AI Prompts
How Lawyers Are Using AI in 2026: 5 Tasks and Prompts That Work
Five legal tasks where AI helps most, with copy-paste prompts, the ethics rules that apply, and the citation mistakes that keep landing lawyers in trouble.
Lawyers mostly use AI for research, drafting correspondence, summarizing documents and brainstorming. The tools save real time, but they also invent case law with complete confidence, so every output needs a lawyer's check before it reaches a client or a court. Below are five tasks where AI is worth using, a prompt for each, and the ethics rules that apply.
How many lawyers use AI in 2026?
Most of them. In the 8am 2026 Legal Industry Report, a survey of 1,300 legal professionals, 69% said they use general-purpose AI tools such as ChatGPT or Claude for work, up from 31% the year before. 42% use legal-specific AI tools, and 46% of firms have rolled out general-purpose AI firm-wide (8am, March 2026).
The rules and training haven't kept up. Only 9% of respondents said their firm has an AI policy that is actively enforced, and 43% have no policy and no plans for one. 54% said their firm offers no AI training and doesn't plan to.
The time savings are real but modest so far. 38% of 8am respondents save 1 to 5 hours a week. In Thomson Reuters' 2025 Future of Professionals survey, legal professionals expected AI to free up nearly 240 hours a year (Thomson Reuters, 2025). Clio's 2026 report on solo and small firms adds a caveat: many are working faster with AI without earning more, and firms that bill by the hour may see revenue shrink (Clio, 2026).
The biggest risk: fake citations
Before the prompts, the part that ends careers. AI tools can produce case names, quotes and holdings that don't exist, and they present them in perfect citation format.
The best-known example is Mata v. Avianca in 2023, where two New York lawyers filed a brief citing six fictitious cases generated by ChatGPT. Judge P. Kevin Castel fined them and their firm $5,000 (Bloomberg Law, June 2023). It was far from the last. Researcher Damien Charlotin tracks court decisions that found a party relied on AI-hallucinated content. As of October 25, 2026, his database lists 2,217 cases, 879 of them involving lawyers and 1,514 in the US (AI Hallucination Cases Database).
In England, the High Court used two such cases in June 2025 to warn the whole profession. In one, a solicitor relied on 18 fictitious cases from research his client had supplied. Dame Victoria Sharp said the court would refer the lawyers involved to their regulators (Legal Cheek, June 2025).
Legal-specific tools reduce the problem without removing it. A Stanford study published in the Journal of Empirical Legal Studies found that Lexis+ AI and Ask Practical Law AI produced hallucinated answers in about 1 in 6 test queries, and Westlaw's AI-Assisted Research in about a third. General-purpose GPT-4 did worse (Magesh et al., 2025).
So treat every case, quote and statute an AI gives you as unverified until you've read it in the original.
Five tasks where AI helps, and where it can hurt
| Task | Where AI helps | Main risk |
|---|---|---|
| Research planning | Mapping issues, standards and search terms | Invented cases and holdings |
| Contract review | A second read for risks, gaps and vague wording | Missing context only you know about the deal |
| Client updates | Turning legal detail into a clear, short email | Over-promising or a tone that doesn't fit the client |
| Deposition prep | A structured question outline | Generic questions that miss the facts of your case |
| Stress-testing | Arguing the other side against your position | Taking the AI's framing as the full picture |
Two rules for all five prompts. Anonymize anything confidential (use Party A and Party B, remove case numbers and identifying details) unless you're using a tool your firm has approved for client data. And read every output as a first draft from a junior colleague.
1. Research planning
General chatbots are good at helping you plan research and bad at being the source of it. Use this prompt to map the issue, then do the actual research in Westlaw, Lexis or the primary sources.
You are a senior legal researcher specializing in [PRACTICE AREA]. I'm building an argument on [DESCRIBE THE LEGAL ISSUE, ANONYMIZED] in [JURISDICTION]. Help me plan the research. Give me: (1) the legal questions I need to answer, in order, (2) the legal standard or test a court is likely to apply and where it usually comes from, (3) the strongest counterarguments I should expect, (4) search terms and synonyms to use in a legal research database. Do not cite any case, statute or quote unless you can link to the source, and mark anything you're unsure of as UNVERIFIED.
