25 ChatGPT Prompts for Lawyers (Contract Review, Discovery, Client Comms)

Important. This article is written for licensed attorneys, paralegals, and law students. ChatGPT is a writing tool, not a legal research database, and not a substitute for professional judgment. Every prompt below is for drafting, organisation, or communication support. None produces legal advice. Do not rely on ChatGPT output for citations to specific statutes, cases, or regulations without independent verification on Westlaw, Lexis, or primary sources; the model will fabricate plausible-looking citations that do not exist. Do not paste privileged material, client confidences, PII, or sealed records into the free tier; check your firm's AI-use policy and your state bar's guidance on confidentiality and competence (ABA Model Rules 1.1 and 1.6) before adopting any of these into a workflow. Nothing here is legal advice. The use of any of these prompts is at your own professional risk and remains subject to your duties to your client, the court, and the bar.

Most ChatGPT prompts for lawyers you find online are written by people who have never sat through a deposition or worked a deadline week. They suggest using the model for things lawyers should never use it for, and they bury the ethics caveats at the bottom. This post does the opposite. The boundaries are set in the disclaimer above and again in the section below before any prompts. The 25 prompts only cover what is actually safe and useful: drafting starting points, organisational structure, internal memos, client communications, and practice-management text.

We tested one of these prompts on free ChatGPT and reproduced the actual response further down so you can see exactly what the output looks like before you commit a single billable minute to it.

What NOT to ask ChatGPT (read this before any prompt below)

These are out of scope for the prompts in this post and for ChatGPT generally on the free tier. Each one has a reason rooted in either an ethics rule, a confidentiality risk, or the model's known failure modes.

Citations to specific cases, statutes, or regulations. ChatGPT routinely produces realistic-looking but fabricated citations. Two attorneys in Mata v. Avianca (2023) were sanctioned for filing a brief with hallucinated cases pulled from ChatGPT. Use the model to outline structure and arguments; never let it pick the cases.

Anything containing client confidences or privileged material on the free tier. Inputs to free ChatGPT are not protected by attorney-client privilege and may be retained by the provider. Always use placeholders ([Client], [Matter], [Date], [Counterparty]) and substitute real values into the output yourself, locally, after generation.

Final court filings or signed documents. A first draft is fine. Anything you put your bar number on must be reviewed line by line by an attorney, with citations independently verified.

Conflicts checks, ethical-screen analysis, or trust-account decisions. These are professional-responsibility judgments, not drafting tasks.

Sealed records, grand jury material, attorney-client emails, or anything covered by a protective order. Treat the prompt window as a publication.

If a use case is not on this list and not explicitly inside one of the categories below, treat it as out of scope until your firm has approved it.

How to use these ChatGPT prompts for lawyers

Open free ChatGPT in a browser. Read the boundary list above first. Pick a prompt. Replace the bracketed parts with your real specifics, with placeholders for anything sensitive. Read the output with the same scepticism you read a junior associate's first draft, because that is what it is. If something is wrong, push back in the same chat: "the second paragraph cites a rule that does not exist; rewrite without citing any specific authority." That second turn is where ChatGPT earns its keep.

A practical rule we follow on every prompt below: the model proposes, the lawyer disposes. None of these prompts replaces analysis. They replace the blank page.

Contract review and drafting (Prompts 1-5)

1. Plain-language contract summary for a client

Prompt: "You are a senior associate at a transactional firm. I am pasting a section of a commercial contract. Summarise it in plain English for a non-lawyer client at a 9th-grade reading level. Include: what each party is agreeing to do, key dates and deadlines, what triggers termination, and any clause that should prompt a follow-up question to me. Format: numbered list of points, then a one-paragraph 'what to ask about' summary. Do not provide legal conclusions; flag unclear parts for attorney review. Section: [paste section]."

The "what to ask about" close is the line clients actually use. It transforms a wall of text into three concrete decisions they can make.

2. Spot risk-shifting clauses in a draft section

Prompt: "You are a contracts attorney reviewing a draft for buyer-side risk. Read the section below and flag every clause that shifts risk to the buyer in a way that should be negotiated. For each, output a 3-column row: clause snippet, why it is risky, suggested counter-language. Be specific about which words you would change. Do not provide final filing language; this is a redline starting point. Section: [paste section]."

