How to Brief ChatGPT Like a Junior Associate: Writing Prompts That Work for Legal Tasks

A two-word prompt like “review this” gives ChatGPT almost nothing to work with. The model does not know the client, the jurisdiction, or which side the lawyer represents. So ChatGPT guesses at those details. The answer still sounds confident, which makes a wrong guess easy to miss.

Lawyers already know how to fix this, since briefing a junior associate works the same way. A partner who wants a useful first draft explains the matter before handing it off. ChatGPT needs that same briefing. This guide covers what to put in it and what to leave out.

Name the Governing Law and Your Client’s Side Before ChatGPT Guesses

ChatGPT cannot reliably tell which jurisdiction applies unless the prompt says so. Without that detail, the model answers in general terms or mixes rules from different jurisdictions. Limitation periods, notice requirements, and enforceability rules all change from one jurisdiction to the next. Naming the governing law at the top of the prompt keeps the answer inside the right rules.

The client’s position shapes the answer just as much. An indemnity clause reads as a risk to one party and a protection to the other. A prompt that says “we represent the buyer” gets a review aimed at the buyer’s exposure. Leaving the side out produces a neutral summary that serves no one.

Leverage belongs in the briefing too. A party that can walk away negotiates differently from one that must close by Friday. ChatGPT cannot see that pressure unless the lawyer describes it in the prompt.

Writing these details from scratch on every matter slows the work down. Starting from a tested ChatGPT prompt for lawyers puts the jurisdiction, the client, and the facts into fill-in slots. 

Describe the Finished Work Product So ChatGPT Writes in the Right Format

A partner handing off work always says what the result should look like. ChatGPT needs the same instruction, since a request to “summarize this contract” can return three lines or three pages. Stating the format up front saves an editing pass later.

Four details give the model a clear target.

• Who will read it, such as a partner, a general counsel, or a business owner with no legal training.
• What form it takes, such as a memo, a checklist, or a short client email.
• How long it should run, in words or paragraphs.
• What tone it needs, from formal for opposing counsel to plain for a client update.

A prompt that ends with “Format the answer as a one-page memo for a general counsel, under 400 words” leaves the model little to guess. The lawyer gets a draft in the right shape on the first pass.

Format instructions work best on tasks that suit a general chatbot. Legal Desire’s guide to artificial intelligence (AI) for lawyers names a few good fits for these tools. Those include summarizing a document the lawyer already holds and rewriting a client email in plain English. 

Correct a Near-Miss in the Same Chat Instead of Starting Over

A draft that comes close needs a targeted correction. Lawyers mark up a junior’s work with specific notes. ChatGPT responds well to the same treatment, since the model keeps earlier instructions in view within one conversation, though very long threads can lose track of early details.

Specific feedback gets specific fixes. “Make the limitation of liability clause mutual and cap it at 12 months of fees” gives the model a clear task. “Make it better” gives the model nothing to act on.

Quoting the exact sentence that needs work keeps the edit narrow. That focus makes it easier to compare the new version against the old one.

Opening a fresh chat usually throws away the facts, the jurisdiction, and the format already provided. The lawyer then has to brief the model again from the top. Sending one correction per message also helps, because it shows which instruction the model followed and which it missed.

Keep Strategy and Final Judgment Out of the Prompt

A good briefing also sets limits on what ChatGPT decides. The model can list the positions available on an indemnity clause. Choosing which one to concede depends on the client relationship and the deal pressure, which the model cannot see.

Outcome predictions carry the same problem. A confident forecast from ChatGPT comes from patterns in text. The model knows nothing about the judge, opposing counsel, or the case record, so its prediction works only as a starting point for discussion.

Professional rules point the same way. The American Bar Association (ABA) issued Formal Opinion 512 on July 29, 2024. The opinion applies existing duties of competence, confidentiality, communication, and fees to generative AI tools. Under Opinion 512, the lawyer stays responsible for reviewing AI output before relying on it.

Client details need the same restraint. Consumer ChatGPT plans can use conversations to train OpenAI’s models unless the user opts out, while business plans exclude that data by default. 

Opinion 512 calls for informed consent before client information goes into a tool that learns from its inputs, so remove names, deal values, and identifiers before pasting.

Careful prompting reduces cleanup, but it does not eliminate the review. Every citation, statute, and figure in a ChatGPT draft still needs checking against a primary source before it reaches a client or a court.

A Well-Briefed Prompt Gives Lawyers a Better First Draft

ChatGPT works from the briefing it receives, so a thin prompt produces a thin draft. Naming the governing law and the client’s side gives the model the context a junior associate would ask for. Describing the finished format gives it a target to aim at.

Corrections in the same thread keep that context intact while the draft improves. The lawyer still makes the strategic calls and checks the final work. With a clear briefing, that review starts from a draft much closer to usable.