Neither is built or certified for legal work, and no published test we found shows one is better. Your ethics duties are the same whichever you pick. Compare them on terms: whether inputs train the model, who can see them, retention, and the administrator controls on the plan you would buy. Then test both yourself.
What the evidence covers
A published head-to-head of Claude and ChatGPT on legal tasks is NF; the nearest evidence tested ChatGPT 4 alone. On specific, verifiable questions about random federal court cases it produced legal hallucinations 58% of the time1. The sanctions cases that reached court involved ChatGPT, including a $5,000 penalty in Mata v. Avianca2, and that says something about how widely it is used and nothing about whether another assistant would do better. Do not read a missing Claude case as a clean record.
Terms depend on the plan
The same assistant can be covered by different terms on a free account, a paid individual plan and a business plan. In our review of vendors' business associate agreement practice, ChatGPT and Claude each sign one only on certain plans3. That is a health-privacy point and not a legal-ethics one, but it shows the pattern: read the terms of the specific plan, not the product's marketing page.
ABA Formal Opinion 512 says many self-learning tools are designed so their output could lead directly or indirectly to disclosure of client information, and a client's informed consent is required before you input information relating to the representation4. Whether a given plan is a self-learning tool is a question about its terms, which is where the comparison belongs.
Answers vary from run to run
In SparkToro and Gumshoe.ai's 2026 study, volunteers ran 12 prompts through ChatGPT, Claude and Google's AI Overview, 2,961 times in all. There was less than a 1 in 100 chance that a prompt asked 100 times would return the same list of brands twice5. That study concerned brand lists and not legal work, but it is a reason to run each test prompt several times before you believe a single result.
A test you can run this week
Pick five questions from closed matters, with the client details removed. Give each assistant the same prompts. Check every authority it cites against the reporter or the database of record. Record fabricated citations, wrong holdings and anything it refused. Compare how each handles a question when you say you are unsure of the law.
Then compare terms, line by line, for the plan you would pay for: training on inputs, human review, retention and deletion, who can read your content at the vendor, and what an administrator can switch off. Put the findings in front of whoever approves tools at your firm.
We are not lawyers. Read the rule yourself or ask your counsel how it applies to you.
People also ask
Do I need to tell clients I use generative AI?
Sometimes. You must disclose if a client asks or requires it, and you need informed consent before client information goes into a self-learning tool. Internal use with no client information may need none, and your state's opinion may add to this.
What is the best AI for lawyers?
No single product is best, and we have not tested enough to rank them. Match the tool to the task and read the vendor's terms before it sees a file.
What kinds of AI tools do lawyers use?
General assistants such as these two, legal research platforms, document tools and front-office tools such as AI receptionists. They carry different data terms and different risks, so approve each kind separately.
How should I train my lawyers and staff on AI?
Teach your written policy and run exercises on anonymized matters, including one where people find invented citations in a draft. Include nonlawyer staff and repeat the session when you add a tool.
Can AI handle intake for a law firm?
It can collect a name, a number and the matter type and pass them on, while saying it is an AI assistant. It should not advise callers or accept a matter.
How to read this page: NF not found, ( ) our arithmetic on cited figures, and a small number points to a source below.
Sources
- Matthew Dahl, Varun Magesh, Mirac Suzgun, Daniel E. Ho. Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models. arXiv (v2, last revised 21 Jun 2024), 2024. arxiv.org/abs/2401.01301
- Judge P. Kevin Castel, U.S. District Court, S.D.N.Y.. Mata v. Avianca, Inc., No. 22-cv-1461 (PKC), Opinion and Order on Sanctions (Doc. 54). U.S. District Court for the Southern District of New York (RECAP/CourtListener copy of the filed order), 2023-06-22. storage.courtlistener.com/recap/gov.uscourts.nysd.575368/gov.uscourts.
- Technical Marketing. Which software will sign a HIPAA BAA? 28 common tools, checked. Technical Marketing, 2026-11. technicalmarketing.ai/reports/hipaa-baa-availability/ Our own data.
- ABA Standing Committee on Ethics and Professional Responsibility. ABA Formal Opinion 512: Generative Artificial Intelligence Tools. American Bar Association, 2024-07-29. americanbar.org/content/dam/aba/administrative/professional_responsibi
- 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.