com/2025/08/21/ignition-report-shows-shift-in-prici Secondary source.
- National Association of Tax Professionals. NATP TAXPRO guide to the 2025 Fee Study (TAXPROPages). National Association of Tax Professionals, 2025. dam.natptax.com/m/39796eb5ee56a576/original/TAXPROPages.pdf
- NATP Media Relations Team. 2025 NATP Fee Study - Media Highlights. National Association of Tax Professionals, 2025. dam.natptax.com/m/5f112786acbd1130/original/2025_NATP_Fee_Study_MediaH
- Ignition. The end of hourly billing: Ignition's 2025 benchmark signals pricing shift for firms. Ignition (Ignition Software), 2025. ignitionapp.com/news/the-end-of-hourly-billing-ignitions-2025-benchmar 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.
- 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
- Invoca. The Healthcare Lead Conversion Benchmarks Report 2026. Invoca, 2026. invoca.com/reports/the-invoca-healthcare-lead-conversion-benchmarks-re Secondary source.
- 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
- 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
- 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
- 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.
- Conductor. Financials Industry: 2026 AEO / GEO Benchmarks. Conductor, 2025. conductor.com/academy/financials-aeo-geo-benchmarks/ 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.
- 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.
- 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.
- 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
- Bain & Company, Global Consumer Lab. Why Consumers Choose an AI Chatbot Over a Search Engine (snap chart). Bain & Company, 2025. bain.com/insights/why-consumers-choose-an-ai-chatbot-over-a-search-eng Secondary source.
- Sina Gogani-Khiabani, Ashutosh Trivedi, ShinPing Chyi, Saeid Tizpaz-Niari. Performance of LLMs on VITA test: potential for AI-assisted tax returns for low income taxpayers. Artificial Intelligence and Law (Springer); record in the NSF Public Access Repository, 2025. par.nsf.gov/biblio/10649839
- 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
- Vivek Pandya. AI Traffic Drives 83% Higher Revenue Per Visit. Adobe, 2026. business.adobe.com/blog/ai-traffic-higher-revenue-per-visit-retail-hol Secondary source.
- 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
- Liu, Zhang, and Liang. Evaluating Verifiability in Generative Search Engines. Association for Computational Linguistics, 2023. arxiv.org/abs/2304.09848
- 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.
- Adam E. Block, Medha Reddy and Ella Corkum. The Impact of 1-Star Physician Ratings on New Patient Volume. The American Journal of Managed Care, 2015-01-01 to 2018-04-25. ajmc.com/view/the-impact-of-1-star-physician-ratings-on-new-patient-vo
- American Institute of CPAs. AICPA Code of Professional Conduct, 1.600.001 Advertising and Other Forms of Solicitation Rule. AICPA (copy hosted by the Minnesota Board of Accountancy; text bears © 2026 AICPA), 2026. mn.gov/boa/assets/AICPA%20Code%20of%20Professional%20Conduct_tcm1233-7
- Texas State Board of Public Accountancy. 22 Tex. Admin. Code section 501.82 - Advertising. Cornell Law School Legal Information Institute (reproduction of the Texas Administrative Code), 2026-11-10. law.cornell.edu/regulations/texas/22-Tex-Admin-Code-SS-501-82
- U.S. Department of the Treasury. 31 CFR 10.30 - Solicitation. Legal Information Institute, Cornell Law School (e-CFR text), 2026-11-10. law.cornell.edu/cfr/text/31/10.30
- Kelly D. Mullins (AICPA Professional Ethics Division). Accounting ethics in the age of AI. Journal of Accountancy (AICPA & CIMA), 2026-08-05. journalofaccountancy.com/news/2026/aug/accounting-ethics-in-the-age-of
- United States Congress (Internal Revenue Code). 26 U.S. Code 7216 - Disclosure or use of information by preparers of returns. Legal Information Institute, Cornell Law School (U.S. Code text), 2026-11-10. law.cornell.edu/uscode/text/26/7216
