Why most “ChatGPT for contract review” prompts produce a mess
Most contract review prompts online are written by people who do not redline contracts for a living. They give the model a one-line instruction like “review this NDA and flag issues” and act surprised when the output is a generic risk list that ignores the playbook, the deal, and the client. Contract review is a comparison against a standard, a judgment about commercial risk, and a set of redlines you would actually send to the other side.
This article is the contract-review slice of a wider prompt library. For a broader catch-all set across intake, research, and drafting, see ChatGPT prompts for lawyers. For briefs, motions, and demand letters, see ChatGPT prompts for legal writing. The 12 prompts below cover first-pass triage, risk and issue spotting, playbook redlining, fallback drafting, and clause comparison. Each one is built to be pasted into ChatGPT, Claude, or Gemini and edited for your matter, with bracketed placeholders for the parts you swap in. Read the redaction checklist near the end before you paste anything that touches a real client.
How to use these prompts
A useful contract-review prompt follows the same five-part recipe every time: tell the model what role it is playing, name the jurisdiction and governing law, define the task, give it context it cannot guess (your playbook, the client’s risk appetite, the deal type), and specify the output format you want. Skip any of those and you get a confident but generic answer that you have to throw out.
Three habits make the prompts below work. First, paste the actual contract or the relevant clauses, not a summary. The model cannot redline a document it has not seen. Second, paste your playbook or your standard form alongside the counterparty draft, so the model has something to compare against. Without a standard, “risk” becomes whatever the model thinks risk usually means. Third, treat the output as a triage layer, not a final redline. The model surfaces candidates; you decide which ones matter for this deal and this client.
For a survey of purpose-built tools that automate this workflow inside a contract platform, see best AI contract review software. The prompts below give you most of that value with a general-purpose chatbot; the dedicated tools add clause libraries, version tracking, and Word integration that matter once review volume scales.
First-pass triage and summarization
The first job is figuring out what you are looking at. These prompts give you a fast read of the deal before you start redlining.
1. One-page contract summary
You are a transactional attorney summarizing a contract for a partner in a hurry. Contract type: [E.G. MUTUAL NDA, MSA, SAAS ORDER FORM, DPA] Governing law: [STATE OR COUNTRY] My client's role: [E.G. CUSTOMER, VENDOR, LICENSOR] Paste of the contract (no identifying information): "[PASTE CLEAN COPY]" Produce a one-page summary in this order: 1. Deal in one sentence. 2. Term and termination (length, renewal, exit rights). 3. Payment and pricing (amounts, schedule, late fees, true-up). 4. Scope of services or license. 5. Key risk allocation points (indemnity, limitation of liability, IP ownership). 6. Anything unusual a reader should flag. Use plain English. Cite the specific section number for each item. Do not invent provisions that are not in the text.
2. Obligation and deadline extraction
Extract every operative obligation and deadline from this contract. For each item, give me: - Section number - Who owes the obligation - What is owed - When it is due (specific date, or "X days after Y trigger") - Consequence of missing it, if stated Format as a table with columns: Section, Party, Obligation, Deadline, Consequence. Do not paraphrase the obligation in a way that changes its meaning. If a deadline is ambiguous, write "ambiguous" and quote the language. Contract: "[PASTE CONTRACT]"
3. Defined terms audit
Audit the defined terms in this contract. Produce two lists. List A: every defined term, in alphabetical order, with the section where it is defined and a one-sentence plain-English explanation. List B: any defined term that is used in the body of the contract but never actually defined, or defined inconsistently. For each one, quote the language that creates the problem. Do not flag a term as undefined if it is defined later in the document. Read the full text before answering. Contract: "[PASTE CONTRACT]"
Risk and issue spotting
The next pass is finding what to push back on. These prompts surface candidates; you decide which ones to take to the other side.
