AI Contract Review: What It Catches and What It Misses

Most small businesses do not review contracts. They skim them. A supplier sends fourteen pages, somebody reads the price and the term, and it gets signed because paying a lawyer to look at a routine agreement feels like an overreaction. AI changes the maths on that, and it is genuinely worth using. It is also blind to the single most expensive category of contract problem, which is worth knowing before you rely on it.

What it does well

It reads fast and never gets bored on page eleven. That sounds trivial and it is most of the value, because the realistic alternative is not a careful human review. It is a skim.

Hand it a supplier agreement and you can reasonably expect a plain-language summary of the commercial terms, every date and renewal window pulled into one place, flags on clauses that sit heavily on one side, and a clean comparison against the version you signed last year. Two minutes, and you know more than you did. For a business that was going to sign after a skim, that is a real upgrade.

The blind spot

Here is the thing to internalise. AI reads the document in front of it. The costly problems in contracts are usually about what is not in the document at all.

A model asked to review a document will describe the document. Catching an omission requires knowing what ought to be there, which is knowledge the contract itself does not contain. You can partly close this by asking directly: what protections would you normally expect in an agreement like this that are not present here? That prompt gets you a useful list. It does not make the tool reliable at the job.

Do not assume it knows Canadian law

These models have read vastly more American contract material than Canadian. In practice that shows up as US framing applied to a Canadian agreement, references to the wrong statute, and silence on provincial specifics: Quebec operating under civil law rather than common law, employment standards that differ by province, consumer protection rules that are not uniform.

A useful question is which jurisdiction is this contract written for, and what does the governing law clause say. That is a reading task and it does it well. A dangerous question is whether this clause is enforceable in Ontario, because it will answer confidently either way and you have no signal about which time it is right.

Check the contract before you upload it

Worth pausing on, because it catches people out. Many agreements carry confidentiality obligations, and pasting one into a consumer AI product may breach the very document you are reviewing. Add your own privacy duties if there is personal information in there.

Use a business tier with a written commitment that your data does not train the vendor's models, check what retention actually is, and redact names and numbers where you can still get a useful answer without them. If the contract is under an NDA, read the NDA first. That advice sounds like a joke and it is the most common mistake we see. It sits alongside the broader question of what your team is putting into AI tools.

Where the line sits

Use AI freely on the routine flow: NDAs, standard supplier terms, renewals of things you have signed before, and first passes on anything you would otherwise have skimmed.

Call a lawyer when the contract is large relative to your business, runs for years, touches your intellectual property, involves employment or shareholders, or carries a downside you could not absorb. And notice the best use of all, which is using AI to arrive at that lawyer better prepared. Twenty dollars of AI time that turns a vague worry into three specific questions makes a four hundred dollar hour go much further. That is not replacing professional judgment. It is showing up ready.

Keep a record of what it said

One habit worth adopting from the start. When AI review informs a decision to sign, save the output alongside the contract. Six months later, when there is a dispute about what everyone understood, having the analysis you relied on is useful, and if the analysis was wrong you want to know that too. This is the same accountability point as knowing who can explain AI-produced work, applied to a decision rather than a document. Firms doing this at volume should also read how law firms handle AI document review, which is the professional version of the same workflow.

Frequently Asked Questions