A contract before you sign: let AI find the catches

Contracts get signed in the worst frame of mind. You've been apartment-hunting for three weeks, the bank is waiting on an answer, the new job starts Monday — and in front of you sits fourteen pages of text you lose focus on after the second paragraph. Most people deal with that by skimming it and signing. AI won't tell you whether to sign. What it can do is pull out exactly the passages that are hardest to read and hurt the most — auto-renewals, penalties, one-sided changes, notice periods — and turn them into concrete questions you take to the other party or to a lawyer.

A typical scenario

Tom is thirty-one, does freelance graphic design, and after a month of searching finally found an apartment. The landlord sent him the lease on a Friday afternoon, signing Monday. Fourteen pages, small print.

Tom uploads the contract and has a list within ten minutes. A one-year fixed term with automatic renewal unless the tenant gives written notice three months in advance — a deadline that's easy to forget in practice. A deposit of three months' rent, but the contract is missing a sentence about when and how it gets returned. A penalty of a fixed amount per day of late payment, with no cap at all. And a clause letting the landlord raise the rent unilaterally “in line with price developments,” with no defined mechanism.

Tom didn't send a legal analysis. He sent six politely worded questions. The landlord fixed two things himself (the penalty cap, the deposit-return deadline) and added a reference to the official inflation index for the rent-increase clause. Tom signed on Monday — just knowing what he was signing.

How to do it

1. Handle the data first, the content second. A contract contains your ID number, address, income, sometimes even health information. Use a paid or business account with contractual data protection, not a freely available chat — and anonymize what you can: replace ID numbers, account numbers, and names with “TENANT” and “BANK.” Nobody needs them for a risk analysis.
2. Give it a role and a purpose, not just “look this over.” Say who you are in the relationship (tenant, employee, borrower, policyholder). An analysis written from the tenant's side looks different from one written from the landlord's side — and you need your own.
3. Have it work through a fixed list of categories. Give AI an outline so it isn't guessing what to check: automatic renewals and the deadlines to reject them; termination grounds and notice periods on both sides; penalties and interest, including whether they're capped; unilateral changes to terms; security (deposit, guarantee, promissory note); and above all what's missing from the contract — silence is often more dangerous than a paragraph in small print.
4. Ask about what's non-standard. A useful question: “Which clauses in this type of contract are unusual or significantly disadvantageous for my side?” Ask for a reference to the specific article and a quote of the sentence, so you can verify every claim against the original document — see extracting a document with page numbers.
5. Make the output a list of questions, not a verdict. The last step of your prompt: “Prepare five to ten specific questions for the other party, politely worded, each with a reason why I'm asking.” Either the other side answers them and you're clear, or they start dodging — which is an answer too.
6. Decide whether you need a lawyer. For an ordinary lease or insurance policy, the list of questions is often enough. Once it's serious money, a non-compete clause, personal liability, or a loan spanning decades, go see a lawyer — but bring questions, not a stack of paper. An hour of consultation over five specific points is worth more than three hours of reading blind.

The best tools

• Claude — upload a PDF or a scan and get an analysis with references to specific articles; it can also compare two versions of a contract and list what changed between them.
• Projects (persistent context) — if you sign contracts often, set up a project with your template and standards; every new contract then gets assessed against what you normally accept (projects and persistent context).
• A lawyer or legal clinic — the only tool on this list that carries any liability. Students often have a free university legal clinic; unions handle employment contracts.

What you get out of it

• Peace of mind: the difference between “I signed something” and “I know what I signed” mostly comes down to how well you sleep in the following months.
• Time: fourteen pages of legalese take an hour of focused reading. You get a risk map in a few minutes and read with purpose after that.
• Money: an auto-renewal caught in time, or an uncapped penalty you get a limit added to, is worth real money — often in the thousands.
• Better negotiations: the other side reacts differently to “something feels off” than to “Article 7.3 has no cap on the penalty, can we add one?”

Pro tip

AI is not a lawyer and shouldn't be treated as one. It doesn't know your local case law in detail, it can miss the connection between two distant paragraphs, and it sometimes fills in things that aren't actually in the contract. Treat its output as an attentive colleague who read the text before you did — not as legal advice. Verify every significant claim against the original document, and keep the decision for yourself. The same rule applies here as everywhere: AI proposes, the person approves — and only a person signs.