How to Draft a Contract With AI (Without Signing Away the Wrong Things)

Last spring a graphic designer I'll call Priya sent me a services agreement she had produced with a chatbot in about four minutes. It was eleven pages long, beautifully formatted, and full of words like "heretofore." It also gave her client ownership of every sketch, mood board and rejected concept she had ever shown them, from the moment she showed it, whether or not they paid. She had been about to send it.

I'm a solo transactional lawyer. Most of my clients are small businesses and freelancers: designers, developers, photographers, consultants, the occasional bakery. I use AI contract drafting every week, and I'm not here to tell you to stop. I'm here to show you how I actually do it, so the draft you end up with protects you instead of just looking like it does.

This guide follows one agreement from start to finish: a freelance design services agreement between a brand designer and a small business client. You'll see the brief I give the AI, what comes back, the three edits I make to every single draft, and the checklist I run before anyone signs.

The short version: AI is very good at producing a clean, well-organised first draft from a clear brief. It is not good at knowing what your deal actually is, which law applies, or which risks you can live with. Give it a detailed brief, treat the output as a starting point, and always rework ownership, money and liability by hand before you send it.

Why AI contract drafting works (and where it quietly fails)

A language model has read an enormous number of contracts. That makes it excellent at structure: it knows a services agreement needs a scope section, payment terms, a term and termination clause, confidentiality, and the usual closing provisions. It will rarely forget to include a severability clause. It will produce consistent defined terms, number the sections properly, and do it in less time than it takes me to open my precedent folder.

What it doesn't have is your deal. It doesn't know that Priya wants to reuse her custom illustration style for other clients, that her client is a two-person coffee roaster rather than a multinational, or that Priya lives in Ontario while the client is in Washington State. Unless you tell it, it fills those gaps with the most statistically common answer. And the most common answer in the training data is usually a contract written by the bigger party's lawyer, for the bigger party.

That's the core thing to understand about using AI to write a contract: the tool is neutral about drafting quality but not neutral about whose side it takes. Left alone, it tends to drift toward whoever usually pays for contracts to be written. If you're the freelancer, that's not you.

Good tasks for AI vs tasks to keep human

Before we start the walkthrough, here is how I split the work. I've refined this over a couple of years of using these tools on real client files.

Notice the pattern. The left column is about language and structure. The right column is about judgment and consequences. Whenever a task has a consequence you'd have to live with if the AI got it wrong, it belongs on the right.

Step 1: Write a brief the AI can't misread

Most bad AI drafts come from thin prompts. "Write me a freelance design contract" gives the model almost nothing, so it invents everything. A useful brief reads more like the intake notes I'd take in a first client meeting.

Here's roughly what I typed for Priya's agreement:

Draft a design services agreement. My client is the designer, Priya Shah, a sole proprietor based in Toronto, Ontario. The customer is Ridgeline Coffee LLC, a small coffee roaster in Spokane, Washington. Scope: a brand identity package consisting of a primary logo, two logo variations, a colour palette, a typography guide and a 12-page brand guidelines PDF. Fee: USD 6,500, paid 40% on signing, 30% on approval of the logo concept, 30% on final delivery. Two rounds of revisions included per deliverable; extra rounds billed at USD 95 per hour. Timeline: eight weeks from receipt of the deposit. Ownership of the final approved deliverables should transfer to the client only once all fees are paid. Priya keeps ownership of unused concepts, her sketches and working files, and any pre-existing fonts, illustrations or templates she brings in. She wants the right to show the work in her portfolio. Either party may end the agreement with 14 days' written notice; if the client ends it, they pay for work done to date plus the next payment milestone as a cancellation fee. Write in plain English, short sentences, no archaic legal words. Leave governing law as a placeholder for me to decide.

A few things to notice about that brief:

• It names who I'm acting for. "My client is the designer" tells the model whose interests to protect. This one line changes the whole draft.
• It describes the deliverables concretely. Vague scope is the single biggest source of freelance disputes I see.
• It says what I want for the tricky parts (ownership, cancellation, revisions) instead of leaving them to the model.
• It asks for plain English. Courts don't award points for "witnesseth," and clients actually read plain contracts.
• It deliberately parks governing law. I'll explain why in a later section.

If you're using a tool built for legal work rather than a general chatbot, you can often attach background material, too: an old contract you liked, the client's purchase order, or the proposal you already sent. In LegesGPT, I'll start from the design services template, paste the brief into the AI drafting workspace, and let it adapt the template to these facts. Starting from a template that a legal tool already knows rather than a blank page tends to give a tighter first draft.

