Best AI for Legal Writing: Tools, Prompts, and a Review Process for Lawyers

The best AI for legal writing is the one matched to the document: a general assistant such as ChatGPT or Claude for rewriting, outlines, and first drafts without client data, and a legal platform grounded in case law for memos and briefs that cite authority. Whatever the tool, the lawyer checks every citation and signs the final work.

This guide separates the documents AI drafts well from the ones where it fails, matches each kind of legal drafting software to its job, and gives you prompts, a review routine, and the ABA rules to follow.

Which AI are law firms using for legal writing?

Writing fills a large share of a lawyer's billable week: research memos, briefs, motions, demand letters, contracts, and client emails. It is also where generative AI produces the most visible time savings, because almost every legal document follows a structure the model has seen many times.

The adoption numbers

The ABA's 2024 Legal Technology Survey Report found that 30% of responding lawyers used AI-based tools, up from 11% in 2023. Among firms using or considering AI, ChatGPT led at 52%. Solo practitioners and small firms leaned on it most, while large firms invested in legal platforms such as Harvey.

Drafting is a core use. According to the Thomson Reuters 2025 Generative AI in Professional Services report, 51% of legal professionals who use generative AI apply it to contract drafting, and 74% to document review. A 2024 Wolters Kluwer survey of lawyers in the US and nine European countries found that 68% of law firm lawyers used generative AI weekly and 58% of firms planned to invest more in AI within three years. The surveys measure different groups, but they point the same way.

Start from the tasks that eat your time

Before you pick a tool, list the writing tasks that take the most hours without needing your judgment, the same logic that drives a sound law firm content marketing plan.

What AI legal writing does well, and what it does badly

For writing, the useful question is narrow: which documents can a model draft well enough that your work shifts from writing to reviewing?

Where AI drafting saves the most time

Documents with a predictable structure come out best: a demand letter, an NDA, an engagement letter, or a memo built from authorities you already pulled. The model reproduces the mechanics; you adapt the draft to the facts. Long, disorganized material is the other sweet spot: AI turns exhibits, transcripts, or dictated notes into a clean chronology.

Where it fails

AI fails where the document depends on strategy or on authority. The argument section of a brief, a clause built to survive a specific challenge, or advice on a contested issue requires a reading of the record, the case law, and sometimes the judge. A model has none of that. Its draft sounds right, stays generic, can cite a case that does not exist, and is polished enough to pass a quick read.

Examples by practice area

• Business and transactional law: a first read of NDAs and commercial contracts, comparison of versions, and flags on sensitive clauses.
• Employment law: demand letter templates, summaries of personnel files and deposition transcripts, and plain-English updates for clients.
• Family law: factual timelines and first drafts of motions and supporting declarations.
• Litigation: brief outlines and a statement of facts from your notes.

The best AI for legal writing depends on the document

There is no single best AI for legal writing. Legal drafting software falls into three families, and most firms end up combining two of them. The question is not which brand tops a ranking, but which family fits the document in front of you.

General-purpose assistants

ChatGPT, Claude, Gemini, and Microsoft Copilot rewrite, outline, summarize, and produce first drafts quickly. They are excellent for prose and structure, but they are not connected to a verified case law database, so they can never be your source of authority. Use a business plan for anything close to client work, as our guide to ChatGPT for lawyers explains. If your firm runs on Microsoft 365, see our guide to Microsoft Copilot for lawyers.

Is there a legal version of ChatGPT?

Yes. Legal platforms such as CoCounsel from Thomson Reuters, the Lexis+ platform from LexisNexis, and Harvey let you ask questions and request drafts in plain English. Their answers are grounded in legal sources or your own documents, with links to citations you can check. They suit memos and briefs where authority matters. Our guides to AI legal research and legal AI software go further on these platforms.

Drafting tools inside Microsoft Word

Transactional lawyers often prefer a tool that works where the contract lives. Spellbook, for example, suggests redlines as tracked changes in a Word sidebar. For contract work in detail, read our guide to AI contract review, and for a broader list organized by job, our roundup of legal AI tools.

How to choose legal drafting software: 4 criteria

Before you compare brands, set your own grid. ABA Formal Opinion 512 expects you to understand a tool's terms of use and privacy policy before you rely on it, so four filters come first:

• Data terms: no training on your inputs, clear retention periods, a stated storage location, and an independent audit such as SOC 2.
• Confidentiality fit: contract terms that let you meet Rule 1.6, which requires reasonable efforts to prevent the unauthorized disclosure of client information.
• Grounding in US law: federal and state sources for every jurisdiction where you practice, a source behind each answer, and a citator check such as KeyCite or Shepard's.
• Integration: Word, Outlook, your practice management software, and your document management system, so drafts do not travel by copy and paste.

Test any candidate on a matter you know well before you sign.

Can ChatGPT write a legal contract?

It can write a first draft, not a contract you can send. That distinction answers most questions about AI drafting.

A first draft, not a final document

Ask a general model for a nondisclosure agreement or a services agreement and you get a plausible structure in seconds. But models learned from contracts written everywhere, so their drafts drift toward a generic standard. A US contract is governed by one state's law, and the rules that matter, such as limits on noncompetes or the conditions for a valid warranty disclaimer, change from state to state.

What to check before it goes out

• Governing law: name it in the prompt and check every clause against it.
• Your client's positions: the model does not know what your client can concede.
• Invented references: a statute or a "market standard" that does not exist.
• Client data: replace names and amounts with tags such as [CLIENT_A] or [AMOUNT], then restore them in Word.

Our article on ChatGPT in legal practice covers the broader risks, and our guide to legal document automation covers reusable templates.

Prompts for AI legal writing that produce usable drafts

A vague prompt gets generic text that takes longer to fix than to write. A useful legal prompt sets a role, the jurisdiction, the document type, the audience, and the output format, and asks the model to flag what it is unsure of. It never includes names, case numbers, or other identifying details.

Four prompts to adapt

• Research memo. "You are a litigation associate in Illinois. Using only the authorities I paste below, draft a two-page memo with a question presented, a short answer, and a discussion. Add no other case, and write [VERIFY] wherever you are unsure."
• Client email after a hearing. "Rewrite my hearing notes as a 200-word email to the client in plain English: what happened, what comes next, and what we need from them. Add no legal position."
• Chronology. "From these anonymized exhibit summaries, build a table of dates, events, and the exhibit each one comes from. Flag gaps and inconsistencies instead of filling them."
• Brief outline. "You are an appellate lawyer in Georgia. Outline the argument section of a brief on the issue below, with headings written as full-sentence assertions. Leave every citation as [AUTHORITY NEEDED]."

Why the placeholders matter

Telling the model to work only from what you supply, and to mark gaps instead of filling them, does not eliminate invented authority, but it shows you where to look first. Our collection of AI prompts for lawyers organizes more templates by task.

The risks of AI legal drafting and the ABA rules that apply

No US jurisdiction bans AI drafting. Your existing duties apply in full, and ABA Formal Opinion 512, issued on July 29, 2024, explains how. Two risks matter most for legal writing: invented citations and client confidentiality.

Hallucinated citations and court sanctions

In Mata v. Avianca, Judge P. Kevin Castel of the Southern District of New York sanctioned two lawyers and their firm $5,000 on June 22, 2023, for a brief citing six opinions ChatGPT had invented. The problem has not gone away: researcher Damien Charlotin's AI Hallucination Cases database listed 1,501 US court decisions involving hallucinated content as of October 16, 2026.

Legal platforms reduce the risk without removing it. Opinion 512 cites a 2024 Stanford study finding that the generative AI tools of leading legal research companies hallucinated between 17% and 33% of the time.

Client confidentiality under Rule 1.6

Pasting a client's facts into a third-party model discloses them to that provider. Opinion 512 requires the client's informed consent before you enter information relating to the representation into a self-learning tool, and a boilerplate clause in an engagement letter is not enough. The business tiers of ChatGPT and Claude do not train on your data by default, while ChatGPT's consumer plans may unless you turn training off. Anonymize before you paste, whatever the plan.

Competence, candor, communication, and fees

• Competence (Rule 1.1): relying on AI output without independent verification can breach this duty.
• Candor (Rule 3.3): a fake citation in a brief becomes a false statement of law to the tribunal, and in federal court your Rule 11 signature certifies that your legal contentions are warranted by existing law.
• Communication (Rule 1.4): tell clients about AI use when they ask, or when it matters to a significant decision in the matter.
• Supervision (Rules 5.1 and 5.3): managing lawyers set AI policies for lawyers and staff.
• Fees (Rule 1.5): bill hourly work only for time actually spent, prompting and review included.

Courts add their own layer: since Judge Brantley Starr's May 2023 standing order in the Northern District of Texas, a number of federal judges require lawyers to certify or disclose AI use in filings. Marketing has its own rules, covered in our guide to lawyer solicitation rules.

How to review AI-drafted legal writing before you sign

Review is where AI drafting becomes safe or sanctionable. Make it a fixed routine, written down and shared with everyone who drafts at the firm.

1. Read the draft line by line, not as a skim. Polished prose hides weak reasoning.
2. Pull every authority in Westlaw, Lexis, or the official court source. If you cannot find a case by citation and by party name, treat it as nonexistent.
3. Check that each case says what the draft claims, and run it through KeyCite or Shepard's to confirm it is still good law.
4. Check quotations and pin cites word for word, and confirm Bluebook format or your court's local citation rules.
5. Rewrite the strategic passages yourself: arguments, concessions, and advice are not cosmetic edits.
6. Label AI-assisted work in the file. Opinion 512 suggests marking AI-generated materials so future readers know the work may be fallible.

What AI legal writing is worth to your firm

The math is simple: time saved multiplied by your hourly rate. If a tool saves you a few hours a week on first drafts, it pays for itself quickly, even against a paid license. Measure the gain on your own matters, not a vendor's average.

Hourly billing adds a US twist. Under Rule 1.5 and Opinion 512, you cannot bill time you did not spend, so efficiency gains do not show up on an hourly invoice. The real return comes from what you do with the freed hours: more matters, more careful review, and business development.

Saved time should serve quality and growth, never replace human review. To turn it into new matters, see our guides to growing a law firm, LinkedIn for lawyers, and law firm SEO strategies.

How to bring AI drafting into your firm in 5 steps

Integration does not happen by itself. Roll it out in this order:

1. Map your time-consuming writing tasks for one week: which documents, how many, and how long each one takes.
2. Choose a tool that fits the job: a general assistant on a business plan for prose, a legal platform grounded in US law for anything that cites authority.
3. Read the data terms: no training on your inputs, defined retention, and no reuse of client documents.
4. Make human review mandatory, with the routine above written into firm policy.
5. Train the team and write a one-page AI policy: approved tools, approved uses, and the data that never goes into a prompt. Check each court's standing orders on AI disclosure as well.

Training matters most: the sanction cases all started with lawyers who trusted output they had not checked. A marketing agency for lawyers that knows your ethics constraints can help you structure the rest. For the broader picture, see our guide to AI for lawyers and our analysis of whether AI will replace lawyers.

Key takeaways

The best AI for legal writing is the one that fits the document: a general assistant for prose without client data, a legal platform grounded in case law for memos and briefs, and a Word tool for contracts. The time saved is real, but it only has value with systematic human review, citation checks, and a written firm policy.

Chosen and supervised well, AI legal writing returns time to the work clients pay you for. For a clear view of your tools and your online visibility, tell us about your firm.