Asking for a research plan instead of a list of cases plays to what general AI tools do well, and the instruction to mark uncertain items gives you a checklist for verification. If you do use an AI research tool inside Westlaw or Lexis, the same rule applies: open and read every authority it cites.
2. Contract review
AI is useful as a second reader, especially late in the day when you're on the seventh contract. Anonymize the clause before you paste it in.
You are an experienced contract review attorney. Review the clause below for [MY CLIENT'S ROLE, e.g. the buyer / the seller / the licensee] and identify: (1) potential risks or exposure for my client, (2) ambiguous language that could lead to disputes, (3) protections that are usually included in this type of agreement but are missing, (4) suggested revisions with alternative wording, (5) any terms that may be unenforceable in [JURISDICTION], and why. Be specific about what to change and why. Clause: [PASTE ANONYMIZED CLAUSE]
Asking for alternative wording gets you something close to a redline instead of a list of concerns. The AI doesn't know the commercial background or what your client agreed to off the page, so its suggestions are a starting point for your judgment.
3. Client updates
Clients want to know where their matter stands, what it means for them and what happens next. This prompt turns your notes into a short, plain-language email.
You are a client-facing attorney at a [TYPE OF FIRM]. Draft a short email to my client, a [CLIENT TYPE, e.g. small business owner], with a status update. Current status: [WHERE THE MATTER STANDS]. Recent developments: [WHAT HAPPENED]. Next steps: [WHAT HAPPENS NEXT AND WHEN]. The email should be reassuring but honest about the timeline, explain any legal terms in plain language, include one clear action for the client if they need to do something, and offer a call to discuss. Under 150 words. Tone: professional and warm.
"Reassuring but honest" stops the AI from over-promising, and the word limit keeps the email short enough to be read. Add the client's name yourself after drafting, so it never goes into the tool.
4. Deposition prep
AI won't take the deposition for you, but it can give you a structured outline to work from.
You are an experienced trial attorney preparing for a deposition. The deponent is [ROLE, e.g. the opposing party / an expert witness / a corporate representative]. Case summary: [BRIEF, ANONYMIZED DESCRIPTION]. Key issues: [WHAT I NEED TO ESTABLISH OR CHALLENGE]. Prepare: (1) ten questions organized by topic, starting with foundation and building toward the key admissions, (2) three follow-up questions for likely evasive answers, (3) two questions to lock in testimony that supports our position, (4) the documents I should have ready for each topic. Approach: methodical and calm, focused on building a clear record.
The order from foundation to admissions follows how depositions are usually built, and asking for documents per topic helps you prepare exhibits. Expect to cut and rewrite: the AI only knows what you put in the summary.
5. Stress-testing your argument
Before you file a motion, ask the AI to argue the other side.
You are opposing counsel reviewing my argument. I plan to argue: [YOUR ARGUMENT]. Jurisdiction: [JURISDICTION]. Type of case: [TYPE]. Attack it as hard as you can: (1) the three strongest counterarguments, (2) factual weaknesses in my position, (3) the kind of authority opposing counsel would rely on, marked UNVERIFIED, (4) procedural or evidentiary problems I haven't addressed, (5) a score for my argument from 1 to 10, with reasons. Be blunt. I want problems, not reassurance.
The "be blunt" instruction is there for a reason. AI assistants tend to agree with the person using them. OpenAI rolled back a ChatGPT update in April 2025 because it had become too sycophantic (OpenAI, May 2025). Asking the model to take the other side gets you a more honest read of the weak spots.
The ethics rules that apply
In July 2024, the ABA issued Formal Opinion 512, its first ethics opinion on generative AI. It applies the existing Model Rules to AI tools (ABA, July 2024):
- Competence (Rule 1.1). Lawyers need a reasonable understanding of the benefits and risks of the AI tools they use.
- Confidentiality (Rule 1.6). Client information must stay confidential unless the client gives informed consent, which matters before you enter client information into an AI tool.
- Communication (Rule 1.4). You have to keep clients reasonably informed about how their matter is handled, and that can include your use of AI.
- Fees (Rule 1.5). You may bill for the time spent entering information and reviewing the output, but in most cases you can't charge the client for learning how to use the tool.
