← Learn

AI contract review: check every contract before you sign

AI contract review runs in a Claude or ChatGPT project that holds your playbook, a one-page list of terms you accept and refuse. Drop in a contract and the agent returns a summary, a key-terms table with exact quotes, red and yellow flags, suggested wording, questions for a lawyer, and every deadline. You check the quotes and sign yourself.

A founder in sunglasses and a dark suit reclining on a sun lounger by the sea, white Santorini houses and a dome behind him, laptop with a glowing orange screen beside him, drawn in black ink with orange accents

AI contract review lets you read every contract properly before you sign it, even when no lawyer is on hand. ZeroCrew built this use case around one AI assistant that holds your own rules for contracts, reads each agreement you drop in, and returns a short review: what you're agreeing to, which clauses break your rules, and what to ask the other side to change.

The review takes minutes, and ZeroCrew's prompt tells the agent to quote the exact words of every clause it flags, so you can check each one in the contract yourself. The agent never signs, sends, or replies for you. Every step below names at least two tools from ZeroCrew's directory, so you can build the review on the assistant and signing tool you already use.

What does it cost to sign a contract you haven't read?

Signing an unread contract costs you the terms you never saw: an auto-renewal you can't cancel, a liability you can't afford, or work you don't own. ZeroCrew sees founders sign client agreements, supplier contracts, and NDAs every month, usually alone and usually in a hurry, with no legal team to catch the one clause that matters.

Full legal review is priced for large companies, not solo founders. The researchers behind the CUAD contract dataset (Hendrycks et al., 2021) write that lawyers at large US firms typically bill around $500 to $900 an hour, and that small companies and individuals often sign contracts without reading them. ZeroCrew's use case gives a solo founder a first reading of every contract, at the price of an AI assistant plan.

Why is reviewing a contract by hand so hard?

Reviewing a contract by hand is hard because the clauses that matter hide in long, dense text, and you have to know what to look for. ZeroCrew counts a dozen terms a founder checks in almost every agreement: length, renewal, notice, fees, payment, liability, indemnity, ownership of the work, confidentiality, exclusivity, termination, and governing law.

Reading contracts well takes real training, even for people studying law. To build the CUAD dataset (Hendrycks et al., 2021), law students trained for 70 to 100 hours, then labeled 13,101 clauses across 510 contracts in 41 categories that lawyers pay attention to. ZeroCrew's prompt turns the most common of those categories into a checklist the agent works through on every contract, so you don't have to hold the list in your head.

Contract review takes time even for professionals. In the Better Call GPT study (Martin et al., 2024), senior lawyers averaged about 43 minutes per contract and junior lawyers about 56 minutes, while the language models finished in under five minutes. ZeroCrew reads those numbers as the case for an AI first pass: the agent does the slow reading, and you spend your time on the flags.

How does AI contract review work?

AI contract review works in one chat: you drop in the contract, and an assistant compares it with your playbook, a one-page list of the terms you accept and the terms you refuse. ZeroCrew's version runs as a Claude project, where the playbook lives in the project's knowledge and the prompt lives in the project's instructions.

Each contract review follows the same six parts, which ZeroCrew wrote into the prompt on this page, so every agreement, from a one-page NDA to a long supplier contract, gets the same reading in the same order and you always know where to look for the flags:

  1. A plain summary: who does what, for how long, and for how much.
  2. A table of key terms, each with the exact words and the section number.
  3. Flags marked red, yellow, or green against your playbook.
  4. Polite replacement wording for every red and yellow flag.
  5. Questions to take to a lawyer.
  6. Every deadline and notice window, ready for your calendar.

The review asks for quotes because an AI model can miss where an issue sits even when it spots the issue. The Better Call GPT study (Martin et al., 2024) found that the best models matched or beat junior lawyers at spotting legal issues, while legal process outsourcers stayed slightly ahead at pinpointing where each issue sat. ZeroCrew's prompt makes the agent quote the clause and its section number, so a wrong pointer shows up the moment you check it.

