AI Workflows · Workflow playbook · Updated June 2026
AI Contract Review for Lawyers: The Five Pass Method
A vendor SaaS promises to redline your contract for you. The senior move is the opposite: keep the judgment, and use AI as a fast, structured first reader you check. Here is the exact five pass method, what to look for in each pass, and where it must never go.
Key takeaways
- Method beats product. A repeatable five pass prompt method on a general model you control gives a senior lawyer more reach and less lock in than a black box contract review SaaS that hides how it reached its conclusions.
- Run it in passes, not one prompt. Orient, map structure, pressure test key terms, find what is missing, then draft edits. Each pass has a different job and a different prompt, and the missing terms pass is the one AI is quietly best at.
- Verify every flag against the text. AI points you at a clause; you open the clause and confirm it. An unverified AI observation is a hypothesis, not a finding, and acting on it unread is malpractice waiting to happen.
- Confidentiality is the hard line. Privileged and client confidential agreements do not belong in a public consumer model. Use an approved enterprise deployment with no training on your inputs, or redact, or do not use AI on that document at all.
The professional's real problem
Contract review is where a lot of legal hours quietly disappear. A forty page services agreement, a master agreement with three amendments, a stack of nearly identical vendor contracts that each need a careful pass: the work is essential, the client is rarely thrilled to pay full freight for it, and the part that actually requires your expertise is buried inside a lot of reading that does not. You are paid for the judgment calls on indemnification, liability caps, termination rights, and the clause that is conspicuously absent. You are not paid, in any satisfying sense, for the two hours of orienting yourself to the document before you can even make those calls.
That is the gap AI fills, and it is also where the market wants to sell you the wrong thing. A wave of contract review products promise to ingest an agreement and hand back a finished redline, as if the review were the deliverable rather than your reasoning about the deal. For a senior lawyer that framing is backwards and a little dangerous. The redline is not the hard part. Deciding which risks matter for this client, in this deal, against this counterparty is the hard part, and no tool owns that. The lawyers getting real leverage from AI are not outsourcing the review. They are using AI to do the reading and the first structured markup at speed, then spending their time where it counts: on the calls only they can make.
The question is not whether AI can review your contract. It is whether you will let a black box make the judgment calls, or use AI as the fast first reader while you keep the judgment.
The practitioner method vs the vendor SaaS
Before the steps, it is worth being clear about why a method on a model you control beats most one click contract tools for serious work. This is not anti software. It is about where the leverage and the risk actually sit.
| Dimension | Five pass method (you drive a model) | One click contract review SaaS |
|---|---|---|
| Who holds the judgment | You do. The tool reads and flags; you decide what matters for this client and this deal. | The product decides what to surface and how to redline, often against a generic playbook you cannot fully see. |
| Transparency | You see the prompt, the reasoning, and the clause it points to, and you can push back in plain language. | Frequently a black box. You get an output, not the reasoning, which makes it hard to certify. |
| Flexibility | Works on any contract type and any client posture. You adjust the prompt to the deal in seconds. | Strong on the contract types it was built and tuned for; weaker or rigid outside them. |
| Data handling | You choose the deployment. An approved enterprise plan can keep inputs out of training; you control what goes in. | Varies widely. Some are built for confidential legal data; many consumer tools are not. Read the terms. |
| Best use | Senior review where judgment, client specifics, and defensibility matter most. | High volume, low variation contracts against a fixed standard, where a tuned playbook genuinely helps. |
Read that as a routing guide, not a verdict. For a large book of near identical low risk agreements measured against a fixed standard, a purpose built tool can be the right call. For the review that carries your name and your judgment, the method below keeps you in control of both.
What to check, pass by pass
The whole method rests on five passes, each with a single job. Run them in order. The point of separating them is that a contract reviewed in one undifferentiated pass gets a shallow read of everything; five focused passes get a deep read of the things that decide the deal.
| Pass | The job | What you are looking for |
|---|---|---|
| 1. Orient | Understand the deal in plain terms before reading clauses. | Parties, what is being exchanged, term, governing law, your client's side, and the two or three things that actually matter commercially. |
| 2. Map structure | Build a clause by clause map of the whole agreement. | Every operative section, defined terms, cross references, exhibits and schedules, and anything labeled but never defined. |
| 3. Pressure test key terms | Read the risk and money clauses hard. | Indemnification, limitation of liability and caps, termination and cure, IP ownership and license scope, payment, warranties, and dispute resolution. |
| 4. Find what is missing | Surface the clauses that should be there and are not. | Absent liability cap, no termination for convenience, missing confidentiality or data terms, one sided indemnity with no mutual provision, no assignment clause. |
| 5. Draft suggested edits | Turn findings into proposed redline language in your client's favor. | Specific clause rewrites, fallback positions, and a short rationale you can paste into a negotiation note, all subject to your edit. |
How to run the review, step by step
Here is the method as a workflow you can run today on an approved AI deployment. Each step pairs a focused prompt with the verification you do alongside it. The prompts are starting points; tighten them to your client and your contract type.
