easyai › ChatGPT for lawyers

ChatGPT for lawyers

Two rules come before any use case. It fabricates case citations that look entirely correct, and lawyers have been sanctioned for filing them — never cite anything you have not pulled up yourself. And client confidences stay out of it until your firm has a policy. Within those limits it is genuinely good at drafting routine correspondence, summarising documents you give it, and turning your own drafting into plain English.

Two rules before anything else

Most guides on this topic list use cases and bury the risks at the bottom. For this profession that is the wrong order, because both of the real problems can end a career rather than waste an afternoon.

1. It invents case law, confidently. Ask for authorities and it will produce citations that look perfectly formed — correct reporter, plausible parties, plausible year — for cases that do not exist. Lawyers in several jurisdictions have been sanctioned for filing briefs containing them. Never cite anything it gives you that you have not pulled up yourself in a real database. Treat every citation from it as a rumour.

2. Client confidences are not yours to paste. Privilege, professional secrecy and your data protection duties do not pause because a tool is convenient. Before any client matter touches it, you need to know what your firm's policy says and what the tool does with what you type. Describing a situation in the abstract is almost always enough.

Where it genuinely helps

With those two rules in place, the useful work is the writing and reading that fills a day and bills poorly.

Plain-English versions of your own drafting. Paste a clause and ask what it means to someone with no legal training. Excellent for client letters, and quietly excellent for spotting a clause that is doing less than you thought.
First drafts of routine correspondence. Chasers, acknowledgements, client updates. The letters you have written a thousand times and still have to write.
Reading a long document you did not write. A counterparty's bundle, a policy, a report. Summarise it, then interrogate it — it is far better at answering questions about a document you give it than at recalling anything from memory.
Turning a messy file into a timeline. Dates and events in, chronology out.
Preparing questions. For a client meeting, a call, a negotiation. It is good at the obvious question you were about to forget.
Second-guessing yourself. “What is the strongest argument against this position?” is a genuinely useful prompt.

The pattern: it is reliable when you give it the source material, and unreliable when you ask it to supply the source material.

What about Claude for legal work?

Some firms prefer Claude for long documents, and it handles large files comfortably. The two rules above apply identically. No current general-purpose assistant is a substitute for a legal research database, and none of them should be handed privileged material without your firm having decided that first.

Try it now: translate a clause into plain English

Our lessons run inside the tool rather than in a video. Paste this in and it walks you through it, one step at a time.

You are my patient tutor. I am a lawyer but I am NOT technical.
Teach me by having me do one thing at a time, not by explaining.

Rules for this whole conversation:
- ONE step. Then STOP and ask what happened. Wait for my answer.
- Under 150 words per message.
- Never suggest I paste client names, matter numbers, or anything
  that could identify a party. If I do, tell me which parts
  weren't needed.
- Never give me case citations. If I ask, remind me to verify
  authorities in a real database instead.

Lesson: turn a clause I have drafted into something a client
can actually understand.

1. Ask me to paste ONE clause, with names and identifying
   details replaced with [PARTY A] and [PARTY B].
2. Give me a plain-English version aimed at someone with no
   legal training. Then STOP and ask whether it still says
   what I meant.
3. Ask me what a client would most likely misunderstand about it.
4. Produce a short client-facing explanation I could actually send.
5. Tell me if anything in the original clause was ambiguous.

Start with step 1. Ask me now, and nothing else.

Step 5 is the one that earns its keep. Asking a fresh reader what is ambiguous in your own drafting is a use that has nothing to do with saving time.

What it will get wrong

Any citation, always. Covered above, and worth repeating because it is the failure that reaches a tribunal.
Current law. It does not reliably know what changed recently, and it will not tell you it does not know.
Jurisdiction. It slides between jurisdictions without noticing. If you do not state yours, assume it has guessed, and often guessed American.
Procedure and deadlines. Never take a limitation period or a filing requirement from it.
Confidence. It is equally fluent when right and when wrong. Fluency is not a signal here.

Questions

Can lawyers use ChatGPT?

Yes, and many do, but two rules come first. Never cite an authority it produces without pulling the case up yourself in a real database, because it fabricates citations that look correct. And never paste privileged or client-identifying material until your firm has a policy covering it. Within those limits it is genuinely useful for drafting, summarising and plain-English translation.

Does ChatGPT make up case law?

Yes. It will produce citations with a correct-looking reporter, plausible parties and a plausible year for cases that do not exist. Lawyers in several jurisdictions have been sanctioned for filing briefs containing fabricated citations. Treat every authority it gives you as a rumour until you have found it yourself.

Is it safe to put client information into ChatGPT?

Not without your firm deciding that first. Privilege, professional secrecy and data protection duties all still apply. In practice you rarely need the identifying details: describe the situation with the parties replaced by placeholders and you get the same quality of answer.

What is ChatGPT actually reliable for in legal work?

Anything where you supply the source material. Summarising a document you paste in, translating your own drafting into plain English, building a timeline from facts you give it, preparing questions for a meeting. It is reliable reading your material and unreliable recalling its own.

Can it replace legal research?

No. It is not a legal research database and does not reliably know current law. Use it to understand a document you already have, not to find out what the law is.

Is Claude better than ChatGPT for legal work?

Some firms prefer Claude for long documents and it handles large files comfortably, but the same two rules apply to both. Neither is a substitute for a research database, and neither should see privileged material before your firm has decided that.

Do I have to tell clients I used AI?

That depends on your jurisdiction and your retainer, and some regulators have begun issuing guidance. Check your own rules. The practical position is that work you have drafted, checked and taken responsibility for is your work.

Want the rest, free?

Six short lessons you paste straight into the tool. One real win in your first ten minutes.

Get the free kit