Question · At work
Can AI review contracts?
AI can do a useful first pass on a contract, but it does not replace legal advice on anything that matters.
Short answer
Yes, AI can do a useful first-pass review of a contract: summarizing it, flagging unusual or one-sided clauses, and comparing terms against your standard positions. It can miss issues, misread context and state wrong law confidently, so it does not replace a lawyer for significant agreements. Use it to prepare and speed up review, not to approve.
Contract review is one of the most time-consuming office tasks, and one where AI helps most and errors matter most. The sensible middle path is to use it as a fast first reader.
What AI does well
- Plain-English summary: who owes what to whom, by when, and for how much.
- Clause extraction: term, renewal, termination, payment terms, liability caps, governing law.
- Flagging unusual terms: one-sided indemnities, auto-renewals, broad IP assignments.
- Comparison to your standards: you provide your standard positions, and it lists where the document deviates.
- Drafting suggestions: proposed alternative wording you can hand to counsel.
In our class, the day-two session on OpenClaw and AI for contracts walks through real contract review, NDA analysis and compliance checks; the aim is the same: shorter first-pass time.
Where it fails
- Missing context: it does not know your negotiating leverage, your other contracts, or what the business actually needs.
- Jurisdiction: it may state legal rules confidently that are wrong or outdated for your location.
- Cross-references: long contracts with definitions and exhibits are easy for a model to misread.
- Omissions: it may not flag what is missing altogether, like a required clause.
- Fabrication: see how to stop AI hallucinations.
What the research says about the failures
- An academic study of legal research tools found that Lexis+ AI from LexisNexis and the Westlaw research and Ask Practical Law AI tools from Thomson Reuters each hallucinate between 17% and 33% of the time. That study tested legal research questions, not contract review, but it shows that tools sold for law still need checking.
- The Lost in the Middle paper found that language model performance can degrade significantly when the position of relevant information changes in a long input. That is one reason a clause buried in the middle of a long contract can be missed.
- Anthropic guidance for long documents is to ask Claude to extract word-for-word quotes first before performing its task. That matches the clause-by-clause quoting in the workflow below.
A safer workflow
- Get permission to use AI on the document. Check confidentiality terms and company policy; see whether it is safe to put company data in ChatGPT.
- Provide your checklist, e.g., “we require mutual confidentiality, a 2-year term, and Massachusetts law.”
- Ask for a table of clauses, what the contract says, and how it differs from your checklist, with clause numbers.
- Ask for questions, not conclusions: “What should I ask the other side about?”
- Verify each flagged clause in the original text.
- Escalate to counsel for anything material.
Example prompt
You are helping me review a vendor agreement. My standards are below. For each standard, quote the relevant clause with its number, say whether the contract meets it, and rate the deviation as minor, moderate or major. If the contract is silent, say “silent.” Do not give legal advice; list questions for counsel.
The professional question
If you are a lawyer or work under professional-conduct rules, additional duties apply. The ABA ethics opinion on generative AI says lawyers using generative AI must consider duties to provide competent legal representation, to protect client information, to communicate with clients and to charge reasonable fees. On confidentiality, a lawyer must keep confidential all information relating to the representation of a client, regardless of its source, unless the client gives informed consent. Check with your regulator or firm for the rules that apply to you. For related concerns see using AI for client work.
Bottom line
Use AI to read faster and prepare better questions. Use humans for decisions that carry legal risk.
Sources
- Magesh et al., Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools (arXiv 2405.20362): “Westlaw AI-Assisted Research and Ask Practical Law AI) each hallucinate between 17% and 33% of the time”
- Liu et al., Lost in the Middle: How Language Models Use Long Contexts (arXiv 2307.03172): “performance can degrade significantly when changing the position of relevant information”
- Anthropic, Reduce hallucinations (Claude Platform Docs): “ask Claude to extract word-for-word quotes first”
- American Bar Association, ABA issues first ethics guidance on a lawyer’s use of AI tools: “duties to provide competent legal representation, to protect client information, to communicate with clients and to charge reasonable fees”
- American Bar Association, ABA issues first ethics guidance on a lawyer’s use of AI tools: “keep confidential all information relating to the representation of a client, regardless of its source, unless the client gives informed consent”
Frequently asked
Can I use AI to review an NDA?
It is a common first use: extracting the term, scope, exclusions and governing law. Have counsel review anything you will sign that matters.
Is it safe to upload contracts to AI?
Only if your company and the counterparty allow it and the tool's data terms fit. Confidentiality clauses may restrict sharing.
Applied AI Workflow Automation · Boston
Build this with us, in person
Two days, hands-on, 20 seats. You practice AI workflow automation on provided case files for email, documents and spreadsheets. Next class: early 2027, dates to be announced · $1,299 tuition.