AI Contract Review: What to Automate and What Never to Automate
Quick answer
Automate locating, never judging. AI is excellent at finding every deadline, termination right, liability cap and defined term in a contract, and it is no substitute for legal judgment about whether a clause is acceptable or enforceable in your jurisdiction. Start with an obligations-and-dates inventory in Chat with PDF, ask for the text of each exposure clause rather than a paraphrase, and redact confidential values before uploading anything you are bound to protect.
The distinction that matters: locating versus judging
AI is genuinely excellent at locating things in a contract — finding every deadline, every termination right, every reference to indemnity, every defined term and where it's used. That's a retrieval problem, and retrieval is what these systems do well.
AI is not a substitute for legal judgment about whether a clause is acceptable, enforceable in your jurisdiction, or unusual for your industry. That requires knowing things the document doesn't contain. The workflow below leans hard on the first capability and doesn't pretend about the second.
Start with an obligations and dates inventory
The highest-value first pass on any contract is a complete list of what each party must do and by when. Open the document in Chat with PDF and ask for every commitment, deadline, notice period, and renewal or termination date, along with which section each appears in.
This is the pass that catches the auto-renewal buried in section 14 and the 90-day notice requirement that determines whether you can still get out. It takes minutes and it's the item most commonly missed in a manual first read, because those clauses are deliberately unremarkable in the surrounding text.
Then check the commercial exposure clauses
Ask specifically for liability caps, indemnity obligations, warranty disclaimers, governing law, and dispute-resolution mechanism. These are the terms that determine what happens when things go wrong, and they're the terms a reviewer under time pressure is most likely to skim.
Ask for the text of each, not a paraphrase. You want to read the actual clause — the AI's job here is to find it and tell you where it is, not to characterize it for you.
Comparing a redline against the version you agreed to
When a counterparty returns a marked-up draft, the risk isn't the changes they told you about — it's the one they didn't. Compare PDF puts the two versions side by side and surfaces every difference between them, including the single-word changes that quietly reverse a clause's meaning.
A changed "shall" to "may", a removed "not", a cap that moved from 12 months' fees to 12 months' revenue: these are easy to miss reading two drafts in sequence and hard to miss when the differences are extracted for you.
Confidentiality: redact before you upload
Contracts routinely contain client names, deal values, and personal data of individuals who never agreed to have it processed by a third-party service. If you're bound by confidentiality obligations — and on a contract, you usually are — that's a real constraint, not a theoretical one.
Redact PDF and Smart Redact let you remove identifying values before analysis, so you can review structure and terms without the sensitive specifics leaving your control. Check your own firm's or organization's policy on third-party processing first; that policy, not a vendor's assurance, is what governs.
What to never hand over
Do not use AI output as the basis for advising a client, signing an agreement, or concluding that a clause is enforceable. It has no knowledge of your jurisdiction's case law, your client's risk tolerance, or the negotiation history that explains why a term is written the way it is.
The right framing is that this turns a two-hour first read into a fifteen-minute one and hands a lawyer a mapped document with the important clauses already located. The judgment stays exactly where it was.