Step 2: Read the first draft like an opposing lawyer would

The first draft came back in under a minute. It was genuinely decent. Eight sections, sensible headings, the payment schedule exactly as specified, a revisions clause, a confidentiality clause, and the standard closing provisions. If I'd been in a hurry, I could have convinced myself it was done.

So I read it the way I'd read a draft that landed on my desk from the other side. I'm not looking for typos at this stage. I'm asking three questions of every clause:

1. What does this actually make someone do?
2. What happens if they don't?
3. Who does this favour when something goes wrong?

That pass turned up a handful of problems. The revisions clause said "reasonable revisions" in one place and "two rounds" in another. The confidentiality clause had no end date. The termination section required "written notice" but the notices clause only allowed delivery by post, which nobody does anymore. And there was a warranty that the work would be "fit for the Client's purposes," which is a promise no designer should make about something as subjective as a logo.

None of those were catastrophic, and I fixed them quickly. AI can even help here: I'll often ask the tool to "list every place this draft uses an undefined standard like 'reasonable' or 'promptly'" or "find any inconsistency between the revisions clause and the payment schedule." It's good at that kind of internal cross-checking.

But the three edits that really matter are the ones I'd make even if the draft looked perfect.

The three edits I make to every AI-drafted contract

Edit 1: Pin down who owns what, and when ownership moves

Even with my brief, the draft's intellectual property clause opened with "All work product created under this Agreement shall be the sole property of the Client." That's the clause Priya's chatbot version had, and it's the default you'll see over and over.

Here's why it matters. In the United States, a freelancer's work is generally not a "work made for hire" just because a contract says so. Under the Copyright Act, commissioned work from an independent contractor only qualifies if it falls into one of nine specific categories listed in 17 U.S.C. § 101 and the parties agree to it in writing. A logo isn't on that list. So for the client to own it, the designer has to assign the copyright. In Canada and the UK the starting point is similar: the author owns the copyright unless they're an employee, and a transfer needs to be in writing.

That puts the designer in a strong position, and the AI's default clause throws it away. My rewrite does four things:

• Ownership of the final approved deliverables transfers on full payment, not on creation. Until then, the client gets a licence to review and test the work.
• Unused concepts, sketches and working files stay with the designer. The client is paying for the chosen identity, not every idea along the way.
• Pre-existing materials are carved out. If Priya uses her own custom brush set or a licensed font, the client gets a licence to use those as part of the brand, not ownership.
• A portfolio right is spelled out, with a note that she won't show the work publicly before the client launches it.

The AI wrote a version of most of this once I pointed it at the problem. But it took a human to notice that the opening sentence quietly contradicted everything I'd asked for in the brief.

Edit 2: Make the money mechanics enforceable

The draft had the payment schedule right. What it lacked was everything that happens when payment doesn't go to plan, and that's where freelancers lose money.

I tightened four things:

• Invoice timing and due dates. "Payment on delivery" became "Designer will invoice on each milestone; invoices are due within 14 days."
• What a milestone approval means. A client can stall forever by never "approving" the logo concept. I added deemed approval: if the client doesn't respond within ten business days, the milestone counts as approved.
• Late payment consequences. Interest on overdue amounts at a stated rate, and the right to pause work until the account is current. Check the rate against local rules; some places cap contractual interest, and some give businesses a statutory rate anyway (the UK's Late Payment of Commercial Debts (Interest) Act 1998 is the best-known example).
• Out-of-scope work. The draft said extra revisions were billable but didn't say how anyone would agree to them. I added a short change request process: Priya sends a written estimate, the client approves it in writing, and the fee is added to the next invoice.

If you only make one of these changes, make it the deemed approval clause. I've watched more freelance projects die in the "waiting for feedback" stage than in any actual disagreement.

Edit 3: Decide how much risk each side carries

This is the edit people skip because the clauses look like boilerplate. They aren't.

The AI draft included a one-way indemnity in which Priya promised to "defend, indemnify and hold harmless the Client from any and all claims arising out of or relating to the Services." Read that slowly. Any claim, related in any way to her work, with no cap and no requirement that she was actually at fault. For a solo designer charging USD 6,500, that's a lopsided promise.

My version:

• Narrows the indemnity to claims that the designer's original work infringes someone else's intellectual property, and excludes anything the client supplied (their photos, their old logo, their copy).
• Adds a reciprocal indemnity for materials the client provides.
• Caps total liability at the fees paid under the agreement, with the usual exceptions for things like fraud.
• Excludes indirect and consequential losses such as lost profits from a delayed product launch.
• Removes the "fit for purpose" warranty and replaces it with a promise that the work will be original and delivered with reasonable skill and care.

