TL;DR
On this page
You are about to start a $5,000 project. The client wants something in writing, so you open ChatGPT and type "write me a contract." Ten seconds later you have a clean, professional-looking document with numbered sections and legal-sounding language. It looks finished.
This walkthrough is part of the complete guide to generating client documents with AI.
The draft reads well enough that most people send it without checking the three clauses that decide whether the contract actually protects them. ChatGPT is good at producing the shape of a contract. It is unreliable at the parts that matter when a project goes sideways: the scope boundary, the payment terms, and who owns the work before you have been paid.
This post gives you a prompt that produces a strong first draft, then walks through the three clauses you need to rewrite before you send it.
The prompt that gets a usable draft
Paste this into ChatGPT, Claude, or Gemini. Use placeholders for client names and figures rather than real details (more on why below), then fill the real values in afterward.
You are a contract lawyer who works with independent contractors. Draft
a service contract for the following engagement:
- Contractor: [your name / business name], based in [your country/state]
- Client: [Client A], based in [client country/state]
- Project: [one-line description, e.g. "design and build a 5-page website"]
- Deliverables: [list each concrete deliverable]
- Total fee: [amount + currency]
- Payment schedule: [e.g. 50% deposit, 50% on final delivery]
- Timeline: [start date, milestones, final due date]
- Revision rounds included: [e.g. 2]
Requirements for the draft:
1. Itemize the deliverables in a numbered scope section, and state that
anything not listed is out of scope and billed separately.
2. Include payment terms with the deposit, a Net 14 term on the final
invoice, and a 1.5% per month late fee on overdue balances.
3. Include an intellectual-property clause stating that ownership of the
work transfers to the Client only upon payment in full, and that the
Contractor retains copyright until then.
4. Cap included revisions and price additional rounds.
5. Include a kill fee / cancellation clause.
6. Use plain English. Flag any clause that depends on my jurisdiction
and where I should consult a local lawyer.
Output the full contract, then list the placeholders I still need to fill.
The requirements list is doing the heavy lifting. A bare "write me a contract" prompt produces a generic template; spelling out the scope boundary, the payment mechanics, and the IP condition forces the model to include the protective language instead of leaving it out. Even with explicit instructions, the model is inconsistent on all three. Check each one below before you send.
Here is the quick version before the detail:
| Clause | What ChatGPT usually writes | Rewrite it to |
|---|---|---|
| Scope | A list of deliverables, no boundary | The same list, plus a line that anything not listed is out of scope and billed separately |
| Payment | Net 30, no deposit, no late fee | A deposit, a Net 14 final term, and a 1.5% per month late fee |
| Intellectual property | Ownership transfers on completion or delivery | Ownership transfers only on payment in full, and you keep copyright until then |
Clause 1: the scope section needs a boundary, not just a list
ChatGPT will list your deliverables neatly. What it usually fails to add, unless you push, is the single sentence that stops scope creep: anything not listed here is out of scope and billed separately.
That sentence matters more than the list. The list says what you will do. The boundary says what happens when the client asks for the thing that was never agreed, which is where most unpaid work comes from. It is not a small risk: of the people who get stiffed by a client, 37% blame vague or poorly written contracts.
Read the scope section ChatGPT produced and check for three things: every deliverable is concrete (not "design assets" but "three logo concepts, one final logo in SVG and PNG"), the number of items is fixed, and the out-of-scope sentence is present. If any deliverable is fuzzy, tighten it. If the boundary sentence is missing, add it. The scope-of-work structure and the guide to handling scope creep go deeper on what a tight scope section looks like.
Clause 2: the payment terms are almost always too soft
Left to its defaults, ChatGPT writes payment terms that favor the client: Net 30, no deposit, no late fee. That is the arrangement that turns into a collection problem. About 29% of invoices are paid at least a day late, and soft terms only make that worse.
Rewrite the payment section to add three things the draft probably omits. First, a deposit, commonly 30 to 50 percent, before work begins, so you are never fully exposed. Second, a short final term: Net 14 rather than Net 30, because a shorter clock gets you paid sooner. Third, a late fee, typically 1.5% per month on overdue balances, which gives you both a deterrent and a documented basis for collection. The contract essentials guide and the post on late-paying clients cover the mechanics.
Clause 3: ownership should transfer on payment, not on delivery
This is the one that costs people the most and the one ChatGPT gets wrong most often. Ask for an IP clause and the model frequently writes that ownership transfers to the client "on completion" or "on delivery." That sounds reasonable. It also means that the moment you hand over the work, you have given away your only leverage, even if you have not been paid.
The clause should condition the transfer on payment in full: you retain copyright until the final invoice is paid, and ownership transfers only then. Until that condition is met, the work is yours, and the client's use of it without paying is unauthorized. That single change is what keeps a non-paying client from walking away with your work for free.
Check the IP section of the draft and confirm it says "upon payment in full" and not "upon completion" or "upon delivery." If it does not, rewrite it.
Read it once, because the AI can invent things
The three fixes above are the predictable gaps. The unpredictable risk is that the model occasionally states something with total confidence that is simply wrong. This is not hypothetical for legal text: in Mata v. Avianca (2023), lawyers were fined $5,000 after submitting a brief that ChatGPT had filled with fabricated cases that looked real.
You are not filing in federal court, but the lesson holds. Read the whole contract once before you send it, and treat any clause that cites a specific law or names a legal doctrine with suspicion unless you can confirm it. The practitioners at Pactly put the standard plainly:
Any text generated by an AI, no matter how sophisticated, must be treated as a draft and requires final legal oversight.
Source: Pactly, "Can I Use ChatGPT to Write a Legal Contract?"
They also note where the risk lands if the AI gets it wrong:
If the AI generates an error that leads to a financial dispute, the entire financial and legal risk falls squarely on your organization and the employee who used the tool.
Source: Pactly, "Can I Use ChatGPT to Write a Legal Contract?"
For a routine project, the prompt plus the three fixes plus one careful read is enough. For a high-value or unusual engagement, have a local lawyer review it once, then reuse that reviewed version as your template.
pro tip
Generate the draft with placeholders, not real client data. Consumer ChatGPT may retain what you paste, and one analysis found that 11% of what employees paste into ChatGPT is confidential. Use "Client A" and "[amount]" while drafting, then fill in the real details privately once the document is on your own machine.
Or skip the prompt entirely
The loop is fine once. Repeat it across every project and the tax adds up: paste the prompt, fix the same three clauses, copy the result into Word, strip the formatting, add your details, export a PDF. Every project, start to finish.
If you would rather not run that loop on repeat, FreelanceDesk builds contracts with the scope boundary, the deposit-plus-late-fee payment terms, and the transfer-on-payment IP clause already in place. You fill in the project details and export a clean PDF, and the document never leaves your browser, so there is no question about pasting client data anywhere. It is free to use.