The three-column output is what makes this useful. A long bullet list of "concerns" is hard to action. A row per clause is reviewable in five minutes.

3. Draft a confidentiality clause for a specific scenario

Prompt: "You are drafting a mutual NDA clause for [the scenario, e.g., 'a 60-day evaluation period where two SaaS companies share roadmaps and customer pipelines to discuss a possible reseller deal']. Write a single confidentiality section covering: definition of confidential information, exclusions (already public, independently developed, etc.), permitted disclosures, term length, and return-or-destroy obligation at the end of the period. Length: under 350 words. Plain language where possible."

Use this for the first draft of a one-off NDA, then run it past your standard NDA library before sending. Do not skip the standard library.

4. Compare two redline versions and surface meaningful changes

Prompt: "You are a senior associate. Two versions of a contract section are below. Version A is our last draft; Version B came back from opposing counsel. Identify every change that is more than a stylistic edit, and group them into 'concession to us', 'concession from us', and 'neutral cleanup'. Output as a table. Then write a 4-sentence summary of where the negotiation now stands. Do not assume bad faith; describe the changes neutrally. Version A: [paste]. Version B: [paste]."

This is where ChatGPT genuinely beats a human first read. The model is consistent at flagging which side moved, where, and by how much. The lawyer still picks which fights to take.

5. Draft a termination-for-convenience clause

Prompt: "You are a contracts attorney. Draft a termination-for-convenience clause for [the contract type, e.g., 'a 12-month services agreement with a 30-day notice period and pro-rata refund of unused prepaid fees']. Include: notice period, format of notice, effect on payment obligations, transition assistance, and survival of confidentiality. Write the clause itself, then a 2-sentence note explaining the trade-offs of this version vs. a 60-day-notice version."

The trade-off note is the part that gets cut from most form-bank versions. It is the part the partner asks about in review.

Discovery and document review (Prompts 6-10)

6. Generate a document review tag taxonomy

Prompt: "You are a litigation associate building a document review protocol for [the case theory, e.g., 'a breach of fiduciary duty case alleging a former CFO diverted vendor payments to a related entity']. Suggest 8-12 review tags grouped into: hot documents, key dates, key people, key topics, and privilege flags. For each tag, give a one-sentence definition reviewers can apply consistently. Format: bulleted list grouped by category."

A consistent tag definition saves a re-review cycle two weeks in. The model is good at the definitions; you fix the tag list to your case.

7. Draft a deficiency letter to opposing counsel

Prompt: "You are responding to opposing counsel's incomplete document production. The deficiencies are: [list specifics, e.g., 'no responsive emails for the three custodians for January 2024; no metadata on Excel files; privilege log lacks subject lines for 47 entries']. Draft a 4-paragraph deficiency letter that identifies each gap with reference to the relevant Rule (do not invent specific rule numbers; use placeholders like '[Rule X]' that I will fill in), asks for a meet-and-confer within 7 days, preserves objections, and is professional in tone. Sign-off block included. Plain English where it works."

Note the explicit instruction to not invent rule numbers. This is the single most useful constraint to add to any legal prompt. The model will otherwise produce confident-looking citations that map to nothing.

8. Outline a document production schedule

Prompt: "You are project-managing a document production for a case with [estimated volume, e.g., '120,000 documents across 14 custodians']. Sketch a week-by-week production schedule covering: collection, deduplication, search-term application, first-pass review, privilege review, redaction, production export, and privilege log. Note assumptions (review rate, vendor turnaround). Output as a table with weeks, deliverable, and lead-responsible role."

This is a project-management prompt, not a legal one, but it is the discovery prep work that most often runs late. The output gives you a defensible plan to share with the client and the partner.

9. Spot privilege issues in a document description

Prompt: "You are conducting a privilege review. The document description is: [paste description, e.g., 'email from in-house GC to VP of Sales and outside accountant, subject: Q3 forecast and contract risk']. Flag the privilege questions raised by this description (attorney-client, work product, common interest, waiver) and what additional facts you would need to make a final call. Frame as questions, not conclusions. Then suggest the most likely correct privilege call if forced to decide on the metadata alone."

