Before You Prompt
The five-step ChatGPT setup that actually reviews a contract — from a GC who reviewed 12 of them this way.
A freelancer pastes a 14-page services agreement into ChatGPT. "Is this contract okay?" ChatGPT replies: "This appears to be a standard services agreement. There are a few clauses you may wish to review with a legal professional." So they sign. Six months later the auto-renewal locks them in for another year at a rate they wanted to renegotiate. The AI did not fail them. They asked the wrong question.
The question that gets you nowhere
Most people ask AI to review a contract the way they would ask a lawyer: "Is this standard?" "Is this okay?" "What are the risks?"
These are legal questions. AI gives legal-flavoured answers: vague, hedged, covered in disclaimers, and ultimately useless. A summary of the contract. A list of things to "consider."
The right question is commercial. Not "is this clause standard?" but "what does this clause cost me if it is triggered?" Not "what are the risks?" but "which clause could take money out of my pocket, and how much?"
That shift changes everything the AI returns to you. The same contract. The same AI. A completely different output.
Before you paste a single clause
This is where most people go wrong. They open a new chat, paste the contract, and type "review this." What follows is the setup that actually works. Five things to do before you ask AI anything about a contract.
1. Tell it who you are and what side you are on. AI does not know if you are the client or the supplier. It does not know if you are a sole trader or a limited company. It does not know your industry or your location. Tell it before you paste anything.
"I am a freelance designer. I am the one being hired. I am based in the UK."
That context changes what it flags. A liability clause reads very differently depending on whether you are the one accepting liability or the one enforcing it. An IP ownership clause matters differently to an agency than to an individual. Context is not a nice-to-have. It is the whole frame.
2. Tell it what you want in commercial terms, not legal terms. Do not ask "review this contract." Ask: "Read this contract and tell me which clause could cost me the most money if things go wrong. What is the worst-case financial outcome of signing this as written?"
Commercial language gets commercial answers. Legal language gets legal answers. You are not trying to pass an exam. You are trying to know what you are signing up for.
3. Give it the right lens. Tell it which clauses matter most for your situation.
For a freelancer: payment terms, IP ownership, liability cap, auto-renewal, kill fee / termination for convenience.
For a founder receiving a SaaS contract: data use, AI training restrictions, termination rights, price escalation, exclusivity / change-of-control.
If you do not tell it what to look for, it will tell you everything and nothing. You will get a ten-point summary where point three (auto-renewal with no opt-out) is buried between point two (governing law) and point four (notice period). You need to tell it what actually matters to you commercially.
4. Ask for consequences, not classifications. There is a real difference between these two outputs:
“This is an unlimited liability clause.” (Legal classification. Useless.)
“If the client claims you caused them a loss, this clause means they can come after everything you have, with no ceiling on the amount.” (Commercial consequence. Actionable.)
Ask explicitly: "For each risk you find, tell me what it costs me in plain English. Not what the clause says. What it means for my money, my time, or my business."
5. Ask what to do, not just what it says. "What should I push back on, and what is a reasonable alternative?" is a better question than "what does this mean?"
You want action, not analysis. A good commercial clarity check does not just describe the problem. It tells you what a fairer version of that clause looks like, and whether this is a must-fix or a nice-to-fix. That is what turns a contract review into a negotiation plan.
The prompt, ready to paste
Copy this. Fill in the brackets. Then paste your contract underneath.
I am a [role] based in [country]. I am the party [being hired / hiring]. I sign this kind of contract roughly [X] times a year. Read the contract below and tell me:
1. The three clauses most likely to cost me money if triggered.
2. The worst-case financial outcome of each, in plain English.
3. What to push back on and a fairer alternative for each.
4. Anything missing that I would normally expect to see in this type of agreement.
Frame everything in commercial consequence, not legal classification. Tell me what each risk means for my money, my time, or my business.
That is the setup. Use it and the same contract through the same AI will read very differently.
What it still will not give you
The five steps above will give you a better ChatGPT output. They will not give you the four things that actually matter when something goes wrong.
They will not give you severity. Even with the best framing, ChatGPT returns a flat list of concerns. The one that could bankrupt you sits beside the one that does not matter at all. You still have to work out which is which, and you are doing that work without a framework.
They will not tell you what is missing from the contract, only what is in it. A general-purpose AI is descriptive, not evaluative. The clauses that cost people money are usually the ones that should be there and are not: no liability cap, no kill fee, no late-payment terms, no IP-on-payment trigger. No prompt template can surface a clause that does not exist in the document.
They will not give you the exact words to send back. A good review tells you what to push back on, in what order, with the precise counter-language a counterparty's lawyer will read and accept. Without that, the review ends where the negotiation begins.
And they will not give you English law. ChatGPT defaults to US "work made for hire" conventions, US damages frameworks, US payment norms. Some of it does not translate. Some of it actively misleads. No framework on top of the model fixes the model.
Those four things - severity, what is missing, exact counter-language, English law - are not prompt-engineering problems. They are product problems. They are why BeforeYouSign exists.
BeforeYouSign reads your contract against two decades of clause-by-clause negotiation, in plain English, anchored in English law. Every flag carries a severity rating. Every risk comes with the exact words to send back. Every gap gets surfaced. It is what the prompt was always going to leave behind.
If the contract in front of you is worth getting right, that is the difference.
One contract clause, one commercial risk, every month — written for the people who actually sign contracts.
Louise
BeforeYouSign Ltd. All rights reserved. | beforeyousign.uk