4. Issue spotter against a standard checklist
You are a senior transactional attorney reviewing a [CONTRACT TYPE] on behalf of [CLIENT ROLE, E.G. THE BUYER] in [JURISDICTION]. Read the contract below and flag every clause that creates a meaningful risk for my client. For each issue, give me: 1. Section number and a short quote of the offending language. 2. Why it is a problem for my side. 3. How serious it is (low, medium, high) and the basis for that rating. 4. What you would propose to change. Focus on these standard risk categories: limitation of liability, indemnification scope, IP ownership and license-back, confidentiality term and survival, termination rights, auto-renewal, payment and audit rights, warranties and disclaimers, dispute resolution and venue, assignment. Do not invent risks. If the contract is silent on a category, say so and note whether that silence helps or hurts my client. Contract: "[PASTE CONTRACT]"
5. Asymmetry check
Compare the contract below for asymmetric terms that favor one party over the other. I represent [CLIENT ROLE]. For each asymmetric provision, list: - Section number - The asymmetric language (quoted) - How it favors the other side - A symmetric or balanced redraft Look closely at: indemnity (one-way vs. mutual), termination rights, cure periods, cap on liability, exclusions from the cap, audit rights, IP ownership, confidentiality survival, and dispute resolution. Be specific. "The indemnity is one-sided" is not enough. Tell me which party owes what to whom and why that allocation is unbalanced for this deal. Contract: "[PASTE CONTRACT]"
6. Limitation of liability stress test
Analyze the limitation of liability and indemnification provisions in this contract. I represent [CLIENT ROLE]. Walk through three scenarios and tell me, based only on the contract text, what each party owes the other in each case. If the contract is silent or ambiguous on a point, say so. Scenario A: The other side breaches a confidentiality obligation and my client's trade secret is exposed. Scenario B: A third party sues my client for IP infringement based on the deliverable the other side provided. Scenario C: My client misses a payment deadline by 45 days. For each scenario, identify: 1. Which provisions apply. 2. What is recoverable and what is excluded. 3. Whether the cap, the carve-outs, or the indemnity controls. 4. The dollar exposure if you can estimate it from the contract. Do not pull in outside law unless the contract incorporates it by reference. Contract: "[PASTE CONTRACT]"
Playbook redlining and counter-drafting
This is where the leverage shows up. When you give the model your playbook, it stops guessing and starts working from your standard.
7. Redline against a clause playbook
You have two documents. Document 1 is my firm's playbook for [CONTRACT TYPE]. For each topic, the playbook gives a preferred position, an acceptable fallback, and an unacceptable position. Document 2 is the counterparty's draft. Compare the draft to the playbook clause by clause. For each clause that does not match a preferred position, tell me: 1. Section number and current language (quoted). 2. Which playbook category it falls into (preferred / fallback / unacceptable / silent). 3. Proposed redline language that moves the clause to the closest acceptable playbook position. 4. A one-line negotiation note explaining why we are asking for the change. Where the draft already matches the playbook, leave it alone. Do not redline for style. Document 1 (playbook): "[PASTE PLAYBOOK]" Document 2 (counterparty draft): "[PASTE DRAFT]"
8. Counter-redline to opposing markup
The counterparty returned the attached draft with their redlines accepted (or rejected and replaced with their language). I am preparing the counter-redline on behalf of [CLIENT ROLE]. For each change the other side made, do the following: 1. Quote the change. 2. Classify it: acceptable, push back, reject. 3. If push back: propose the smallest edit that preserves my side's position. 4. If reject: explain why and propose a fallback we could live with. Focus on substantive changes. Ignore typo fixes and pure style edits. Contract (with opposing redlines visible in [BRACKETS]): "[PASTE MARKED-UP DRAFT]"
9. Fallback language menu for a single clause
The counterparty rejected the language I proposed for [CLAUSE, E.G. CAP ON LIABILITY]. They are pushing for [SHORT DESCRIPTION OF THEIR POSITION]. Draft three fallback positions for my side, in order from strongest to weakest. For each one: 1. Full clause text my associate can paste into the document. 2. The concession we are making. 3. The concession we are asking for in return (the "give" we want). 4. A one-line internal note on why this fallback might or might not fly with this counterparty type. Assume my client's bottom line is [CLIENT'S BOTTOM LINE, E.G. CAP CANNOT GO ABOVE 2X FEES PAID]. Governing law: [STATE].
Clause comparison and extraction
The last group is the high-volume work: turning a stack of contracts into structured information you can use.
