How to run an AI legal review process for contracts without a full attorney
Outside counsel at $400–$600 per hour is the right answer for a $2M acquisition, a contested employment termination, or a novel IP licensing structure.
It is the wrong answer for the twelfth NDA variation of the year, the vendor software subscription renewal, and the client MSA that looks like the last twelve you have signed.
AI does not replace the lawyer for complex matters. It eliminates the need for a lawyer on the ones that do not require one; and prepares a better brief for the ones that do.
The average mid-market company sends routine contracts to outside counsel because nobody on the team is confident enough to review them alone.
Most of those contracts are standard vendor terms, NDA variations, and client MSAs that follow a predictable pattern.
AI can read, extract, flag, and summarize every significant clause in a standard contract in under five minutes; which changes the economics of contract review without removing the human judgment that matters.
What AI can and cannot do in contract review: the honest capability map
What AI does reliably:
- Reads and extracts all major clauses from standard commercial contracts; regardless of length, formatting, or PDF quality
- Identifies the type and structure of each clause (limitation of liability, indemnification, termination, data processing, auto-renewal, change of control, governing law, dispute resolution)
- Produces a plain-language summary of what each clause actually means in operational terms; not legalese
- Compares the contract’s terms against the company’s standard positions and flags any clause that deviates
- Identifies missing clauses; terms that should be present in a contract of this type and are not
- Calculates financial exposure from specific clauses (liability caps, penalty provisions, auto-renewal costs)
What AI does not do reliably:
- Assess how a specific clause will be interpreted by a court in the applicable jurisdiction
- Evaluate the negotiating leverage the company has and whether flagged issues are worth pushing on
- Identify novel legal risk that requires awareness of recent case law or regulatory changes
- Provide a professional legal opinion the company can rely on if a dispute arises
- Determine whether a non-standard clause is acceptable for this specific counterparty and relationship context
The decision gate that makes the workflow safe:
AI produces the summary and flags. A responsible person on the company’s team; the founder, COO, or general commercial manager; reviews the summary and flags, and decides: proceed as-is, request negotiation, or escalate to outside counsel.
The AI never makes the legal decision. The team member who makes it is accountable for it.
The contract playbook: the context that makes AI review specific
A generic AI prompt (“review this contract and flag risky clauses”) produces a generic review. A review against the company’s specific playbook produces a targeted analysis that tells the team whether this contract is acceptable, negotiable, or concerning; without a lawyer present.
Section 1: Standard positions by clause type
| Clause type | Company standard position | Non-negotiable? |
|---|---|---|
| Limitation of liability | Cap at 12 months of fees paid under the agreement | Negotiable to 24 months; anything above: rejected |
| Indemnification | Mutual; limited to IP infringement and gross negligence | Non-mutual indemnification: escalate |
| Auto-renewal | No auto-renewal without 30-day written notice | Less than 30-day notice: flag |
| Governing law | Company’s home state or mutual agreement | Counterparty-only jurisdiction: escalate |
| Data processing | GDPR/CCPA-compliant DPA required for any personal data processing | Missing DPA for data-processing agreements: non-negotiable |
| IP ownership | Company retains all IP created under the agreement | Work-for-hire clauses assigning IP to counterparty: escalate |
| Payment terms | Net 30 | Net 60+: flag for discussion |
| Termination for convenience | Minimum 30-day notice period | Less than 30 days or no termination right: flag |
Section 2: Non-negotiable terms
Specific clauses the company will not sign without modification regardless of counterparty or deal size. These are the automatic escalation triggers; the AI flags them, the team treats them as immediate referrals to counsel.
Section 3: Common counterparty variations and how to handle them
For the counterparty types the company regularly contracts with (major software vendors, enterprise clients, staffing agencies, professional services providers), notes on common deviations from the standard and the standard response to each.
Building the playbook:
This is a one-time, 3–4 hour exercise completed in collaboration with outside counsel; ideally in a 90-minute working session. The lawyer reviews the playbook structure, fills in the company’s actual positions, and flags the non-negotiables. This consultation costs $500–$1,000 once; it replaces ad hoc counsel involvement on dozens of routine contracts annually.
