The Best ChatGPT and Claude Prompts for Lawyers (2026), with the Verification Step Each One Needs

I published my first collection of legal prompts in March 2023, a few months after ChatGPT appeared. The tools have changed enormously since then. The principles of a good legal prompt have not. This article gives you prompts you can copy today for the tasks lawyers ask about most, and pairs each one with the verification step it needs. That pairing is the point: a prompt without a check is how lawyers end up in sanctions judgments.

Before you prompt: three ground rules

1. Use an approved tool and account. Consumer accounts may use your inputs to improve the service unless you switch that off. Use the enterprise or team version your firm has approved, check that training on your data is disabled, and follow your firm’s policy on client information. ABA Formal Opinion 512 makes the point directly: lawyers must understand how a tool handles information before putting client data into it.

2. Anonymise when in doubt. Replace names, addresses, account numbers and other identifiers with placeholders such as [Client], [Counterparty] or [Property]. The model does not need them to analyse structure, risk or drafting.

3. You verify, every time. The tool drafts. You remain responsible for every fact, citation and conclusion that leaves your office.

The anatomy of a legal prompt

The prompts that consistently produce usable work share five parts:

• Role: who the model should act as (“You are a senior commercial lawyer in England and Wales”).
• Context: the facts, the client’s position and the goal.
• Jurisdiction: the governing law, because a clause that works in Delaware may fail in Dubai.
• Output format: a table, a redline, a numbered list, a letter of a set length.
• Limits: what not to do, and an instruction to flag uncertainty rather than guess.

The prompts below follow that structure. Replace everything in square brackets.

Contract review prompts

First-pass risk review

You are a senior commercial lawyer acting for [the customer] under [governing law].
Review the contract below. Produce a table with columns: Clause | Issue | Risk to
my client (High/Medium/Low) | Suggested fix. Focus on liability caps, indemnities,
termination, IP ownership, data protection, payment and governing law. Do not
invent clauses that are not in the text. If something is ambiguous, say so.

[Paste anonymised contract]

Verify: read every flagged clause yourself, then read the clauses the table did not mention. Models skip issues as often as they invent them.

Missing protections check

Acting for [the supplier], list the standard protections that are ABSENT from
this agreement, compared with a balanced [SaaS / services / supply] contract
under [governing law]. For each, explain the risk in one sentence and propose
neutral wording.

Verify: confirm each “missing” clause is genuinely absent (search the document), and adapt proposed wording to your client’s commercial position. The free Contract Clause Checker runs a similar check in your browser without sending the text anywhere.

Negotiation position paper

Here are the other side's proposed amendments. For each, give: what it changes,
why they want it, how much it hurts my client, a fallback position, and a
walk-away point. Keep the tone suitable to share internally with the client.

Verify: the commercial judgement is yours. Check each characterisation against the marked-up text.

Drafting prompts

Clause drafting with options

Draft a [limitation of liability] clause for a [B2B SaaS agreement] governed by
[law], acting for [the provider]. Give three versions: provider-friendly,
balanced and customer-friendly. Use defined terms consistently and explain in
two lines when each version is appropriate.

Verify: check enforceability under the governing law (for example, statutory limits on excluding liability), and run the final text through the Defined Terms Checker to catch terms that are used but never defined.

Plain-English rewrite

Rewrite the clause below in plain English for a business reader without changing
its legal effect. Keep defined terms. Then list any point where simplification
could change the meaning, so I can decide.

[Clause]

Verify: compare old and new line by line for any change in obligation, condition or exception. The Plain-English Legal Scanner scores readability before and after.

Research planning prompts

AI is excellent at planning research and poor at being the research. Use it to structure your work, not to supply authorities.

I need to research [issue] under [jurisdiction] for [type of client]. Produce a
research plan: the legal questions in order, the likely sources of law (statute,
regulation, case law, guidance), search terms for a legal database, and the
counter-arguments I should anticipate. Do not cite specific cases.

Verify: all authority comes from your own research in primary sources. If you later ask a tool to summarise a judgment, paste the judgment in and check the summary against it.

Client communication prompts

Advice letter from your notes

Turn my notes below into a clear advice letter to a [non-lawyer founder]. Order:
the answer, the reasons, the risks, the options, next steps. Under [600] words.
Plain English. Do not add any legal point that is not in my notes.

[Your notes]

Verify: the instruction not to add law is essential. Read for any sentence that goes beyond your notes and remove it.

Explaining a document to a client

Explain this [lease / shareholders' agreement] to my client in a one-page summary:
what they must do, what they must not do, key dates, costs, and the three clauses
most likely to cause problems. Use headings and short bullets.

Verify: check every date, amount and obligation against the document. Summaries are where silent errors hide.

Litigation preparation prompts

Chronology from documents

From the documents below, build a chronology table: Date | Event | Source document
and page. Only include events that appear in the documents. Mark any date you have
inferred rather than read directly.

Verify: spot-check entries against the source pages, especially inferred dates.

Stress-testing your argument

Here is my argument on [issue]. Act as opposing counsel and as a sceptical judge.
Give me the five strongest objections, ranked by likely impact, and the factual
gaps I need to fill.

Verify: treat the objections as prompts for your own analysis. Any authority the model suggests must be confirmed before use.

Prompts that should never be used as written

Some requests invite exactly the failures courts sanction:

• “Find me cases that support [proposition].” This invites invented authorities.
• “Write my brief on [issue].” Wholesale drafting of advocacy without your analysis produces confident nonsense.
• “Is this contract enforceable?” Without jurisdiction, facts and the full document, any answer is a guess.

Reframe each as a planning, structuring or critique task, and keep the legal conclusions for yourself.

Choosing between ChatGPT and Claude

Both tools handle the prompts above well. In my experience Claude is particularly strong with long documents and measured tone, which suits contract work and client letters, while ChatGPT offers a wide range of tools and integrations. The decision that matters more is governance: which account your firm has approved, what the data terms say, and whether your team has been trained to verify. A capable model used carelessly is more dangerous than a modest one used well.

A final check before anything leaves your desk

Before you send or file AI-assisted work, ask four questions. Have I verified every authority and quotation in a primary source? Have I checked every fact, date and figure against the documents? Did any confidential information go into an unapproved tool? Would I be comfortable explaining to a judge or client exactly how this was produced? If the answer to all four is yes, the tool has done its job and so have you.