Whether a cap like this holds up depends on where you are. Some legal systems test limitation clauses for reasonableness, especially in standard-form business contracts (the UK's Unfair Contract Terms Act 1977 does exactly this), and consumer contracts face stricter rules almost everywhere. This is precisely the kind of question to answer with sources rather than instinct. When I'm unsure, I'll ask LegesGPT the specific question ("is a liability cap equal to fees paid enforceable in a B2B services contract in Washington?") and then read the cases or statutes it cites before I rely on the answer.

Governing law and jurisdiction: where AI drafts go wrong most often

Remember that I told the AI to leave governing law as a placeholder. Here's why.

Ask a general-purpose model for a contract with no location and it will often pick Delaware, New York or California, because those states dominate the contracts it learned from. Ask it for a contract "under Ontario law" and you may still get a draft full of US concepts. I've seen AI drafts that choose English law and then refer to "punitive damages" and "the Uniform Commercial Code" in the same document. The model is matching patterns, not applying a legal system.

These are the pitfalls I check for on every draft:

Governing law and forum are two different things. Governing law decides which rules interpret the contract. The jurisdiction or forum clause decides where a dispute gets heard. A contract can be governed by Ontario law and still allow the client to sue in Washington. Make both choices deliberately, and make sure they make sense together.

The default isn't neutral. For Priya, choosing Ontario law and Ontario courts is convenient for her and expensive for a Spokane coffee roaster to fight. That is a bargaining chip, but it can also slow down signing. Sometimes the right answer is the other party's home law plus a cheap dispute process like mediation or online arbitration. That's a business decision, and it's hers to make.

Some rules apply whatever you choose. Mandatory local protections don't disappear because the contract picks a different law. In the EU, the Rome I Regulation stops a business from using a choice-of-law clause to strip consumers of the protections of the country where they live. California's Business and Professions Code section 16600 makes most non-compete clauses void, whatever the contract says. If you're dealing with consumers or with restrictive covenants, check the rules where the other side is.

Local freelancer laws may dictate the contract itself. New York City's Freelance Isn't Free Act, and since 2024 a statewide New York version, require a written contract with specific content for many freelance engagements and set payment deadlines. A generic AI draft won't know to include those terms unless you tell it the client or freelancer is there.

Worker classification sits on top of everything. If the "freelancer" is really working like an employee, the contract's label won't save anyone. States like California apply strict tests for this. No clause fixes a relationship that doesn't match it, so this is a conversation, not a drafting task.

Placeholders get signed. The single most common error I see in AI-drafted contracts that reach me is a literal "[State]" or "the laws of the jurisdiction in which the Client resides" left in the final version. Search the document for brackets before anything goes out.

My practical rule: decide the governing law and forum yourself, then tell the AI exactly what you've decided and ask it to review the whole draft for consistency with that choice. Then check the output against a source you trust, not against the model's own confidence.

Step 3: Sign it properly

Once the edits were in and Priya had read the final version (out loud, which I recommend for anything you'll sign), it was ready to go. The last mile matters more than people think. I've seen more than one well-drafted agreement fail because the signed copy was an older version, or because one party "signed" by replying "looks good" to an email.

Electronic signatures are valid for most commercial contracts in the US under the federal ESIGN Act and the state-level Uniform Electronic Transactions Act, and in Canada under provincial electronic commerce laws, with broadly similar rules in the EU under eIDAS. A few documents, such as some real estate and family law papers, still have special formalities, so check if you're outside ordinary commercial work.

For this agreement, we sent it for signature straight from LegesGPT, which meant the version Priya reviewed was exactly the version Ridgeline signed, and both sides got a completed copy with a record of when each party signed. Keeping drafting, review and e-signature in one place is the main reason I use it for small-client work: there's no exporting, emailing a PDF, and hoping nobody edited it in between. It runs in the browser, so there's nothing to install, and the plans start at USD 19.99 a month with a USD 1 three-day trial if you want to test it on a real agreement first.

The pre-signature checklist

This is the list I run on every AI-drafted agreement before it leaves my desk. It takes about fifteen minutes.

The deal

• Are both parties named with full legal names, entity types and addresses?
• Does the scope list every deliverable specifically enough that a stranger could tell whether it was delivered?
• Do the fee, the payment milestones and the invoice due dates all match each other?
• Is there a process for changes and extra work?

Ownership and confidentiality

• When does ownership transfer, and is it tied to payment?
• Are pre-existing materials and unused work carved out?
• Does the confidentiality clause have a sensible end date and normal exceptions?

Risk

• Is any indemnity limited to things the indemnifying party controls?
• Is there a liability cap, and is it realistic for the size of the deal?
• Have you removed promises that can't be kept, like "fit for purpose" or "error-free"?

Ending it

• Can either party exit, and what's owed if they do?
• What happens to work in progress and to files on termination?