The AI contract review workflow: step by step
Step 1: Contract intake and triage (5 minutes)
Receive the contract. Before running AI review, determine which track it belongs to:
- AI-first track: NDAs, standard vendor/supplier agreements, software subscription terms, standard client MSAs based on the company’s own form, lease renewals, standard consulting agreements
- AI-assisted attorney track: employment agreements with non-standard provisions, acquisition-related documents, IP licensing with novel structures, any contract above a specified deal value threshold (e.g. $100K+ annual value)
- Direct attorney track: litigation-related documents, regulatory filings, complex M&A, employment disputes
If the contract is on the AI-first track: proceed. If not: move directly to outside counsel with the appropriate brief.
Step 2: Run the AI review (5–10 minutes)
You are reviewing a [contract type] between [Company] and [Counterparty].
COMPANY STANDARD POSITIONS: [Paste relevant sections of the contract playbook]
Review the contract and produce:
1. CONTRACT SUMMARY — Plain-language description of what this agreement
does, what obligations it creates for each party, and the key commercial
terms (value, duration, payment terms)
2. CLAUSE INVENTORY — List every major clause with: clause name, what it
says in plain English, and whether it is standard or non-standard relative
to our playbook
3. FLAGS — Every clause that deviates from our standard positions, with:
(a) what the contract says
(b) what our standard position is
(c) the practical implication of the deviation
(d) whether this is a minor flag, a negotiation flag, or an escalation flag
4. MISSING CLAUSES — Any clause types that should be present in this type
of agreement and are not
5. FINANCIAL EXPOSURE — Any clauses with financial consequences (liability
caps, penalties, auto-renewal costs) with the specific amounts calculated
where possible
6. RECOMMENDATION — Proceed as-is / Proceed with negotiation on [specific
flagged items] / Escalate to outside counsel
Contract text: [paste contract]Step 3: Team review of the AI output (15–20 minutes)
A responsible team member reviews the AI output; not the original contract. The AI output is the brief; the team member makes the decision.
The review focuses on:
- Are all flagged items understood? If any flag is unclear, the team member asks the AI for clarification on that specific clause.
- Are the escalation flags genuine? The AI may occasionally flag something as an escalation item that the team member’s business context clarifies as acceptable.
- Is the recommendation appropriate?
Step 4: Decision and action (5 minutes)
- Proceed as-is: file the AI review output with the signed contract as the company’s review record
- Proceed with negotiation: use the AI-drafted redlines (from Step 5) as the basis for negotiation
- Escalate to outside counsel: send the contract and the AI review output to outside counsel, with a cover note identifying the specific issues requiring professional judgment
The AI review compresses the lawyer’s briefing time by 50–70%.
Step 5: Redline drafting (optional, 5–10 minutes)
When negotiation is warranted, the AI drafts redline language for each flagged clause:
For each of the following flagged clauses, draft replacement language
that reflects our standard position while remaining commercially
reasonable for a counterparty of this type:
[List of flagged clauses from Step 2 output]
Our standard positions: [Paste relevant playbook sections]The AI-drafted redlines are reviewed by the team member before being sent. For significant negotiations, outside counsel reviews the AI-drafted redlines; but the drafting time is compressed.
The three contract types in detail: how the workflow differs
NDAs and confidentiality agreements
The most routine contract type and the best starting point for building AI review confidence. NDAs follow a very consistent structure; the primary variables are scope of confidentiality, exclusions, duration, and what happens to confidential information on termination.
Primary flags to look for: one-way versus mutual obligations; overly broad definition of confidential information; no carve-outs for information already known or publicly available; auto-renewal without notice; jurisdiction that creates enforcement difficulty.
Escalation trigger: one-way NDA that requires the company to protect information without reciprocal protection; or an NDA that covers operational information at a scope that would affect normal business operations.
Standard vendor and supplier agreements
Higher stakes than NDAs but still highly patterned. The primary variables are payment terms, IP ownership, liability caps, indemnification scope, and data processing obligations.
Primary flags to look for: IP assignment clauses that transfer ownership of work product beyond what was intended; limitation of liability caps below the annual contract value; indemnification provisions significantly broader than mutual; missing DPA where personal data is involved; auto-renewal with insufficient notice period.
Escalation trigger: any clause that would assign IP created under the agreement to the vendor; any limitation of liability below 3 months of contract value; any indemnification provision that is non-mutual and broad.
Client service agreements based on the company’s standard form
When the company is sending its own standard MSA to a client, the AI review focuses on what the client has changed from the standard form; a redline comparison rather than a full review.
Prompt adjustment:
“Compare this agreement against our standard form [paste standard form]. Identify every change the counterparty has made from our standard language, explain the commercial implication of each change, and flag which changes are acceptable, which require negotiation, and which are non-negotiable.”
