ChatGPT for Legal Writing: Prompts, Review, and Better Drafts

Updated October 18, 2026.

AI can help you explore an outline, revise a difficult passage, or prepare a first draft. The useful skill is directing that work and deciding what deserves to make it into the final document. This guide offers prompts, a review sequence, and a worked example you can use to practice.

Training a law firm or legal department? Our live workshops help teams move from experimenting with AI to producing work they can review and use. See live AI training for legal teams.

In this guide

1. What legal writers need to know about AI
2. Where AI can help with legal writing
3. Limits to plan around
4. Set up a legal writing assignment
5. Write a prompt that is easy to review
6. Four legal writing prompts to adapt
7. Review AI-assisted legal work before sending
8. Worked example: polished wording, changed meaning
9. Terms for everyday AI work
10. Build your team’s AI writing skills

Start with the work you need to deliver: a client update, a research memo, a contract-risk summary, or a passage in a brief. Decide what the reader needs to understand or do. Then choose the parts of the assignment for which AI assistance would be useful.

A useful response gives you something you can inspect: alternative openings, a proposed outline, a list of issues to investigate, or a summary tied to identified sources. Fluent wording alone does not tell you whether the work is accurate, complete, or responsive to the assignment.

Keep a distinction between generating a draft and accepting it. You can ask a tool to flag uncertainty or identify support, but you still need to check whether its answer follows the instructions and whether the cited material supports the claim.

Consider these uses when you can provide appropriate inputs and evaluate the result:

• Explore structure. Ask for two ways to organize an argument or update around the reader’s decision. Compare what each version emphasizes or leaves out.
• Revise a passage. Request clearer sentences while preserving facts, qualifications, defined terms, and the degree of certainty. Inspect the changes, including changes that sound harmless.
• Summarize supplied material. Ask for source references alongside each material proposition, then compare the summary with the original. Identify what the summary omits.
• Find questions to investigate. Ask which facts, exceptions, or opposing arguments could change the analysis. Treat the response as a starting point for further work.
• Adapt for an audience. Turn a technical explanation into a proposed client update, then check that simplifying the language has not simplified away the problem.

Choose a task you can assess. If you cannot recognize a misleading result, arrange for the necessary expertise and review before relying on it.

Plan for plausible errors: a citation that does not support its proposition, a missing exception, an invented fact, or a confident recommendation that goes beyond the available evidence. An instruction to “use only these sources” helps define the task; it does not prove that the response complied.

Tool capabilities, document handling, and data terms vary. Do not assume a tool has read every attachment, can access a linked source, or will retain all the context from an earlier exchange. Check the particular product and configuration your team uses.

ABA Formal Opinion 512 (2024) explains that lawyers need an appropriate understanding of their AI tools and independent review suited to the task. It also addresses confidentiality, communication, supervision, candor, and fees. The lawyer remains responsible for the work. The opinion interprets the ABA Model Rules; check applicable jurisdictional rules, court requirements, and client obligations for your matter.

Before entering client information, evaluate the tool’s terms, access, and data handling with the people responsible for your organization’s approvals. Determine whether the intended use and inputs are permitted and whether client consent or other steps are required. A generic prompt instruction cannot supply those protections.

Write a short assignment brief before opening the chat. Record the recipient, the decision or question, the deliverable, the deadline, the relevant sources, and the constraints. Name the person who will review the final work.

For a contract-risk summary, for example, identify the agreement and version, the provisions to assess, the business objective, and any related definitions or exceptions. If the packet is incomplete, record the gap instead of asking the tool to fill it with an assumption.

Divide a large assignment into useful stages: extract relevant passages, identify issues, propose an outline, draft, and review. Carry necessary context into each stage. A shorter prompt is not better if it leaves out the governing definition, audience, or purpose.

When a new source or client instruction arrives, update the assignment brief and identify which earlier conclusions need to be checked again.

A practical prompt gives the tool enough information to do the assigned work and gives you a way to evaluate the response. Include the task, audience, sources, constraints, and output format. Use an approved example when it helps explain the standard you want.

Example prompt to adapt:

Draft a 200-word update for the business lead using the approved source packet below. Explain the decision needed, the supported recommendation, and the unresolved questions. Preserve dates, conditions, and the distinction between confirmed facts and assumptions. After the draft, list the source location for each material claim and any information you could not find. Do not invent missing facts or authorities. [Insert the permitted source packet and matter-specific instructions.]

Check the source list yourself. A confident explanation of why a draft is correct is not independent verification. Ask for observable work products, such as a claim-and-source table or a comparison of two versions, and assess those against the underlying material.

Prompt Pilot helps you develop a prompt for your legal task. The Prompt Library provides starting points to adapt. Use either within your organization’s rules for tools and inputs.

Replace the bracketed instructions with the context the task requires. These prompts are starting points for reviewable work, not guarantees of accuracy.