AI models can also state legal facts that aren't true. A study of legal hallucinations (Dahl et al., 2024) found that popular models hallucinated in 58% to 88% of answers to specific questions about real federal court cases. ZeroCrew keeps this use case to the text in front of the agent: the prompt forbids quoting words that aren't in the contract and sends questions about the law to a lawyer.

Which tools do you need for AI contract review?

You need three tools for AI contract review: an assistant that reads the contract, a signing tool, and a place to keep the dates. ZeroCrew lists two or more options for each step, and the table shows when to pick which, from a general assistant to a contract specialist that works inside Word.

StepOption AOption BWhen to pick which
Read and reviewClaude in a projectChatGPT in a projectClaude or ChatGPT for a low-cost first pass on any contract, with your playbook saved once.
Review inside WordSpellbookGenie AISpellbook or Genie AI if you redline in Microsoft Word and review contracts every week.
SignDocusignPandaDoc or SkalaDocusign when the other side expects it. PandaDoc when the contract is also a proposal or quote. Skala if it already runs your company paperwork.
Keep the datesNotionAirtableNotion if your notes already live there. Airtable if you want a contract register with views and automations.

The review runs in a general AI assistant, and ZeroCrew's AI assistants page ranks the options. Anthropic's help center says projects keep a knowledge base and project instructions together, and that free accounts can create up to five projects. ZeroCrew's ChatGPT page notes that ChatGPT also keeps chats and files together in projects. As of November 2026, Anthropic's pricing page lists Claude Pro at $20 a month billed monthly, or $17 a month billed yearly.

A contract specialist fits founders who review contracts every week, and ZeroCrew's contracts and signing page lists Spellbook among the tools for that job. ZeroCrew's Spellbook page describes AI redlines, playbooks that apply your standards to every review, a Microsoft Word integration, and a 7-day free trial, and founders' posts it summarizes praise redlining inside Word. ZeroCrew's Genie AI page describes an AI legal assistant that drafts and reviews contracts and connects with Microsoft Word.

The signing step happens after you agree on the changes, and ZeroCrew's contracts page ranks Docusign as the most searched tool in the group. As of November 2026, Docusign's pricing page lists the Personal plan at $11 a month billed yearly, with 5 envelopes a month. ZeroCrew's PandaDoc page lists a free e-signature plan and an official PandaDoc MCP server that can track signature status from Claude or ChatGPT.

E-signatures carry legal weight in the US and the UK, and ZeroCrew lists signing tools for that reason. For transactions in or affecting interstate or foreign commerce, the US ESIGN Act says a contract or signature may not be denied legal effect solely because it is electronic. The Law Commission of England and Wales concluded that an electronic signature can execute a document when the signer intends to authenticate it and any required formalities are met.

Copy the prompt

You review contracts for {company name} before I sign them. In each contract, I am the {my side: customer, supplier, contractor, or tenant}. Contracts I sign are governed by the law of {country or state}.

Read the contract I attach and my playbook in this project's knowledge. Then write a review with these six parts:

1. Summary: in five plain sentences, what the contract is, who must do what, for how long, and for how much.
2. Key terms: a table with term length, renewal and auto-renewal, notice to cancel, fees, payment terms, liability cap, indemnities, who owns the work and the IP, confidentiality, exclusivity or non-compete, termination rights, governing law, and anything I must do by a date. For each term, quote the exact words and give the section number. If a term is missing, write "not in the contract".
3. Flags: every clause that breaks my playbook or is unusual for this kind of contract. Mark each one red (don't sign as written), yellow (ask to change), or green (fine). Quote the clause, give its section number, and say in one sentence why it matters to me.
4. Suggested changes: for each red and yellow flag, polite replacement wording I can send to the other party.
5. Questions for a lawyer: anything that depends on the law of {country or state}, anything high-value or unusual, and anything you aren't sure about.
6. Dates: every deadline, renewal date, and notice window, as a list I can copy into my calendar.

Rules:
- Quote only words that appear in the contract. If you can't find a clause, say so instead of guessing.
- Check schedules, exhibits, and attachments before you write "not in the contract".
- Don't contact anyone, send anything, or sign anything.
- End the review with: "This is a first-pass review, not legal advice."