- United States Congress (Internal Revenue Code). 26 U.S. Code 6713 - Disclosure or use of information by preparers of returns. Legal Information Institute, Cornell Law School (U.S. Code text), 2026-11-10. law.cornell.edu/uscode/text/26/6713
- U.S. Department of the Treasury / Internal Revenue Service. 26 CFR 301.7216-1 - Penalty for disclosure or use of tax return information. Legal Information Institute, Cornell Law School (e-CFR text), 2026-11-10. law.cornell.edu/cfr/text/26/301.7216-1
- U.S. Department of the Treasury / Internal Revenue Service. 26 CFR 301.7216-2 - Permissible disclosures or uses without consent of the taxpayer. Legal Information Institute, Cornell Law School (e-CFR text), 2026-11-10. law.cornell.edu/cfr/text/26/301.7216-2
- Federal Trade Commission. FTC Safeguards Rule: What Your Business Needs to Know. Federal Trade Commission, 2026-11-10. ftc.gov/business-guidance/resources/ftc-safeguards-rule-what-your-busi
- Internal Revenue Service. Publication 5708: Creating a Written Information Security Plan for your Tax & Accounting Practice (Rev. 8-2024). Internal Revenue Service, 2024-08. irs.gov/pub/irs-pdf/p5708.pdf
- Federal Trade Commission. 16 CFR 314.4 - Elements. Legal Information Institute, Cornell Law School (e-CFR text), 2024. law.cornell.edu/cfr/text/16/314.4
- Federal Trade Commission. 16 CFR 314.5 - Effective date. Legal Information Institute, Cornell Law School (e-CFR text), 2024-05-13. law.cornell.edu/cfr/text/16/314.5
- 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
- Federal Trade Commission. 16 CFR 465.2 Fake or false consumer reviews, consumer testimonials, or celebrity testimonials. Legal Information Institute (copy of the e-CFR 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. 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
- Federal Trade Commission. Civil Penalty Inflation Adjustments (Federal Trade Commission notice, document 2026-18853). Federal Register (public inspection copy), 2026. public-inspection.federalregister.gov/2026-18853.pdf
- HHS Office for Civil Rights. Dental Practice Pays $10,000 to Settle Social Media Disclosures of Patients’ Protected Health Information. U.S. Department of Health and Human Services, 2019-10-02. hhs.gov/hipaa/for-professionals/compliance-enforcement/agreements/elit
- HHS Office for Civil Rights and Elite Dental Associates. Resolution Agreement and Corrective Action Plan, Elite Dental Associates (HHS TN 16-240465). U.S. Department of Health and Human Services, 2016-06-04. hhs.gov/sites/default/files/elite-dental-ra-cap.pdf
- U.S. Securities and Exchange Commission. Investment Adviser Marketing: A Small Entity Compliance Guide. U.S. Securities and Exchange Commission, 2022-11-04. sec.gov/resources-small-businesses/small-business-compliance-guides/in
- Financial Industry Regulatory Authority. FINRA Rule 2210: Communications with the Public. FINRA, 2026-11-10. finra.org/rules-guidance/rulebooks/finra-rules/2210
- George J. Stigler. The Economics of Information. Journal of Political Economy (University of Chicago Press), 1961. doi.org/10.1086/258464
- Michael Spence. Job Market Signaling. The Quarterly Journal of Economics (Oxford University Press), 1973. doi.org/10.2307/1882010
- A. Parasuraman; Valarie A. Zeithaml; Leonard L. Berry. SERVQUAL: A Multiple-Item Scale for Measuring Customer Perceptions of Service Quality. Marketing Science Institute, 1986. msi.org/working-paper/servqual-a-multipleitem-scale-for-measuring-cust
- 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.
- CPA Practice Advisor. Senior Accountants Crushed by Busy Season Stress. CPA Practice Advisor, 2025. cpapracticeadvisor.com/2025/08/26/senior-accountants-crushed-by-busy-s Secondary source.
