AI Contract Review Governance
Contract review is the most-deployed AI use case in legal practice. The "AI beats lawyers" headline is narrower than it sounds — and the supervision rules are tighter than they look.
The Market in One Glance
Four established platforms plus generative-AI new entrants
Kira Systems
Acquired by Litera in 2021. Clause extraction, M&A diligence — the long-pedigree leader in the category.
Evisort
Acquired by Workday in 2024. Contract lifecycle management with AI extraction.
LinkSquares
Analytics-focused contract review platform.
Spellbook
Word-integrated AI drafting and review for transactional work.
Harvey
Generative AI extending into contract drafting and negotiation workflows.
Lexis+ AI Contract Drafting
LexisNexis's GenAI play in transactional drafting and review.
The LawGeex Benchmark Every Vendor Cites
2018, NDA-only, vendor-funded — but the operating data point in the category
Cite the LawGeex Number Carefully
The 2018 LawGeex study is the most-cited "AI beats lawyers on contracts" data point. It is also vendor-funded, limited to NDAs (the most standardized contract type), seven years old, and not replicated by a peer-reviewed independent study on complex commercial contracts at equivalent rigor.
The "AI is faster than lawyers on NDAs" claim is well-supported. The "AI is more accurate than lawyers on contracts generally" claim is not.
Where AI Helps in Contract Review
Five high-fit use cases
Clause Extraction
Pulling defined terms, indemnification, termination, IP assignment from large contract sets. Mature, reliable, well-deployed.
NDA Review at Speed
Standard NDAs are well-suited to AI review — limited variability, deterministic issues.
First-Pass Redlines
Initial markup of routine agreements, with attorney review of every change.
M&A Diligence
Pulling key terms from large M&A document sets for human review.
Compliance Checks
Confirming a counterparty-drafted agreement matches firm-approved positions.
Renewal Tracking
Surfacing renewal dates, auto-renewals, and notice windows across contract portfolios.
Where AI Fails in Contract Review
Four failure modes that determine where the firm must require human-first review
Novel or Non-Standard Contracts
Anything outside the training distribution — bespoke transaction documents, custom indemnity structures, novel IP arrangements — produces lower accuracy.
Why it matters: The LawGeex NDA benchmark does not generalize to these.
Complex Commercial Contracts
No peer-reviewed evidence matches the LawGeex NDA numbers on master services agreements, JVAs, or transaction documents at scale.
Why it matters: Treat AI output as draft, not as judgment.
Strategic Judgment
Identifying that a clause is technically present but operates poorly given the deal's commercial reality.
Why it matters: Strategic judgment is what clients pay partners for; AI does not substitute.
Drafting Precision
Generating tight contract language that survives litigation scrutiny — AI tends to introduce ambiguity or stylistic inconsistency.
Why it matters: Bespoke drafting is a poor fit; review-and-redline is a good fit.
Governance Under ABA Op. 512
Contract review AI is the textbook Rule 5.3 supervision scenario
Defined scope of AI use
Written policy specifying which contract types AI may first-pass review (e.g., NDAs, standard services agreements) and which require human-first review (e.g., M&A transaction documents, novel agreements).
Mandatory human review
Every AI-flagged issue reviewed by a qualified attorney before client delivery or counterparty exchange.
Verification documentation
Sign-off by reviewing attorney on each AI-assisted document.
Training on tool limits
Associates and paralegals using the tool understand what it does well, where it fails, and how to identify failure modes.
Periodic accuracy benchmarking
Sampling completed AI-assisted reviews against blind attorney review to track real-world accuracy on the firm's actual contract mix.
The Confidentiality Boundary
Contract review almost always involves client confidential information. The tooling tier determines the Rule 1.6 analysis.
Enterprise-grade contract review tools (Kira, Evisort enterprise, etc.) typically include the contractual terms necessary to satisfy Rule 1.6. General-purpose AI used to review contracts (pasting a contract into ChatGPT) fails the same Rule 1.6 analysis covered in attorney-client privilege guidance.
AI Contract Review Governance — FAQ
Is AI more accurate than lawyers on contract review?
On NDAs specifically, yes — LawGeex's 2018 study found AI at 94% accuracy vs. 85% for experienced human attorneys, with AI taking 26 seconds vs. 92 minutes per NDA. But that study was vendor-funded, limited to NDAs, and has not been replicated at equivalent rigor on complex commercial contracts. The 'AI beats lawyers' claim is narrower than it is often presented.
What does ABA Op. 512 require for AI-assisted contract review?
Treat AI as analogous to a nonlawyer assistant under Rule 5.3. Define scope of AI use, require human review of every AI-flagged issue before client delivery, document attorney sign-off on AI-assisted work product, train staff on tool limits, and bill for actual labor under Rule 1.5.
Can a firm use ChatGPT to review client contracts?
Not without specific client consent or enterprise-grade contractual terms. Pasting a client contract into consumer ChatGPT inputs client confidential information into a tool whose terms permit retention and (in some configurations) training — the same Rule 1.6 problem that drives the analysis on legal research.
Where does AI contract review fail?
Novel or non-standard contracts, complex commercial transactions outside the training distribution, strategic judgment (whether a clause operates well commercially), and drafting precision (AI tends to introduce ambiguity in original language). Firms should restrict AI first-pass review to standardized contract types and require human-first review on bespoke documents.
Related Resources
Continue across the silo or bridge to a core hub
ABA Formal Opinion 512
Rule 5.3 supervision in practice — the contract-review reference case
Read article →Attorney-Client Privilege and AI
Rule 1.6 analysis on the contract data going into the tool
Read article →AI Hallucinations in Legal Practice
Sanctioning orders that turned on supervising-attorney failures
Read article →Multi-Model AI Access
Enterprise tooling that supports MNPI-adjacent contract review without leakage
Read article →Governed AI Platform Checklist
Platform features that scale review safely across associate teams
Read article →Build a Contract Review AI Workflow That Survives Malpractice Review
Free Shadow AI Risk Check audits your scope-of-use policy, your Rule 5.3 supervision documentation, and your accuracy benchmarking practice.