What Legal Tasks Is ChatGPT Genuinely Good At?
ChatGPT is a language model. It excels at tasks where the raw material comes from you and the output is prose, not authority. So how can ChatGPT help lawyers day to day? For a litigator, that means:
- Brainstorming and issue-spotting. Ask it to poke holes in your theory of the case, generate counterarguments, or list angles you have not considered. You are using it as a sparring partner, and you verify everything against actual law yourself.
- First-draft correspondence. Client status updates, scheduling emails, engagement follow-ups. Anything routine where the facts are non-confidential or generalized.
- Summarizing and condensing your own text. Paste in your own draft (not client documents on a consumer account) and ask for a tighter version, a plain-English client explanation, or an executive summary.
- Reformatting and restructuring. Converting notes into an outline, turning a dense paragraph into a bullet list, adjusting tone from aggressive to conciliatory.
- Explaining unfamiliar concepts. A quick orientation to a technical or industry topic before you dig into real sources.
Every task on this list assumes a business-tier account with training disabled. Consumer and free ChatGPT tiers use your conversations to train the model by default, so pasting client facts into one is a confidentiality breach the moment you hit enter. And the model fabricates citations regardless of tier, which is sanctions exposure under Noland v. Land of the Free, L.P. (2025) 114 Cal.App.5th 426. Keep client materials off consumer accounts, and never let the model supply authority.
Notice what is missing from that list: legal research, citations, and confidential facts. How can lawyers use ChatGPT productively? By keeping it firmly on the language side of the line and never letting it supply the law.
If you want a repeatable method for structuring these tasks, our guide to the DRAFT method for AI-assisted legal work walks through a prompting framework that works across tools. Most "chatgpt prompts for lawyers" lists boil down to the same principles: give context, define the output, and never ask the model to supply authority.
Where Does ChatGPT Become Dangerous for Legal Work?
Two failure modes account for nearly every disciplinary and sanctions story involving lawyers using ChatGPT: fabricated citations and confidentiality breaches.
Hallucinated Citations: How ChatGPT Gets Lawyers Sanctioned
Language models generate plausible text. A citation that looks real, with a believable case name, reporter volume, and pinpoint page, is exactly the kind of plausible text they produce. The model has no concept of whether the case exists.
The pattern started with Mata v. Avianca, Inc. (S.D.N.Y. 2023) 678 F.Supp.3d 443 in New York, where attorney Steven A. Schwartz cited fake cases generated by ChatGPT in a federal brief. Judge P. Kevin Castel found six of the submitted cases were bogus judicial decisions with bogus quotes and bogus internal citations. Lawyers blame ChatGPT in these cases, but courts have uniformly placed responsibility on the attorney who signed the filing.
California now has its own published warning. In Noland, counsel acknowledged relying on AI to support his citations and claimed he did not know that generative AI frequently fabricates legal sources. The court sanctioned him $10,000 anyway, held the appeal frivolous under Code of Civil Procedure sections 907 and 128.7, and separately found the briefs violated the Rules of Court because points were not supported by citations to real authority (Cal. Rules of Court, rules 8.204(a)(1)(B), 8.276(a)(4)), and directed the clerk to notify the State Bar. (Noland v. Land of the Free, L.P. (2025) 114 Cal.App.5th 426.) Ignorance of how the tool works is not a defense. Under Rule of Professional Conduct 1.1, it is itself a competence problem.
Confidentiality on Consumer Accounts
The second danger is quieter. Consumer ChatGPT accounts use your conversations to train the model by default. Paste in a client's medical records, a confidential settlement demand, or privileged strategy notes, and you have transmitted client confidences to a third party under terms that permit reuse. That is a straight confidentiality problem regardless of whether anything ever surfaces.
Consumer ChatGPT accounts train on user conversations by default, so pasting client medical records, settlement demands, or privileged strategy notes into one is a confidentiality breach the moment you hit enter. Consumer accounts also fabricate citations, which is sanctions exposure under Noland. Never use a consumer ChatGPT account for client work, court filings, legal research, or anything touching confidential client facts.
Business-tier accounts are different: OpenAI states that by default it does not use business data for training. If your firm uses ChatGPT at all for anything touching client matters, an enterprise or business agreement with training disabled is the floor, not a nice-to-have. Our guide on how small firms can adopt AI safely covers the policy and disclosure framework in detail.
Can Lawyers Trust ChatGPT for Legal Research and Citations?
No. And the harder truth is that purpose-built legal research AI does not eliminate the problem either.
The Stanford HAI/RegLab legal hallucinations study tested hallucination rates across tools and found reviewers reported rates of 17% for Lexis+ AI, 33% for Westlaw AI-Assisted Research, and 43% for GPT-4. The purpose-built tools hallucinated less than the general-purpose model, but "less" is doing heavy lifting when one in six answers from the best performer contained hallucinated content – fabricated or misdescribed authority.