10. Clause-by-clause comparison of two drafts
I have two versions of the same contract. Version A is my client's preferred draft. Version B is the counterparty's redline. Compare them clause by clause. For each clause where the text differs, produce a row in a table with these columns: Section, Topic, Version A language (quoted), Version B language (quoted), Substantive impact (who benefits), Recommended action. Order the rows from most material to least material based on commercial impact, not order in the document. Version A: "[PASTE VERSION A]" Version B: "[PASTE VERSION B]"
11. Extract specific terms across multiple contracts
You will receive several contracts in sequence. For each one, extract the following fields and return them as a single row in a CSV-style block. Fields (in this order): 1. Contract title or short name 2. Counterparty name [LEAVE BLANK IF REDACTED] 3. Effective date 4. Term length 5. Renewal type (auto, mutual, none) 6. Notice period to terminate 7. Annual or total contract value 8. Governing law 9. Liability cap (state amount or "uncapped") 10. Indemnity scope (one-way to my client, one-way to counterparty, mutual) 11. Anything unusual If a field is not in the contract, write "not stated." Do not infer values. After all contracts are processed, output a final consolidated table. I will paste the first contract in the next message.
12. Plain-English explainer for a non-lawyer stakeholder
My client's [BUSINESS STAKEHOLDER, E.G. HEAD OF PROCUREMENT] needs to understand this contract before they sign. Rewrite the key terms in plain English a non-lawyer can act on. Aim for an eighth-grade reading level. Cover, in this order: 1. What we are agreeing to in one paragraph. 2. What it costs us, including any variable charges. 3. What we have to do, and by when. 4. What happens if we want out. 5. What happens if something goes wrong (liability, indemnity, dispute resolution) in plain words. 6. Three things they should ask before they sign. Keep legal terms only where there is no plain substitute, and define each one in a short parenthetical the first time it appears. No legal advice phrasing; just clear explanation. Contract: "[PASTE CONTRACT]"
How to adapt these prompts to a real deal
The single biggest unlock for AI contract review is feeding the model your firm’s or your client’s playbook. If you do not have one written down, build a short version for the contract types you see most often. Three columns is enough: topic, preferred position, acceptable fallback. Paste that into the prompt and the output stops being generic risk theater and starts being a redline you can use.
The second adjustment is voice. Paste a paragraph of a past redline you sent and add a line: “Match this voice and level of formality in any drafted language.” The model picks up your sentence length, your tolerance for plain English, and your typical concession phrasing. It is the fastest way to stop AI-assisted redlines from reading like an LLM.
Build a small personal prompt library over time. Save your three or four most-used prompts as text snippets in your contract management system or a notes app, and edit them as your deal mix changes.
Before you paste anything: a redaction checklist
This is the part most prompt articles bury or skip. Free-tier ChatGPT, Claude, and Gemini may use your inputs to improve their models. Even where they do not, you are sending a counterparty’s draft outside your firm’s control. The free tiers are appropriate for hypothetical fact patterns, your own playbook content, and your own past redlines. They are not appropriate for an actual client’s draft contract without redaction.
Before you paste any contract or markup into a public AI tool, strip:
- Party and signatory names on both sides, including affiliates.
- Specific dollar amounts that, combined with public filings, could identify the deal.
- Account numbers, tax IDs, and any number tied to a specific person.
- Trade secrets, technical specs, or product roadmaps in schedules or exhibits.
- Anything subject to a protective order, a third-party NDA, or a sealed filing.
If the contract cannot be meaningfully anonymized, use a tier where the vendor contractually agrees not to train on your inputs. Examples include ChatGPT Enterprise and Claude for Work. Microsoft Copilot for Business sits inside the Microsoft 365 tenancy and offers similar protection. Purpose-built tools like Spellbook and CoCounsel are designed for contract work and run inside the editor itself. Run anything truly sensitive on a local model, or do not run it through AI at all.
The ABA addressed lawyer use of generative AI in Formal Opinion 512 (July 2024), which folds AI use into existing duties of competence (Rule 1.1) and confidentiality (Rule 1.6). Several state bars have followed with their own guidance. Mata v. Avianca is the cautionary tale on what happens when a lawyer skips citation verification, and the same lesson applies to any “the model says this clause is enforceable” output you take at face value. For the broader picture on AI in a small or solo practice, see our pillar guide to AI for law firms.
Related on Business AI Workflows
- ChatGPT prompts for lawyers covers a wider library across intake, research, drafting, and client communication.
- ChatGPT prompts for legal writing goes deep on briefs, motions, demand letters, and client memos.
- Best AI contract review software surveys the purpose-built tools that automate the workflow these prompts approximate.
None of this is legal advice. Verify every output, follow your state bar’s guidance, and check your client’s own AI policy before you paste their contract into a public model.