Step 1: Orient yourself to the deal
Before a single clause, get the shape of the agreement in plain language. This gives you the lens for every later pass and catches gross issues immediately.
Example prompt: "You are assisting a lawyer who represents the [buyer / service recipient]. Here is a contract. In plain language, summarize the deal: who the parties are, what is being exchanged, the term and renewal, the governing law and venue, and the three commercial points that most affect my client. Do not redline yet. Flag anything that looks unusual for this type of agreement."
Verify as you go: read the summary against your own first skim. If the model misstates a party or a term, you have just learned how carefully you need to check the rest.
Step 2: Map the structure
Ask for a clause by clause map so nothing hides. This is where AI's patience genuinely helps, because it will list every section without skimming.
Example prompt: "List every operative section of this contract in order, with a one line description of each. Then list all defined terms, and flag any defined term that is used but never defined, and any cross reference that points to a section or exhibit that does not exist."
Verify as you go: spot check the section list against the actual document headings. Dangling cross references and undefined terms are real, common drafting errors, and a flagged one is a lead you confirm in the text.
Step 3: Pressure test the key terms
Now go deep on the clauses that carry the risk and the money. Do them as a set so the model can see how they interact, for example a broad indemnity colliding with a low liability cap.
Example prompt: "Analyze the indemnification, limitation of liability, termination, IP ownership, warranty, and payment clauses from the perspective of my client, the [buyer]. For each, explain in plain language what it does, whether it favors my client or the counterparty, and the specific risk it creates. Note any place where two clauses interact in a way that increases my client's exposure."
Verify as you go: open each clause the model discusses and read it yourself. This is the pass where a confident wrong reading does the most damage, so this is the pass where you check most carefully.
Step 4: Find what is missing
The hardest thing to catch in a contract is the clause that is not there. A blank space does not announce itself on a read through, and this is where a structured AI prompt earns its place.
Example prompt: "For a [services agreement] of this kind, list the protective clauses a careful lawyer for the [buyer] would expect to see. Then tell me which of those are present, which are weak, and which are entirely absent from this contract. Focus on liability caps, termination for convenience, confidentiality, data protection, mutual indemnity, and assignment."
Verify as you go: for every clause the model says is missing, search the document yourself to confirm it truly is absent and not just located somewhere unexpected. Then decide whether your client actually needs it here. Missing is a prompt for your judgment, not an automatic redline.
Step 5: Draft suggested edits
Finally, turn the findings into proposed language. You are not accepting these edits; you are getting a fast first draft to react to and rewrite.
Example prompt: "Draft proposed redline language for the three clauses we flagged as highest risk to my client. For each, give a suggested rewrite that improves my client's position, a more modest fallback position, and a one sentence rationale I could use in a negotiation note. Keep the drafting plain and consistent with the contract's existing style."
Verify as you go: treat every suggested edit as a first draft from a capable junior. Read it, rewrite it in your voice and your client's strategy, and confirm it actually says what you intend before it leaves your desk.
Honest usage notes
A few things become clear once you run this on real agreements rather than a demo contract.
The orient and map passes are where AI is most reliable and most time saving. Summarizing a deal and listing structure are exactly the kind of patient, comprehensive reading where a model rarely disappoints, and where doing it yourself is the most tedious. Start here and the rest of the review is faster because you already hold the shape of the document.
The missing clauses pass is the quiet star. Experienced lawyers catch absent provisions through pattern memory built over years, and even then a tired reviewer misses things. Asking the model what a careful lawyer would expect and which of those are absent is a genuinely useful second set of eyes, provided you confirm each call yourself.
The pressure test and drafting passes are where you stay most skeptical. A model can state a liability cap with total confidence and have the number wrong, or read an indemnity backwards. The output is fluent regardless of whether it is right, which is precisely the trap. Useful as a fast first analysis, never as a substitute for reading the clause. For the broader split between where AI helps on research versus drafting, our companion piece on AI legal research vs drafting goes deeper, and this method is one room in the larger map of how law firms run on AI.
Privilege, confidentiality, and verification
This is the part of the method you do not get to skip, and the part the vendor demos rarely dwell on.
Never put privileged or confidential data in a public model
A client's contract is confidential, and often privileged. Do not paste it into a free or consumer grade AI tool whose terms may use your inputs to train models or do not contractually protect the data. Use only an approved enterprise deployment that contractually commits not to train on your inputs and meets your firm's security requirements, or redact identifying and sensitive terms first, or do not use AI on that document at all. When in doubt, the document stays out. Our confidentiality guide for attorneys walks through the deployment questions to settle before any client matter touches a model.
Verify every flagged clause against the text
An AI flag is a hypothesis. Before you rely on any observation, redline any clause, or advise your client, open the actual provision and confirm the model read it correctly. Models can misstate a number, miss a defined term, or invert a clause's effect while sounding completely sure. The duty to be right is yours and is not delegable. The same fluent confidence that fabricates case citations in filings, covered in our briefing on AI citation hallucinations in legal filings, can misread a contract clause, so the same verification discipline applies.