- 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
- Federal Communications Commission. Implications of Artificial Intelligence Technologies on Protecting Consumers From Unwanted Robocalls and Robotexts (notice of proposed rulemaking), 89 Fed. Reg. 73321. Federal Register / U.S. Government Publishing Office, 2024-09-10. govinfo.gov/content/pkg/FR-2024-09-10/html/2024-19028.htm
- 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 13-77-101 Definitions. Utah State Legislature (le.utah.gov), 2026-05-06. le.utah.gov/xcode/Title13/Chapter77/C13-77-S101_2026050620260506.html
- United States Congress. 47 U.S.C. 227: Restrictions on use of telephone equipment. Legal Information Institute, Cornell Law School (US Code text), 2026. law.cornell.edu/uscode/text/47/227
- 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 13-77-105 Enforcement. Utah State Legislature (le.utah.gov), 2025-05-07. le.utah.gov/xcode/Title13/Chapter77/C13-77-S105_2025050720250507.html
- Christopher C. Rivers, Tribunal Member. 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
- Utah Legislature. Utah Code 13-77-102 Liability for violation of consumer protection law. Utah State Legislature (le.utah.gov), 2026-05-06. le.utah.gov/xcode/Title13/Chapter77/C13-77-S102_2026050620260506.html
- U.S. Bureau of Labor Statistics. Receptionists, Occupational Outlook Handbook. U.S. Bureau of Labor Statistics, 2025-05. bls.gov/ooh/office-and-administrative-support/receptionists.htm
- Kevin N Griffith et al.. Call Center Performance Affects Patient Perceptions of Access and Satisfaction. American Journal of Managed Care, 2014-10. pmc.ncbi.nlm.nih.gov/articles/PMC8177735/
- 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.
- Intuit QuickBooks. Accountants embrace AI and strategic advisory services to fuel growth, despite tech and talent challenges, according to 2025 Intuit QuickBooks Survey. Intuit (accounting software vendor), 2025-04. quickbooks.intuit.com/r/news/intuit-quickbooks-tech-survey/ Secondary source.
- Intuit. Intuit's 2026 Accountant Technology Survey: AI, fragmentation, and the future of the profession. Accounting Today (partner insights, Intuit), 2026. accountingtoday.com/partnerinsights/article/intuit/intuits-2026-accoun Secondary source.
- Karbon. State of AI in Accounting 2025 Report. Karbon (practice management software vendor), 2025. karbonhq.com/resources/state-of-ai-accounting-2025/ 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.
- 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
- 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.
- 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
- 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
- Noy and Zhang. Experimental evidence on the productivity effects of generative artificial intelligence. Science, 2023. doi.org/10.1126/science.adh2586
- Brynjolfsson, Li, and Raymond. Generative AI at Work. National Bureau of Economic Research, 2023. doi.org/10.3386/w31161
- 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
- Anders Humlum, Emilie Vestergaard. Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI (NBER Working Paper 33777; previously 'Large Language Models, Small Labor Market Effects'). National Bureau of Economic Research, 2026-03. nber.org/papers/w33777
- Internal Revenue Service, Office of Professional Responsibility. Introductory Guidelines for Responsible AI Use in Federal Tax Practice (OPR Alerts, Issue Number 2026-19). Internal Revenue Service (GovDelivery bulletin), 2026-06-24. content.govdelivery.com/accounts/USIRS/bulletins/41d6e70
- U.S. Department of the Treasury. 31 CFR 10.22 - Diligence as to accuracy. Legal Information Institute, Cornell Law School (e-CFR text), 2026-11-10. law.cornell.edu/cfr/text/31/10.22
- U.S. Department of the Treasury. 31 CFR 10.35 - Competence. Legal Information Institute, Cornell Law School (e-CFR text), 2014-06-12. law.cornell.edu/cfr/text/31/10.35
- J. Michael Reese, J.D., LL.M.. Should I disclose my use of gen AI to clients?. Journal of Accountancy (AICPA & CIMA), 2025-04-01. journalofaccountancy.com/issues/2025/apr/should-i-disclose-my-use-of-g Secondary source.
- 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
- 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), 2015. 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
- Micala Ricketts (Canopy). How Tax Firms Are Pricing Their Tax Preparation Services in 2020. Canopy, 2020. getcanopy.com/blog/how-tax-firms-price-tax-preparation-services Secondary source.
- 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.
- Fader. “Counting Your Customers” the Easy Way: An Alternative to the Pareto/NBD Model. INFORMS, 2005. doi.org/10.1287/mksc.1040.0098
Cite this paper
Mike Golliet. "Accounting firm and CPA marketing: what the evidence and the rules say." Technical Marketing, November 10, 2026. https://technicalmarketing.ai/papers/accounting-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.