1. Organize a client update.
“Propose two outlines for an update to [recipient], who needs to decide [decision]. Use [approved sources]. Put the recommendation and its key qualification near the beginning. For each outline, identify any missing information that prevents a useful recommendation.”

2. Improve a sentence without changing its meaning.
“Offer two clearer versions of [passage]. Preserve the actor, timing, conditions, exceptions, and level of certainty. Explain any proposed change that could affect meaning. Do not turn an inference into a confirmed fact.”

3. Prepare a source-checking table.
“List the material factual and legal claims in [draft]. For each, identify the supporting passage and location in [source packet], or mark support as not found. Separate direct support from inference. Identify contrary material in the packet.”

4. Test the reader’s next step.
“Read [draft] from the perspective of [recipient] deciding [question]. Identify where the recommendation, deadline, owner, or unresolved issue is unclear. Point to the passage causing each problem. Suggest a revision without adding facts.”

Review the response against the sources and assignment. A source packet may itself be incomplete; finding support inside it does not establish that the legal analysis is complete or current.

Use separate review passes when one pass would hide important questions:

       
1. Confirm the assignment. Does the draft answer the actual question for this recipient? Has the instruction changed?
       
2. Check the claims. Compare material facts, quotations, citations, and legal propositions with appropriate sources. Confirm that authorities are relevant and current for the matter.
       
3. Inspect omissions and uncertainty. Look for missing exceptions, contrary evidence, unresolved facts, and qualifications the reader needs.
       
4. Test the recommendation. Does it follow from the analysis? Is the proposed action useful, feasible, and clear about what remains unknown?
       
5. Improve organization and style. Make the answer easy to find. Then check that edits preserve the substance you just verified.
   

For each material criticism, identify a source, an unmet reader need, or a reasoning gap. Decide what to keep, revise, verify, or remove. A real matter may require additional sources, specialist input, and several passes.

Try the sequence. In our short source-review exercise, compare two replies with three source notes, decide what to keep or change, then open the annotations and one possible revision. All facts and drafts are constructed; no AI account or client documents are needed.

Our 90-minute workshop Is This Ready to Send? Reviewing AI-Assisted Legal Work gives teams practice making these decisions. Explore the review workshop and other live AI training.

This is a constructed teaching example. The drafts illustrate review decisions; they are not attributed to a person or model and are not a benchmark of AI performance.

Source packet: The supplier expects to finish testing on Thursday. Procurement has not yet approved the revised price. The project lead says a Friday launch is possible only if testing and price approval are complete by noon Thursday.

Draft to review:

We are on track to launch Friday. The supplier will finish testing Thursday, and Procurement is finalizing the price.

The wording is smooth, but the substance has shifted. “Expects” became “will.” “Not yet approved” became “finalizing,” which suggests progress the source does not establish. The noon deadline and both conditions disappeared.

A version supported by this packet:

A Friday launch remains possible if testing and price approval are complete by noon Thursday. The supplier expects to finish testing Thursday, but Procurement has not yet approved the revised price; we still need confirmation that both conditions can be met by the deadline.

This version preserves the uncertainty and tells the reader what still needs confirmation. Before sending, check for later information and decide whether the recipient also needs an owner, a fallback plan, or a specific request.

The same review applies to a human draft. Awkward wording may contain an essential qualification; polished wording may remove it.

Prompt: The instructions and material you give the tool for a task.

Context: The information needed to interpret those instructions, including the reader, purpose, definitions, constraints, and sources.

Grounding: Connecting a response to identified source material. A source reference is useful only if the material actually supports the proposition.

Hallucination: Plausible output that is fabricated or unsupported. A fluent citation or quotation still needs checking.

Workflow: The sequence of inputs, intermediate outputs, checks, revisions, and handoff used to complete an assignment.

Review criteria: The standards used to decide whether the work is acceptable, such as source support, completeness, appropriate qualification, and usefulness to the reader.

A team benefits from a shared process: which assignments are suitable, what information may be used, which checks are required, and who accepts the final work. Practice that process on a manageable example before relying on it in a live matter.

Live AI workshops for law firms and legal departments

• Is This Ready to Send? Practice reviewing AI-assisted legal work and leave with an annotated draft and review guide. 90 minutes.
• From Assignment to Verified Deliverable. Work through a matter from an unclear request to a checked deliverable, including a change in instructions. 90 minutes.
• Build Your Team’s AI Writing Playbook. Develop and test a workflow for recurring team assignments. 90 minutes, with an optional follow-up clinic.

Explore live AI training for your team

Training can also be combined with courses and writing tools. Discuss an AI writing playbook for your team.

Build your own skills. Explore our AI and legal writing courses, use Prompt Pilot to develop prompts, and get writing feedback with Diagnostics. Compare individual plans.

Included for all Write.law users: the AI-Enabled Lawyer. This resource brings together more than a decade of Write.law’s proprietary work with leading legal teams around the world, with guidance for using AI more effectively across legal tasks. Explore the AI-Enabled Lawyer.