This is not e-discovery
This method is for reviewing the terms of an agreement, clause by clause. It is a different task from reviewing large document sets for responsiveness or privilege in discovery, where the tooling and the validation questions are their own discipline. For that, see our comparison of generative AI vs TAR for document review.
How we built this method
This playbook reflects hands on use of leading general purpose models on the kind of contracts senior lawyers actually review: services agreements, master agreements, NDAs, and vendor contracts. The five pass structure is a practitioner workflow, not a product, and not a survey. The Leveraged Years just launched and we do not publish invented statistics or client results we do not have. Where we describe what AI is good and bad at, we mean what holds up in repeated practical use as of June 2026, on documents that contained no real client confidential information. AI capabilities change quickly, so we date this guide and refresh it. None of this is legal advice, and none of it changes your professional duties of competence, confidentiality, and candor. Confirm any approach against your jurisdiction's rules and your firm's policy before using it on a live matter.
What this means for your week
You do not need a new product to review contracts faster. You need a method you trust and run every time. Orient, map, pressure test, find what is missing, draft the edits, and verify every flag against the text. The hours that used to vanish into the first read collapse, and the time you save goes back into the judgment calls clients are actually paying for. The discipline that makes it safe, classify what can go into a model, verify everything, own the result, is the same discipline that makes you good at the rest of your practice.
That is the whole premise of how we train senior lawyers to work with AI: not faster, sloppier output, but the same standard of work reached with far less of the toil. The Leveraged Attorney course is built to install this contract review method and the rest of the system as a habit you can defend.
Part of TLY's AI Workflows → workflow playbooks for senior professionals.
Frequently asked questions
Is it safe to use AI for contract review as a lawyer?
It can be, with the right discipline. Use only an approved enterprise AI deployment that contractually commits not to train on your inputs and meets your firm's security standards, never a free consumer tool, for any client confidential or privileged contract. Treat every AI observation as a lead to verify against the actual clause, not a finding to trust. And remember the duty to be right stays yours: AI speeds the reading and the first markup, but you certify the result. Used that way it is a fast first reader. Used carelessly, by pasting confidential contracts into a public model or relying on unverified output, it is a confidentiality breach and a malpractice risk.
Can AI replace a lawyer for reviewing contracts?
No. AI can summarize a deal, map a contract's structure, flag risky clauses, and surface missing provisions far faster than a person can, which is real leverage. It cannot decide which risks matter for your specific client and deal, weigh them against the relationship and the negotiation, or take responsibility for the advice. Those are the judgment calls you are paid for, and they are not delegable to a model. The right framing is AI as a fast, structured first reader you direct and verify, not a replacement for the lawyer.
What is the best way to use Claude to review a contract?
Run it in five focused passes rather than one prompt. First, ask for a plain language summary of the deal from your client's perspective. Second, ask for a clause by clause structural map and a list of undefined terms and broken cross references. Third, pressure test the key risk and money clauses, indemnity, liability caps, termination, IP, and payment, from your client's side. Fourth, ask which protective clauses a careful lawyer would expect and which are missing. Fifth, ask for suggested redline language for the highest risk clauses. Read the contract yourself in parallel and verify every flag against the text before you act on it.
Should I use a contract review tool or a general AI model?
It depends on the work. For a high volume of near identical, lower risk contracts measured against a fixed standard, a purpose built contract review tool with a tuned playbook can be efficient. For senior review where judgment, client specific strategy, and defensibility matter, a method you run on a general model you control gives you more transparency and flexibility, because you see the reasoning and can adjust the prompt to the deal. Many lawyers use both: a tool for volume triage, the five pass method for the reviews that carry their name.
What should I never put into an AI tool when reviewing contracts?
Never paste a client's confidential or privileged contract, or any sensitive terms, into a free or consumer grade AI tool whose terms may use your inputs to train models or do not contractually protect your data. Confidential client matters belong only in an approved enterprise deployment that commits not to train on your inputs and meets your security requirements, or you redact first, or you do not use AI on that document. When you are unsure whether a deployment is safe, treat the document as off limits until you have confirmed it.
Build the method, not just the opinion
Knowing the five passes is the start. Running them every time, with the verification and confidentiality discipline baked in, is the skill that compounds. We teach the full method, the prompts, and the guardrails as one repeatable system a senior lawyer can defend to a partner and a client alike.
Start with Leveraged Attorney: the AI contract review and drafting system for lawyers Join The Leverage Club for $49 and get the prompts, clause checklists, and review templates Not sure where to start? Take the 2-minute course finderSources: Anthropic Claude enterprise and commercial data usage policies (Anthropic, 2026); ABA Model Rules of Professional Conduct on competence (1.1), confidentiality (1.6), and candor toward the tribunal (3.3); TLY hands on use of leading general purpose models on services agreements, master agreements, NDAs, and vendor contracts containing no real client confidential information (June 2026). Capabilities and vendor policies as published as of June 2026 and subject to change. This guide is not legal advice.