Primary flags to look for: changes to payment terms; changes to IP ownership provisions; changes to limitation of liability; addition of indemnification obligations not in the standard form; removal of governing law or dispute resolution provisions.
What to do with outside counsel when you use AI review: the changed dynamic
The traditional outside counsel engagement on routine contracts: send the contract, lawyer reads it, lawyer produces a summary and flags, lawyer recommends positions.
Cost: 2–4 hours at $400–$600/hour = $800–$2,400 per contract.
The AI-assisted outside counsel engagement: send the AI review output and the original contract, lawyer reviews the AI flags, confirms or adjusts the AI’s analysis, provides professional opinion on the escalation flags.
Cost: 30–60 minutes at $400–$600/hour = $200–$600 per contract.
The compound effect:
For a company that sends 20–30 contracts to outside counsel annually, the AI review workflow reduces counsel engagement to 5–8 contracts (only those with genuine escalation flags) and compresses the time on those.
Annual legal review spend can drop from $30,000–$50,000 to $8,000–$15,000.
What to tell outside counsel:
“Our AI review identified three specific issues I would like your opinion on: [list the escalation flags]. The attached review covers the full contract; I am not asking you to re-read it from scratch, just to advise on these three points and confirm whether our AI review missed anything significant.”
This framing protects the legal relationship, focuses the engagement, and reduces the time cost.
Common questions on AI contract review
”Is an AI contract review legally binding or admissible?”
No. An AI contract review is an internal analysis tool; it is not a legal opinion and is not admissible as one. It identifies what the contract says and how it compares to the company’s standard positions.
The legal interpretation of those terms and the decision to sign or escalate remains with a human; and significant matters remain with a qualified lawyer.
”What if the AI misses something and I sign a bad contract?”
The risk is real and should be managed by calibrating which contract types use the AI-first track. The playbook’s non-negotiable terms list is the primary safety net; any deviation from those terms must be flagged before signing regardless of what else the AI review produced.
For contracts above a specified value threshold or with novel provisions, escalate to outside counsel regardless of the AI’s recommendation.
”Can I use this for employment contracts?”
Employment contracts with standard terms (offer letters, standard employment agreements) can benefit from AI review for completeness checking and flagging non-standard clauses. Employment contracts with non-standard provisions (unusual non-compete language, equity arrangements, separation agreements) should go directly to outside counsel. Employment disputes and termination-related documents: direct attorney track only.
”What is the liability if we rely on AI review and get it wrong?”
The company bears the liability for any contract it signs regardless of the review process used. AI review does not transfer liability; it improves the quality and consistency of the internal review process. For contracts with material financial exposure, professional legal review remains appropriate.
”How do I build the contract playbook if I don’t have any existing legal guidance?”
Book a 90-minute working session with outside counsel specifically to build the playbook. Present the table structure above. The lawyer fills in the standard positions for each clause type and flags the non-negotiables based on the company’s industry and risk tolerance.
This is one of the highest-ROI legal consultations the company will have; the playbook it produces reduces counsel involvement for years.
”What AI tool is best for contract review?”
Claude Opus and GPT-4 both handle contract review well on standard commercial agreements. For a company doing high-volume contract review, purpose-built contract review tools (Kira, Luminance, Spellbook) offer structured extraction features.
For most $5M–$25M companies doing 20–50 contracts per year: Claude or GPT-4 with the prompt structure above is sufficient and significantly cheaper than purpose-built tools.
Want the contract review workflow built and the playbook written; so the process runs from the first document?
AI contract review does not replace legal counsel. It replaces the part of legal counsel that is pattern recognition; reading a contract to find out what it says, comparing it to a standard, and flagging what is different.
What remains after AI handles it; the professional judgment about whether flagged issues are material, how to negotiate them, and whether the company’s legal position is defensible; is exactly what outside counsel is worth paying for.
Path one: build the contract playbook this week. Use the table structure above. Book a 90-minute working session with outside counsel to fill it in. That session produces a document that changes the economics of every routine contract review for years.
Path two: bring in a partner. If you want the contract review workflow built, the playbook written, and the decision rules documented as part of the company’s AI context pack; that is the work Phos AI Labs does in Phase 1. The team behind Phos AI Labs has helped 400+ businesses run on AI. The fastest way to know if it is the right fit is a conversation. Thirty minutes, no deck. Start here.