Check how your tool handles data. On ChatGPT's consumer plans, conversations can be used to train OpenAI's models unless you turn off "Improve the model for everyone" in Data Controls (OpenAI). Claude's consumer plans ask users to choose whether their chats can be used for training, while Anthropic's Team and Enterprise plans are covered by separate commercial terms (Anthropic, August 2025). Many firms keep client work on enterprise or legal-specific tools for this reason. I wrote more on what the 2026 court rulings mean for ChatGPT privacy.
It also helps to keep a short file note of which tool you used, for what task, and how you checked the output. State bars and courts are issuing their own guidance as well, so check the rules where you practise. This section is a plain-English summary, not legal or ethics advice.
Which AI tool should a lawyer use?
For anything involving client data or case law, a legal-specific or enterprise tool your firm has approved is the safer choice, because of the confidentiality terms and the link to real legal databases. Stanford's study shows those tools still make mistakes. General assistants like ChatGPT, Claude and Gemini work well for drafting, summarizing anonymized text and planning. My comparison of ChatGPT, Claude, Gemini and Grok covers how they differ.
Start with one prompt
Pick the task that matches something on your desk right now. The client update prompt is a good low-risk place to begin. Open your firm's approved tool, paste the prompt, fill in the brackets and check the result.
If you want the method behind these prompts, read how to write AI prompts that work. If you're new to this, start with what prompt engineering actually is. If you run a firm, it's also worth checking what AI assistants say when a prospective client asks them for a lawyer like you: see is your business invisible to ChatGPT and Claude? and the AI for law firms page.
Frequently asked questions
How are lawyers using AI in 2026?
Mostly for general research, drafting correspondence, brainstorming and summarizing documents, according to the 8am 2026 Legal Industry Report. Legal-specific tools are used mainly for legal research and document drafting.
What percentage of lawyers use AI?
In the 8am 2026 Legal Industry Report, 69% of legal professionals said they use general-purpose AI tools for work, up from 31% a year earlier. 42% use legal-specific AI tools.
Can lawyers use ChatGPT for legal research?
For planning research, yes. As a source of authority, no. General chatbots can invent cases and quotes. Use them to map issues and search terms, then research in Westlaw, Lexis or the primary sources and verify every citation.
How many court cases involve AI hallucinations?
Damien Charlotin's AI Hallucination Cases Database listed 2,217 court decisions as of October 25, 2026 where a party was found to have relied on AI-hallucinated content. 879 of them involved lawyers.
What is ABA Formal Opinion 512?
It's the American Bar Association's first ethics opinion on generative AI, issued in July 2024. It covers competence, client confidentiality, communication with clients and reasonable fees when lawyers use AI tools.
Do legal AI tools like Lexis+ AI and Westlaw still hallucinate?
Yes, less often than general chatbots. A Stanford study found hallucinations in about 1 in 6 queries for Lexis+ AI and Ask Practical Law AI and about one-third for Westlaw AI-Assisted Research.
Can lawyers bill clients for time saved with AI?
Under ABA Formal Opinion 512, fees must be reasonable. Lawyers may bill for time spent entering information and reviewing AI output, but in most cases can't charge clients for learning how to use the tool.
Where to go from here
Start with one low-risk task, keep a note of what you checked, and build from there. Lawyers who get the most from AI tend to treat it as a fast junior colleague whose work always needs review.
If you'd like one short, practical AI note a week, The Risk Memo goes out every Tuesday.
Researched with AI tools, written and fact-checked by Dominic.
Sources
- 8am, Generative AI data from the 8am 2026 Legal Industry Report (Mar 20, 2026)
- Thomson Reuters, Future of Professionals Report 2025 (Jun 2025)
- Clio, Legal Trends for Solo and Small Law Firms (2026)
- Bloomberg Law, Phony ChatGPT Brief Leads to $5,000 Fine for NY Lawyers (Jun 22, 2023)
- Damien Charlotin, AI Hallucination Cases Database (accessed Oct 26, 2026)
- Legal Cheek, High Court warns lawyers over AI use after 'fake' cases cited in submissions (Jun 9, 2025)
- Magesh et al., Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools, Journal of Empirical Legal Studies 22 (2025)
- OpenAI, Expanding on what we missed with sycophancy (May 2, 2025)
- American Bar Association, ABA issues first ethics guidance on a lawyer's use of AI tools (Jul 29, 2024)
- OpenAI Help Center, Data controls in ChatGPT
- Anthropic, Updates to our consumer terms (Aug 28, 2025)
Last updated: