AI for lawyers: useful for drafting, dangerous without verification

Legal work runs on reading, drafting and precision, all of which language models are good at imitating. That is also the danger: an assistant can produce something that reads like a well-formed legal answer while being wrong. The lawyers getting value from AI are those who use it for the parts of the job where output is easy to check.

Where AI helps lawyers

Reading and summarizing. Give an assistant a long agreement, transcript or set of exhibits and ask for a structured summary: parties, dates, obligations, termination rights, and anything unusual. Ask it to cite the section or page of each point. Then check the ones that matter against the source.

Comparing to your standard. The most dependable contract use is checklist review. Write down your standard positions on indemnity, limitation of liability, governing law, term and assignment, then ask the assistant to compare a counterparty draft and list each deviation with the clause number. This is faster than reading cold, and every finding is verifiable. See can AI review contracts for the fuller picture.

First drafts. Engagement letters, client updates, cover emails and the first pass of an amendment are all faster to edit than to write. Treat the output like work from a junior colleague who has never seen the file.

Organizing the matter. Turn an intake email into a fact chronology, a list of open questions and a document request list.

Where AI does not help

Legal authority. Courts have sanctioned lawyers who filed briefs with citations an AI invented. This is a hallucination, and it is not fixed by asking the model to be careful. If a case, statute or quote appears in AI output, pull the primary source yourself before you rely on it.

Judgment about strategy or risk. A model can list arguments. It does not know your client’s tolerance, the judge, the counterparty’s history or the unwritten facts.

Completeness. Do not assume it caught everything. A clean summary does not mean there was nothing else in the document.

Risks specific to legal work

Privilege and confidentiality. Your duty of confidentiality applies regardless of the tool. Whether entering client information into a given AI service risks waiver or breaches a rule depends on that service’s terms, its data retention and training defaults, and your jurisdiction’s ethics guidance. Several state bars and the ABA have published guidance on generative AI, so read the version for the place you practice. Where the answer is unclear, do not put client-identifying information in.

Data settings. Consumer and business plans differ in how they treat your inputs. Anthropic documents that consumer conversations are used for training only if you opt in, while commercial products sit under separate terms. Check the current terms of whichever tool you use, and get your firm’s decision in writing. Start with the client work question.

Client consent and disclosure. Some clients now require or prohibit AI use in outside-counsel guidelines. Read them.

Supervision. Under the usual rules, a lawyer remains responsible for work product regardless of who or what drafted it.

Starter prompts

Checklist review, using a de-identified document:

You are helping me review a vendor agreement against our standard
positions. For each item below, quote the relevant clause with its
section number, say whether it matches our position, and describe
the difference. If a clause is missing, say it is missing.

Our positions:
1. Liability cap: 12 months of fees.
2. Governing law: Massachusetts.
3. Termination for convenience: 30 days notice.

Summary:

Summarize this document in a table: section, what it requires,
who is obligated, deadline. After the table, list anything
ambiguous. Quote the exact words for every deadline.

Your first week

• Day 1: Read your bar’s AI guidance and your firm’s policy. If there is none, use this question to start one.
• Day 2: Try a summary of a public document, such as a standard form agreement, and verify every point.
• Day 3: Write your standard positions on your five most common agreement types as a checklist.
• Day 4: Run one de-identified contract through the checklist and compare against your own review.
• Day 5: Record what it caught, what it missed, and what you will and will not use it for.

A worked example: NDA intake review

A business client sends a counterparty’s NDA and asks whether it is fine to sign. You remove party names if your policy requires it, and give the assistant your checklist: definition of confidential information, exclusions, term, permitted disclosures, residuals clause, governing law, remedies. Ask for a table with the clause number and a plain-English note for each item, and a separate list of clauses that are not on your checklist but look unusual.

You then read the flagged clauses in the actual document, and the ones the assistant said were fine at a glance. The value is that your attention goes to the right places first. The judgment on whether to sign, push back or walk away stays with you.

Common mistakes to avoid

• Asking for case law and pasting it into a filing. Pull every authority from a real database.
• Uploading a client document before checking the tool’s terms. Decide first, then upload.
• Treating a summary as complete. Ask it what it left out, and sample the source.
• Forgetting the client’s own restrictions. Outside counsel guidelines may address AI use directly.

What the sources say

Labor data from the BLS and the American Bar Association ethics opinion on generative AI.

• The Bureau of Labor Statistics projects employment of lawyers to grow 5 percent from 2025 to 2035, faster than the average for all occupations.
• The Bureau of Labor Statistics says some routine legal work may be automated, but this is not expected to reduce overall demand for lawyers.
• The American Bar Association says technological competence presupposes that lawyers remain vigilant about the tools’ benefits and risks.
• The American Bar Association says a client’s informed consent is required before inputting information relating to the representation into many self-learning generative AI tools.
• The American Bar Association notes that Model Rule 1.5’s prohibition on unreasonable fees may push lawyers toward new technology over slower methods.

If you want hands-on practice

Learn AI Fast is a two-day in-person class in Boston for non-developers that includes a session on contract and legal review, with a session on data privacy and company policy. It is not legal advice or a substitute for your bar’s guidance.

Sources

• U.S. Bureau of Labor Statistics, Occupational Outlook Handbook: Lawyers: “projected to grow 5 percent from 2025 to 2035, faster than the average for all occupations”
• U.S. Bureau of Labor Statistics, Occupational Outlook Handbook: Lawyers: “this increased efficiency is not expected to reduce overall demand for lawyers”
• American Bar Association, Formal Opinion 512: Generative Artificial Intelligence Tools (July 29, 2024): “technological competence presupposes that lawyers remain vigilant”
• American Bar Association, Formal Opinion 512: Generative Artificial Intelligence Tools (July 29, 2024): “informed consent is required prior to inputting information relating to the representation”
• American Bar Association, Formal Opinion 512: Generative Artificial Intelligence Tools (July 29, 2024): “prohibition on unreasonable fees, as well as market forces, may influence lawyers to use new technology”