Should Your Startup Use AI to Review Contracts?

Yes. Run every contract that comes in the door through an AI tool before a human reads it. At the very least, it will give you a more efficient way to understand what it says. But then understand what the tool cannot see. It does not know your leverage, your investors, or the counterparty. It will treat terms that are normal for a large vendor as normal for you. It cannot tell you which of the eleven things it flagged are worth a fight, and when it is wrong, nobody is on the hook. What you pay a lawyer for has moved from the first read to the things the tool cannot do, and the founder's job is deciding which contracts need them.

What the tools are good at

Current AI tools read a contract fast and consistently. Hand one a sixty-page vendor agreement and it will tell you in a minute what the term is, whether it auto-renews, what the liability cap is, which party indemnifies whom, and whether the confidentiality clause survives termination, and if you then ask it to compare the document against a checklist it will tell you what is missing. Ask it to summarize the twelve customer contracts you signed last year and it will produce a table that would have taken a paralegal a day.

That was billable work not very long ago. A founder who runs inbound paper through a tool before anyone reads it will catch the auto-renewal clause, the uncapped indemnity, and the assignment restriction that would have surfaced two years later in diligence. Use it for that. The clients I see doing this well arrive on a call with a summary and three questions instead of a PDF and "is this okay," and we skip the first twenty minutes.

Where it will hurt you

In my experience the tools rarely misread a clause. What they get wrong is what the clause means for you, and the annoying part is that they get it wrong in exactly the tone they use when they are right.

The obvious gap is context. A contract is a record of the trade-offs two parties made with whatever leverage each one had that week, and the tool was not in the room. It does not know that the customer across the table is your first enterprise logo and worth a concession on payment terms, or that the vendor is one of four interchangeable options and should get none, or that your lead investor's side letter already promised a most-favored-nation clause that the new one would trigger. It is reading a document. The deal around it is invisible.

The less obvious gap is calibration, and it took me a while to notice it. These models learned what contracts look like from the contracts that exist, and the contracts that exist were mostly drafted by large companies for large companies. So a liability cap at twelve months of fees, a one-sided indemnity, and a broad license to the customer's data all read as ordinary, because they are ordinary for the party that wrote them. For a seed-stage company signing its fourth customer, two of those three terms can decide whether the company survives a bad quarter, and the tool will shrug at all three.

Clauses also interact. This is the one I would worry about most. An indemnity that looks standard is a different animal once you notice that the limitation of liability three pages later carves indemnities out of the cap, which means the obligation is unlimited. A termination for convenience clause is harmless until you read the exclusivity provision it leaves you stuck inside. The version I see most often is exclusivity. A customer asks you not to sell to three named competitors, which seems like a small price for a good logo, and the tool will call it narrow. Then the definitions section says "competitor" includes affiliates and successors, and the assignment clause lets the customer hand the contract to whoever acquires it. Two years later a large company buys your customer, and you have promised not to sell to a meaningful slice of your market, which is the kind of thing an investor's counsel finds in diligence and asks about on the call you were hoping would be short. (The tool, to be fair, will usually flag the assignment clause on its own. It just will not connect the two.)

Then there is accountability. I know how this sounds coming from a lawyer, but stay with me. When a lawyer tells you a term is acceptable and it is not, there is a person with a license and a strong interest in being right. When a tool tells you the same thing, there is a terms-of-service page that says the output may be inaccurate. That asymmetry is not a reason to avoid the tool. It is a reason to know, before you sign, which documents you are willing to sign on a machine's say-so.

A practical note, since this one bites people. If the other side sent you a draft under a nondisclosure agreement, pasting it into a consumer AI product whose terms allow it to retain or train on your inputs may itself breach the NDA you are reviewing. Use a business tier with contractual confidentiality, or ask the vendor the question in writing. And a conversation with a tool is not privileged; a conversation with your lawyer is. If the dispute ever arrives, one of those is discoverable and the other is not.

What you are paying a lawyer for now

Not the first read. I will say that plainly because a lot of lawyers will not. If a lawyer's value to you is reading the document and telling you what it says, the tool has already replaced that lawyer.