The sanctions record reflects this. Lawyers using AI tools including CoCounsel, Westlaw Precision, and Google Gemini have still submitted briefs with hallucinated citations. The lesson is not "buy the expensive research tool and relax." The lesson is that the verification duty is identical across every tool: pull the case, read it, confirm it says what the AI claims, before anything goes out under your signature.
No AI tool, general-purpose or legal-specific, produces citations you can file without independent verification. Noland makes clear that in California, filing unverified AI citations violates the duties of candor and competence and exposes you to monetary sanctions and a State Bar referral.
Can Lawyers Use ChatGPT Ethically? What California Rules Say
Can lawyers use ChatGPT ethically? Yes, and the framework is clearer in California than almost anywhere else.
Rule 1.1 and the technology competence duty. Comment [1] to Rule of Professional Conduct 1.1, added March 22, 2021, states that the duty of competence includes "the duty to keep abreast of the changes in the law and its practice, including the benefits and risks associated with relevant technology." Not knowing that ChatGPT fabricates citations is a competence failure, not an excuse.
The State Bar Practical Guidance. The State Bar's Practical Guidance on generative AI, originally published November 16, 2023 and since updated to address emerging developments including agentic AI, sets the core principle: your ethical obligations apply to generative AI the same way they apply to any other technology. Confidentiality, competence, candor, supervision, and reasonable fees all carry over.
Proposed rule changes. On August 22, 2025, the California Supreme Court directed the State Bar to consider incorporating the Practical Guidance principles into the Rules of Professional Conduct and to address agentic AI. These changes are proposed, not enacted, but the direction is unmistakable: today's guidance is on track to become tomorrow's enforceable rule.
Courts are regulating themselves too. The Judicial Council adopted Cal. Rules of Court, rule 10.430, requiring any California court that permits generative AI use by staff or judicial officers to adopt a use policy by December 15, 2025. When the judges reading your briefs operate under AI use policies, expect them to notice when you do not.
The practical takeaway: do lawyers use ChatGPT in California? Plenty do, lawfully. The ones who get sanctioned are the ones who skipped verification or ignored confidentiality settings.
Is ChatGPT for Lawyers Free to Use Safely?
Mostly no, with one narrow exception.
Free and consumer AI tools are dangerous for legal work. Consumer free tiers use your conversations to train models by default, which means pasting client facts into them is a confidentiality breach. And they fabricate citations, which is sanctions exposure under Noland. Never use a free or consumer AI tool for client work, filings, legal research, or anything touching confidential facts.
The economics of free tiers are also shifting. According to OpenAI's announcement and subsequent reporting, ChatGPT's free tier began showing ads in the United States in February 2026, which underscores what the product is: a consumer service monetized around your usage, not a professional tool with confidentiality obligations to you.
The narrow exception: free ChatGPT is acceptable for non-confidential language tasks only. Drafting a generic blog post, tightening marketing copy, explaining a technical concept, rewording a template paragraph that contains no client information. If a hypothetical opposing counsel could read the prompt without learning anything about any client, it passes. If you would hesitate before reading the prompt aloud in a crowded elevator, it fails.
Searching for "chatgpt for lawyers free" setups that handle client work safely is searching for something that does not exist. The free tier's terms are the problem, and no prompt phrasing fixes terms of service.
What Is the Best ChatGPT Alternative for Legal Work? Purpose-Built AI Compared
Honestly, and in both directions. ChatGPT is cheap, flexible, and genuinely good at language tasks. Purpose-built legal tools add domain structure and better data terms, but they do not remove your verification duty, and the Stanford HAI/RegLab data proves it.
| Criteria |
ChatGPT (consumer) |
ChatGPT (business tier) |
Harvey |
CoCounsel |
GC AI |
Legion |
| Built for legal work |
No |
No |
Yes |
Yes |
Yes (in-house focus) |
Yes (CA civil litigation) |
| Trains on your data by default |
Yes |
No |
No |
No |
No |
Never |
| Court-ready formatted output |
No |
No |
No |
No |
No |
Yes (CA pleading paper, TOC/TOA, captions) |
| CA procedure built in |
No |
No |
No |
Partial |
No |
Yes (e.g., auto CCP § 2030.050 declaration) |
| Pricing |
Free (ad-supported) |
Per seat, published |
Not published; sales call required |
Bundled into Westlaw/Practical Law contracts |
$500/seat/month, 14-day trial |
Usage-based, per page, no subscription |
| Prompt engineering required |
Yes |
Yes |
Some |
Some |
Some |
No |
Third-party pricing and availability reflect each vendor's published information as of early 2026 and may change; confirm current terms directly with each provider.
A few honest observations:
Harvey and CoCounsel are serious tools with real strengths. Harvey has enterprise traction at large firms. CoCounsel benefits from Thomson Reuters' content integration. But Harvey does not publish pricing, and CoCounsel is typically bundled into existing Westlaw and Practical Law contracts, which makes both hard for a small firm to evaluate on cost. And CoCounsel users have still been sanctioned for hallucinated citations, so neither tool substitutes for verification.