Frame as questions is the load-bearing instruction. The model will otherwise give you a confident-sounding legal conclusion. You do not want that.

10. Draft interrogatory questions for a specific case theory

Prompt: "You are drafting interrogatories for [the case theory, e.g., 'a wrongful termination claim alleging retaliation for an HR complaint about safety violations']. Write 10 interrogatories tailored to the theory, numbered, each calling for specific identifiable information (dates, identities of decision-makers, documents referenced). Avoid compound questions and overbroad time periods. Output as a numbered list. After the list, add a one-line note on the strategic purpose of each interrogatory."

The strategic-purpose note is the QC layer. If the model cannot articulate why an interrogatory matters, you do not need to ask it.

Legal research starters (Prompts 11-15)

11. Outline an issue spotter for a fact pattern

Prompt: "You are a junior associate. Read the fact pattern below and produce an issue spotter for senior-attorney review. Output as: (1) every potential cause of action visible in the facts, with a one-sentence elements check; (2) every potential defense the other side might raise; (3) any threshold issues (statute of limitations, jurisdiction, standing); (4) facts you would want to develop before forming a view. This is a starting point for research, not a legal conclusion. Fact pattern: [paste]."

Treat the model's output as a checklist of things to verify, not as analysis. Half its suggestions will be off-base. The other half will be the issue you would have missed at 10pm on a Friday.

12. Generate research questions from a case theory

Prompt: "You are planning a legal research memo. The case theory is: [the theory, e.g., 'whether a software vendor's automatic-renewal clause is enforceable against a small business buyer in California']. Write 6 specific research questions that, if answered, would determine the strength of this theory. Each question should be narrow enough that a Westlaw search would produce a useful answer in under 30 minutes. Group as: statutory questions, case-law questions, regulatory questions."

Narrow research questions beat broad ones. This prompt forces narrowness; the model is bad at it on its own.

13. Build a memo skeleton for a specific issue

Prompt: "You are outlining an internal research memo on [the issue, e.g., 'whether non-compete clauses in employment agreements are enforceable against W-2 employees in [state] post-FTC rule']. Produce a memo skeleton with: question presented, brief answer, facts assumed, applicable rule placeholder, discussion outline (3-5 sub-issues), and conclusion placeholder. Do not fill in substantive law; this is structure only. All cite-points should be left as [cite needed]."

The "do not fill in substantive law" line is the safety. Without it, the model will write a fluent paragraph of plausible-but-wrong analysis. With it, you get a clean skeleton.

14. Compare statutory frameworks across states (caveats required)

Prompt: "You are a research associate. Compare how [the issue, e.g., 'employee non-solicitation clauses'] are treated in [State 1] vs. [State 2] at a high level. For each: identify the type of authority (statute, leading case, regulatory) without inventing a specific cite, summarise the test in 2-3 sentences, note any safe harbour or exception. End with a one-paragraph note on which jurisdiction is more permissive and what facts would change the answer. I will verify all citations against primary sources before relying on this; treat your output as a research starting point, not authority."

The instruction to not invent cites is, again, doing all the work. Pair this output with a 30-minute Westlaw session. Do not skip the Westlaw session.

15. Identify likely defenses to a claim

Prompt: "You are a defense attorney evaluating [the plaintiff's claim, e.g., 'a tortious-interference claim brought by a former vendor against our client']. List every defense you would investigate, ordered from strongest to weakest given typical fact patterns of this claim type. For each: a one-sentence statement of the defense, what facts you would need to support it, and the risk of pursuing it. End with the one defense you would prioritise on day one."

The "ordered from strongest to weakest" forces a real ranking. Without it, the model produces a flat list of equal-looking options.

Client communications (Prompts 16-20)

16. Status update email about a case milestone

Prompt: "You are a senior associate at a litigation firm. Write a status update email to a business-owner client about [the case milestone, e.g., 'we've completed initial document review and are filing our response to the opposing party's motion to compel by Friday']. Include: a 2-3 sentence plain-language summary of where the case stands, what to expect next and by when, any action item the client needs to take (or write 'no action needed'), and a closing line inviting questions. Tone: professional and reassuring. No Latin or legalese. Length: 150-200 words. Format: subject line, then body."