What is left is judgment, which is a word lawyers use when they do not want to be specific, so here is what I mean by it. Part of that is telling you what the document means for this company at this stage with these investors and this counterparty, which is a question about your business rather than about the text, and the tool has read only the text. A larger part is deciding what is worth a fight, because most redlines cost more in goodwill and calendar than they recover, and a good startup lawyer's main skill is knowing the three terms that matter on this deal and letting the other eight go. Then there is standing behind the advice. The counterparty's lawyer pushes back, or investor counsel asks in diligence why a clause reads the way it does, or the term you accepted turns out to matter after all, and there is a person to call who gave the advice and will defend it.

And one more thing, which gets less attention than it should: somebody has to decide what the tool checks for. A review is only as good as the playbook behind it, and the playbook for a seed-stage software company is not the playbook for a marketplace or a hardware company. The lawyer who knows what has gone wrong across a hundred similar companies is the one who should be writing the checklist the tool runs against.

A rule for which contracts go where

Let the tool handle, on its own, the inbound paper you are not going to negotiate: standard SaaS subscriptions for tools under a few thousand dollars a year, vendor terms where you have no leverage anyway, and nondisclosure agreements that come back clean on a quick triage (mutual, reasonable term, no non-solicit or non-compete hiding in it). Read the summary, note anything odd, and sign.

Send to a lawyer anything that touches your capitalization, your intellectual property, or your freedom to run the company. That covers SAFEs, notes, and anything else investors sign, customer contracts large enough to show up in a data room or that grant exclusivity or most-favored-nation pricing, anything assigning or licensing IP in either direction, employment and contractor paper for the people building the product, and any document a counterparty drafted that you intend to sign more than once. The last one surprises founders. A template you adopt gets signed forty times, so the review that costs a few hours today is amortized across every customer you have for the next three years.

If a document does not fit cleanly in either bucket, send it. The downside is an invoice for a review you did not strictly need, which is a bad outcome in roughly the way a flu shot is a bad outcome.

Where to start

Pick a tool with business-grade confidentiality, build a short checklist of the terms that matter for your kind of company, and run every inbound contract through it this week. Then write down the rule above for which contracts go where, and share it with whoever on your team signs things. If you are not sure where a document lands, the Contact page reaches me, and the cost guide will tell you roughly what a human review should run so you can decide whether it is worth it.

Frequently asked questions

Is it safe to put a contract into ChatGPT or another AI tool?

It depends on the tool's terms and on what the contract says. Consumer tiers of many AI products reserve the right to retain or train on what you type, and if the document arrived under a nondisclosure agreement, uploading it there may breach that agreement. Business and enterprise tiers typically offer contractual confidentiality and no training on your data. Check the terms, use the business tier, and keep anything subject to an NDA or containing customer data out of consumer products.

Can AI replace a startup lawyer?

It replaces the first read, and for low-stakes paper you were never going to negotiate, it can be the whole review. It does not replace the judgment about what a term means for your specific company, which terms are worth negotiating, or the accountability of a licensed professional standing behind the advice. Financings, IP, and anything that will be in a data room still belong with a lawyer.

Can I use AI to draft my contracts instead of reviewing them?

For a first draft of a simple document, yes, with the same caveats: review it against a checklist, and have a lawyer review anything you intend to reuse. The risk in drafting is different from the risk in review. A reviewed contract has a counterparty who caught the other side's mistakes; a drafted one does not. Templates that touch equity, IP, or investors should start from a lawyer's form, not a prompt.

Is my conversation with an AI tool privileged?

No. Attorney-client privilege protects communications with your lawyer for the purpose of legal advice. A chat with a software product is not that, and in litigation it can be requested like any other business record. If you want a question to stay confidential, ask your lawyer.

Which contracts should always go to a lawyer?

Anything that touches your cap table, your intellectual property, or your ability to run the company: financing documents, IP assignments and licenses, customer contracts with exclusivity or most-favored-nation terms or large enough to appear in diligence, employment and contractor agreements for the people building the product, and any counterparty template you plan to sign repeatedly.