Law and location

• Have you chosen governing law and forum yourself, and do they agree with each other?
• Does any local rule (consumer law, freelancer law, non-compete limits) override what the contract says?
• Are there any leftover placeholders, brackets or "[insert]" notes?

Signing

• Is the version being signed the version both sides reviewed?
• Does the signature block match the party names in the opening paragraph?

If you can't answer one of these confidently, that's the clause to take to a lawyer. It usually costs far less to have one clause checked than to have a whole agreement drafted.

Choosing an AI contract drafting tool

You can do everything in this guide with a general chatbot, but you'll do more of the checking yourself. The tools built for legal work tend to differ in three ways: they start from vetted templates rather than a blank prompt, they can review a draft and flag risky clauses rather than just generate text, and the better ones cite the legal sources behind their answers so you can verify them.

If you are weighing up options, the side-by-side of AI contract drafting tools for 2026 covers the main contenders by price and workflow, and the guide to AI contract review tools is useful if most of your work is checking contracts other people send you. Lawyers who want a wider view of what's out there can start with our list of AI tools for lawyers. And if you only need occasional one-off agreements, the comparison of AI contract generators leans toward simpler, template-led options.

A few things I'd check in any trial, whichever tool you're looking at:

• Does it let you say whose side you're on, and does the draft actually change when you do?
• Can it explain a clause in plain English and point to the law behind that explanation?
• Can it review its own draft and flag one-sided terms?
• Does it handle signatures, or will you be juggling PDFs at the end?
• What happens to the documents you upload? Read the privacy terms before you put client information in.

What I'd tell any freelancer starting out

Use AI. It will save you hours, and a clear AI-drafted contract is much better than the handshake-and-email arrangement most freelancers start with. But own the three decisions that matter: who owns the work and when, how and when you get paid, and how much risk you're willing to carry. Make the governing law choice yourself. Then read the whole thing once more before you send it.

Priya's final agreement was five pages instead of eleven. Ridgeline signed it in two days, paid the deposit the same week, and the only question they asked was whether they could have the brand guidelines as an editable file too. That's what a good contract feels like: short, clear, and boring right up until the moment you need it.

---

Frequently Asked Questions

Can I legally use a contract drafted by AI?

Yes. A contract is valid because of what the parties agree to and how they sign it, not because of who or what typed the words. An AI-drafted agreement is as enforceable as any other, provided its terms are clear, lawful where you are, and signed properly. The risk is not the AI itself but terms in the draft that don't match your deal or your jurisdiction.

What is the best way to prompt AI to draft a contract?

Write the prompt like a set of client intake notes. Say whose side the draft should protect, name both parties and where they are based, list the deliverables, fees, payment milestones and timeline, and state what you want on ownership, termination and revisions. Ask for plain English and leave governing law as a placeholder until you have decided it yourself.

Which parts of an AI-drafted contract should I always rewrite?

Rework three areas on every draft: intellectual property ownership and when it transfers, the payment mechanics including late payment and approval deadlines, and the risk allocation clauses such as indemnities, warranties and liability caps. These are where AI drafts most often default to terms that favour the larger party.

Is AI contract drafting good enough to replace a lawyer?

For simple, low-value agreements with a clear brief and a careful review, AI drafting can get you most of the way there. It cannot decide what your deal should be, judge how much risk you can carry, or confirm a clause is enforceable in your jurisdiction. For high-value deals, employment relationships, consumer contracts or anything cross-border, have a lawyer review at least the key clauses.

Why do AI tools get governing law wrong?

General AI models pick the law that appears most often in the contracts they learned from, which is often Delaware, New York or California, and they can mix concepts from different legal systems in one document. Decide the governing law and dispute forum yourself, tell the tool exactly what you chose, and then check the draft for consistency with that choice.

Are electronic signatures valid on AI-drafted contracts?

For most commercial contracts, yes. In the US the ESIGN Act and state electronic transaction laws treat e-signatures as valid, and Canada and the EU have similar rules. Some documents, such as certain real estate or family law papers, still need special formalities, so check before using e-signature outside ordinary business agreements.

How long does it take to draft a contract with AI?

The first draft takes a minute or two. Writing a proper brief takes about ten minutes, and the human review, the three key edits and a final checklist typically take another thirty to sixty minutes for a short services agreement. That is still far faster than drafting from a blank page.

Which AI contract drafting tool should a freelancer or small business use?

A legal-specific tool is usually a better fit than a general chatbot because it starts from vetted templates, can review drafts for risky terms, and cites sources you can check. LegesGPT covers drafting, review, legal questions with cited sources and e-signature in one web app, with plans from 19.99 dollars a month and a 1 dollar three-day trial.