AI contract review is one of the fastest-moving areas in legal tech. Claims range from "it catches everything a junior associate would miss" to "it's a liability waiting to happen." The truth is more specific and more useful than either extreme.
What the AI Actually Does
SignFlow Now uses Claude Sonnet — Anthropic's model optimised for nuanced reasoning — as the engine for contract analysis. When you upload a document, the AI does five things:
1. Document classification — identifies the document type. A tenancy agreement, engagement letter, NDA, and share purchase agreement all have different risk profiles.
2. Clause inventory — maps present clauses against what would be expected for that document type.
3. Gap analysis — flags expected clauses that are absent. Absence is often more significant than presence — a tenancy agreement missing a break clause is a significant risk point.
4. Risk identification — looks for provisions that are unusual, one-sided, or potentially problematic: liability caps at unusual levels, jurisdiction clauses that don't match the parties, auto-renewal clauses with short notice windows, unusually broad indemnities.
5. Scoring — produces an overall risk score (low/medium/high) and a list of specific findings with plain-English descriptions and suggested actions.
What It's Genuinely Good At
Catching missing clauses — this is where AI adds the most consistent value. The AI never gets tired, never rushes, and has analysed more contracts of each type than any individual lawyer.
Flagging unusual provisions — an NDA with a 10-year term, a liability cap set at £1 in a significant agreement, a commercial lease with an unusually low service charge cap.
Consistent application — human reviewers vary in what they focus on. The AI applies the same framework to every document, every time.
Pre-send review for standard documents — for engagement letters, NDAs, tenancy agreements, it adds genuine value as a pre-send check.
Where It Falls Short
Nuanced commercial context — the AI doesn't know your client relationship or the commercial background. A provision that looks unusual in isolation might be entirely reasonable given the context.
Novel or highly bespoke documents — gap analysis depends on having a well-established framework for that document type. Complex bespoke documents are less reliable.
Legal advice — AI contract risk scoring is not legal advice. It's a structured analytical tool. The interpretation, the advice to the client, and the decision about what to do — that's the lawyer's job.
Accuracy: Being Honest
In internal testing against UK legal documents, Claude Sonnet identifies approximately 85-90% of the structural gaps a human reviewer would flag. It has a higher false-positive rate on risk-within-clause analysis. Expect some findings you'll review and dismiss. Don't expect it to catch everything a specialist solicitor reviewing a complex document would catch. Do expect it to catch common structural gaps consistently.
How to Use It in Practice
Pre-send check, not a replacement for review — the most valuable use case is running AI review before you send a document to a client.
Act on structural gaps; review risk findings — when the AI flags a missing clause, take it seriously. When it flags risk within an existing clause, review in context first.
Use findings as a client communication tool — the structured summary is a basis for explaining to your client what's notable about the document before they sign.
Don't use it to justify not reviewing the document — professional responsibility remains with you.
The Bigger Picture
AI contract review will become standard in legal practice, just as document automation and legal research tools did. The firms that adopt it thoughtfully — using it for what it's good at, maintaining professional judgment where it's not — will be more efficient and will catch more problems. SignFlow Now's AI contract review is available on all Pro and Teams plans and runs automatically when you upload a document.