The 'no action needed' line is the part clients remember. They want to know whether they have to do anything, more than they want to know what is going on.

What this prompt actually returns (tested 2026-05-07)

We ran Prompt 16 on free ChatGPT with the example milestone baked in ('we've completed initial document review and are filing our response to the opposing party's motion to compel by Friday'). Reproduced verbatim:

Here's a polished client update email in a professional, reassuring tone:

---

Subject: Case Status Update

Dear [Client Name],

I wanted to provide a brief update on the case. We have completed our initial review of the documents produced to date and have finalized the key points for our response to the opposing party's motion to compel. At this stage, the case is moving forward on schedule, and we are in a solid position to address the issues raised in the motion.

We expect to file our response with the court by Friday. After that filing, the court may either schedule a hearing or issue a written ruling, which typically occurs within the following several weeks. We will keep you informed as soon as we receive any updates or deadlines from the court.

No action is needed from you at this time. If we determine that additional information or documents would be helpful, we will reach out promptly with specific requests.

Please feel free to contact me if you have any questions or would like to discuss the case in more detail.

Best regards, [Your Name]

Editor's notes. Five things worth flagging before you copy this. First, the subject line is too generic. "Case Status Update" identifies nothing; if the client has more than one matter open with the firm, change it to something like "Case Status Update, [Matter Name], Week of May 5" before sending. Second, the opener "I wanted to provide a brief update on the case" is the warm-up the prompt explicitly told the model to skip; lead with the milestone instead ("We've completed our initial document review and are filing our response to the motion to compel this Friday"). Third, the email contains a soft prediction about outcome that was not requested ("we are in a solid position to address the issues raised in the motion"), which makes a representation about case strength that you should delete unless you actually want to make it. Fourth, "Friday" is named but the concrete date is not; swap in "Friday, May 9" so the client does not have to count. Fifth, two placeholders ([Client Name] and [Your Name]) need filling in. Net: usable as a 75 percent draft; total edit time before sending is about two minutes if you are firm about cutting the first opener and the soft prediction.

17. Engagement letter intro for a new matter

Prompt: "You are drafting the cover paragraph and scope-of-engagement section of an engagement letter for [the matter, e.g., 'a single-issue trademark application for a small CPG brand']. Cover: scope of work in plain English, what is NOT included (named explicitly), fee structure (flat fee, hourly, hybrid), responsibility for filing fees and disbursements, and a brief note on conflicts. Length: under 400 words. Tone: warm but precise. Add a bracketed [insert standard terms] tag for the back-end clauses you handle separately."

The "what is NOT included" section reduces fee disputes more than any other piece of letter writing. Make it specific.

18. Difficult-news email about a ruling against the client

Prompt: "You are emailing a long-term client about [the bad ruling, e.g., 'the court denied our motion to dismiss this morning']. Write a 4-paragraph email that leads with the news, explains in 3 sentences what the ruling means for the case, lays out the next 30 days, asks for a 15-minute call to talk through options, and avoids over-promising on outcomes. Tone: direct and steady. No defensiveness, no fluff."

Lead with the news is the rule. Not "I hope this email finds you well." Not "I wanted to share an update." The first sentence carries the news.

19. Bill explanation that anticipates objections

Prompt: "You are a partner explaining a higher-than-expected invoice to a sophisticated business client. The bill total is [amount] for [matter type, e.g., 'a 6-week regulatory response that grew in scope after we received the agency's second request']. Write a 3-paragraph email that describes what was actually done at the timekeeper level, explains the two specific drivers of the variance vs. the estimate, and invites a conversation if the client wants any portion adjusted. No apology that concedes the work; just clarity and openness."

The "no apology that concedes the work" line keeps the email professional. You can be open to discussion without retroactively writing off real time.