GC AI publishes transparent pricing at $500 per seat per month with a 14-day trial, but it targets in-house counsel workflows rather than litigation drafting.
Legion takes a different approach for California litigators specifically. It is a drafting platform, not a chatbot: it asks for what it needs (drag in raw materials or describe the case by voice), routes each workflow to the strongest frontier model for that task, and returns a formatted draft on California pleading paper with compliant captions, line numbering, and auto-generated tables of contents and authorities on briefs over 10 pages. California procedure is baked in; exceed 35 special interrogatories and it automatically prepares the CCP § 2030.050 Declaration for Additional Discovery for your review. Client materials are never used to train models, pricing is per page with no subscription, and you review the full draft before paying. Critically, every draft is prepared for attorney review and signature; the platform is built around the verification duty, not around pretending it does not exist. For a concrete look at outcomes, see how a San Francisco litigation firm cut staffing costs 27% using AI.
If the question is "what is the best ChatGPT for lawyers," the answer is usually reframing the question. For language tasks, a business-tier general model is fine. For litigation drafting, a purpose-built tool that outputs court-ready documents and never trains on client data fits the actual workflow.
How to Use ChatGPT Safely in Your Law Practice
If you decide ChatGPT has a place in your practice, set it up deliberately. Here is how to use ChatGPT for lawyers' everyday work without walking into the failure modes above:
- Use a business-tier account with training disabled. OpenAI's business terms exclude your data from model training by default. Consumer accounts do not.
- Adopt a written firm AI policy. Define permitted tasks (drafting, summarizing your own text, brainstorming) and prohibited ones (research, citations, confidential facts on non-compliant accounts). Our firm AI policy guide includes the full framework, including client disclosure considerations.
- Never ask it for authority. No cases, no statutes, no quotes from opinions. If it volunteers a citation anyway, treat it as presumptively fabricated until you have pulled and read the source. For discovery work, check objection language against a verified reference like our California discovery objections cheat sheet rather than trusting model output.
- Verify everything before filing. Noland is the controlling cautionary tale: $10,000, a State Bar referral, and a published opinion with counsel's name in it.
- Sanitize prompts. Even on business tiers, minimize client-identifying detail where the task does not require it.
- Train everyone who touches it. Rule 1.1's technology competence duty extends to supervising staff use. An associate's unverified ChatGPT citation is your problem.
The impact of ChatGPT on lawyers has been real but narrower than the headlines suggest: it changed how drafts get started, not how law gets verified.
Frequently Asked Questions
Do lawyers actually use ChatGPT?
Yes, widely, mostly for drafting correspondence, summarizing, brainstorming, and reformatting. The lawyers who make headlines are the ones who used it for legal research and filed the output unverified. The tool is common; the sanctions come from skipping verification.
Have lawyers been sanctioned for using ChatGPT?
Yes. The best-known federal example is Mata v. Avianca, Inc. (S.D.N.Y. 2023) 678 F.Supp.3d 443, where six fabricated ChatGPT cases drew sanctions. In California, Noland v. Land of the Free, L.P. (2025) 114 Cal.App.5th 426 imposed a $10,000 sanction, ordered counsel to serve the opinion on his client, and referred him to the State Bar for fabricated AI citations.
Will ChatGPT replace lawyers?
No. It replaces certain drafting labor, not legal judgment, verification, strategy, or accountability. Every published sanctions case reinforces that responsibility for filed work stays with the attorney, which is precisely the part no model can absorb.
Claude vs ChatGPT for lawyers: does the choice matter?
Less than you would think. Both are general-purpose language models, and the same rules apply to each: fine for non-confidential language tasks on properly configured accounts, unfit for legal research and citations, and off-limits for client confidences on consumer tiers. Your verification duty is identical either way.
Can I use ChatGPT for legal research if I double-check the results?
You can use it to brainstorm issues or generate search terms, but do not treat its answers as research. With GPT-4-class models hallucinating at reviewer-reported rates around 43% on legal questions in the Stanford HAI/RegLab study, you are verifying noise. Run actual research in a citator-backed database and read every case yourself.
Is a paid ChatGPT subscription enough to satisfy California ethics rules?
A business-tier account solves the default training-data problem, which addresses confidentiality. It does nothing about hallucinations, so the Rule 1.1 competence duty and the candor duty confirmed in Noland still require independent verification of every factual and legal assertion before filing.
What legal tasks is ChatGPT genuinely good at?
First-draft non-confidential correspondence, summarizing and tightening your own writing, reformatting notes into outlines, adjusting tone, and adversarial brainstorming against your case theory. Anything where you supply the substance and the model supplies the prose.
Take the Next Step
ChatGPT can start a draft. It cannot format a California pleading, track the 35-interrogatory limit, or produce a filing you would sign without rebuilding it. Legion can: court-ready California litigation documents, no prompt engineering, client materials never used for training, and per-page pricing with no subscription. Get started with Legion and see what a purpose-built drafting platform looks like on your next matter.