ZeroCrew wrote this prompt for a Claude project: paste it into the project's instructions, fill in the placeholders for your company, your side of the deal, and your country or state, and add your playbook to the project's knowledge. Then start a new chat in the project for each contract and attach the file.

How do you set up AI contract review in Claude?

Setting up AI contract review in Claude takes six steps: write a playbook, create a project, paste the prompt, attach each contract, check the quotes, then sign and log the dates. ZeroCrew wrote each step from Anthropic's help center article on projects, opened in November 2026. Anthropic says a new version of projects is rolling out in stages, so if a screen looks different from these steps, Anthropic's help center shows the current one.

  1. Write your playbook on one page. List the terms you accept and refuse: your payment terms, the longest commitment you'll sign, the notice you need to cancel, the liability cap you can live with, who owns the work you deliver, and any clause you never sign, such as exclusivity.
  2. In Claude, create a project named Contract review, and add your playbook to the project's knowledge. A ChatGPT project works the same way if ChatGPT is your assistant.
  3. Paste the prompt into the project's instructions, and replace each placeholder.
  4. For each new contract, start a chat in the project and attach the contract as a PDF or Word file.
  5. Open the contract next to the review, and find every quoted clause by its section number.
  6. Send the changes you want yourself, sign in Docusign or PandaDoc once you agree, and copy the dates into Notion or Airtable.

What should you check before you trust an AI contract review?

Trust an AI contract review after it passes a test on a contract you already know well. ZeroCrew suggests running the prompt on an agreement you've signed and read line by line, then comparing the agent's flags with your own notes, before you rely on it for a contract you haven't read.

On every review, ZeroCrew suggests four quick checks against the contract itself, because they catch the mistakes AI models are known to make: a wrong quote, a wrong pointer, a missed schedule, and a confident answer to a question that belongs with a lawyer:

  • Every quote appears in the contract, word for word, at the section the agent names.
  • Nothing in the contract contradicts the summary, especially the price and the length.
  • Every "not in the contract" term is truly missing, and not hidden in a schedule or attachment.
  • Anything large or unusual goes to a lawyer, with the agent's questions as your list.

An AI contract review is a first pass, not legal advice, and ZeroCrew's prompt ends every review by saying so. A lawyer still matters for high-value deals, investment documents, employment terms, and anything governed by an unfamiliar country's law. The agent's question list makes that call shorter, because the lawyer starts from the clauses that matter to you.

How does ZeroCrew help you pick your contract tools?

ZeroCrew helps you pick contract tools by ranking the options for every step, so you sign each contract knowing what's in it and with tools that fit how you work. ZeroCrew's contracts and signing page ranks Docusign, PandaDoc, and Spellbook two ways, as of November 2026: by how many people search for each, and by what founders say about the price.

Every step in this use case has a ZeroCrew page that compares its options. The legal and company page brings contracts, company formation, and privacy compliance together, the AI assistants page ranks Claude, ChatGPT, and the other assistants, and ZeroCrew's MCP servers list shows which tools your agent can connect to directly, such as Notion and Airtable.

Your agent can search ZeroCrew itself. ZeroCrew's public MCP server at https://zerocrew.ai/api/mcp/public has tools to search the directory, read a tool's page, list roles, and list the tools for a job. Add it to Claude or ChatGPT as a custom connector, and the agent can suggest a signing tool or a contract specialist that fits how you work.

If you'd rather have it built for you, ZeroCrew's contracts page offers a company admin agent that keeps your filings, renewals, and contracts on the calendar, set up through ZeroCrew's done-for-you AI automation. ZeroCrew projects start at $1,000 as of November 2026, and you approve what goes out until you trust it.

Sources (5)
  1. Hendrycks et al., CUAD: An Expert-Annotated NLP Dataset for Legal Contract Review (arXiv)
  2. Martin et al., Better Call GPT, Comparing Large Language Models Against Lawyers (arXiv, 2024)
  3. Dahl et al., Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models (arXiv, 2024)
  4. 15 U.S. Code § 7001, General rule of validity (ESIGN Act), Cornell Law School LII
  5. Law Commission of England and Wales: Electronic execution of documents

Zero employees. Never build alone.

Find the agent for every job