20. Settlement offer evaluation talking points

Prompt: "You are preparing for a 20-minute call with [the client, e.g., 'an in-house GC of a mid-size technology company'] about an incoming settlement offer. The offer is [the terms, e.g., '$2.4M, mutual release, 30-day acceptance window, no admission of liability']. Produce talking points covering: the upside of accepting, the upside of refusing, the realistic range of outcomes if we go to trial, a cost-of-trial estimate, and your recommendation with one sentence of reasoning. End with the 3 questions you want answered before the client decides."

Talking points are not the call. They are the structure that lets you actually listen on the call. The 3 questions at the end are the part that earns the next 20 minutes.

Practice management and marketing (Prompts 21-25)

21. Solo-practice intake questionnaire

Prompt: "You are designing the intake questionnaire for a solo employment-law practice that takes plaintiff-side cases. Draft 12 questions, in conversational order, that capture: what happened, when, who was involved, employer details, prior internal complaints, documents already gathered, applicable deadlines, prior attorney consultations, and budget expectations. Each question should be answerable in 1-2 sentences. End with one open-ended question for anything else the prospective client wants to share."

Conversational order matters more than thoroughness. A questionnaire that feels like a deposition causes prospective clients to drop off.

22. CLE attendance summary for compliance reporting

Prompt: "You are summarising [the CLE program] for your bar's annual compliance filing. Write a 4-sentence summary covering: program title, sponsoring organisation, primary substantive content, and the credit-hour breakdown (general, ethics, technology). Plain narrative, suitable to copy into a state bar reporting form. Do not exaggerate scope; describe only what was actually covered."

Sounds trivial. It is. That is why most lawyers procrastinate on it. Make this prompt the way you batch a quarter's filings on the last Friday of the quarter.

23. Bio paragraph for the firm website

Prompt: "Draft a 130-150 word bio for [the lawyer, e.g., 'a fifth-year associate practising commercial litigation with a focus on technology disputes; JD from [school]; admitted in [state]']. Include: practice focus in concrete terms (case types and industries, not buzzwords); one or two representative matter types (anonymised); bar admissions; education; one humanising line at the end (interest, community work, language). Avoid the words 'dynamic', 'passionate', 'results-driven'."

The banned-words instruction is not optional. The default model output uses every one of them.

24. Practice-area landing page outline

Prompt: "You are outlining a practice-area landing page for [the practice area, e.g., 'commercial real estate leasing for tenants']. Produce a page outline with: a one-sentence positioning statement; 4 sub-services covered (specific deliverables, not categories); 3 client types served; 4 representative matters described in one sentence each; an FAQ of 4 questions a prospective client would actually search; and a single call-to-action. Do not include attorney bios; those live elsewhere."

The "questions a prospective client would actually search" is the FAQ instruction that produces useful FAQs. Without it, you get the standard question list every other firm uses.

25. LinkedIn post for a case milestone (de-identified)

Prompt: "You are drafting a LinkedIn post about [the case milestone, e.g., 'a successful summary-judgment motion in a contract case']. Constraints: client and counterparty are not named; jurisdiction stays general; no claims of unique skill; no superlatives ('record-breaking', 'historic'). Structure: one-sentence hook describing the issue, 3 sentences on what made it interesting (the legal question, not the bragging), 2 sentences on the broader takeaway for in-house counsel, no more than 2 hashtags. 110-130 words."

Most lawyer LinkedIn posts read as if drafted by marketing. The constraints in this prompt produce something closer to a colleague-to-colleague note. That is what gets reposted.

Tips for getting better output

If the first response is generic, your bracketed specifics were too generic. Add one concrete fact, then push back. "The second paragraph is too cautious; rewrite it more directly" is a one-second turn that often produces the version you actually wanted.

Keep a private library of prompts that produced good output for your practice. The exact phrasing that hits your tone is hard to reconstruct from memory. Save the winners, version-controlled, in a single place.

Treat ChatGPT as a junior associate in their first month. Useful for first drafts, structure, and brainstorming. Not yet trustworthy on substantive law without verification. Never trustworthy on citations.

FAQ

Is it ethical to use ChatGPT for legal work? The ABA's Formal Opinion 512 (2024) and most state bar guidance treat generative AI as a tool that does not, by itself, create an ethics problem, provided the lawyer maintains competence (Rule 1.1), preserves client co