How to Prompt AI to Draft a Statement of Work

If you want to know how to prompt AI to draft a statement of work, the short answer is: don't ask for a generic SOW. Ask for one built around how AI-assisted projects actually go wrong, which differs from traditional dev projects. A client can ask you to "just try it a different way" five times before lunch, and each try is a new prompt, a new output, and a new argument about what was promised.

Below is a prompt template for a first-draft SOW, plus three clauses that matter more when AI is doing part of the work: change control, ownership of AI-generated output, and revision limits.

Why a Standard SOW Doesn't Cover AI Work

A normal statement of work assumes a rough link between effort and output: you scope it, build it, and revisions happen in discrete rounds because each round costs real time.

AI collapses that cost. Generating a new landing page draft, logo concept, or set of onboarding emails takes a prompt and seconds, not hours. That's the value proposition, and also why unscoped AI work turns into an unpaid loop: if trying again is nearly free for you, the client has no signal for when to stop asking.

There's a second gap: traditional SOWs assume the freelancer wrote the deliverable by hand. When a client's own prompts and feedback shaped an AI output, "who owns this" stops being obvious, and standard work-for-hire language wasn't written with that in mind.

How to Prompt AI to Draft a Statement of Work

Use this as your starting prompt. It's built to produce a statement of work template for an AI project, not a generic freelance contract, and it forces the model to include the three clauses this kind of engagement actually needs.

Copy and adapt the bracketed parts:

"Draft a statement of work for a freelance/agency engagement where I will use AI tools to help produce the deliverables. Project: [describe the work, e.g., 'a 10-page marketing website with AI-generated copy and images']. Client: [name]. Timeline: [start and end dates]. Fee: [flat fee or milestones].

Structure it with: Scope of Work, Deliverables, Timeline and Milestones, Fees and Payment Schedule, Change Control, Ownership of AI-Generated Output, Revision Limits, Client Responsibilities, Termination.

For Change Control: define an in-scope revision versus a new request, require written approval before scope changes, and note that regenerating or reworking AI output beyond the agreed revision count is billed separately.

For Ownership: state that the client owns final approved deliverables upon full payment, that I retain rights to reuse general prompts and workflows, and that neither party warrants the AI output is free of third-party IP claims given current legal uncertainty around AI training data.

For Revision Limits: specify a fixed number of rounds per deliverable, define what one round includes, and state the rate for additional revisions.

Keep the tone plain and specific, avoid vague terms like 'reasonable' without defining them, and flag any placeholder I still need to fill in."

Feed the output back with real details before sending anything to a client, and read every clause it produces. AI drafting tools are good at structure, not at knowing your actual risk tolerance.

Change Control: The Clause AI Projects Need Most

This is the clause most freelancers skip, and it costs the most on AI projects specifically. Because iteration is fast, "small tweaks" accumulate into a materially different project without ever triggering the conversation a slower process would force.

Good SOW change control for an AI project does three things: defines scope precisely enough that both sides can tell when something crosses it, requires written approval before you act on a change, and states what happens to the timeline and fee once a change is approved.

Example wording:

"Change Control. Any request that alters the deliverables, prompts, or approach described in Section [X] beyond correcting errors in approved output is a Change Request. Change Requests must be submitted in writing. Contractor will provide a written estimate of additional fees and timeline impact within [2] business days. Work begins only after Client approves the estimate in writing. Requests to regenerate AI output using materially different prompts, instructions, or source content than what was approved in the current round are Change Requests, not revisions."

That last sentence closes the loophole where a client claims "just try a different prompt" is a minor tweak, not a new request.

Ownership of AI-Generated Code and Assets

Standard work-for-hire language assumes a human author transferring rights they hold. AI output complicates that, which is why an AI-assisted code ownership clause has to say more than a normal freelance contract does.

• The deliverable. Most freelancers are comfortable assigning full ownership of the final, paid deliverable to the client, whether a human or an AI tool produced the underlying draft.

• The process. You likely don't want to sign away reusable prompts or internal workflows just because you used them on this project. Separate the deliverable (client owns it) from your methods (you keep them).

• The warranty. Neither side can fully guarantee AI-generated code or content is free of third-party IP claims, since the legal treatment of AI training data is unsettled and varies by jurisdiction. Say so instead of promising something you can't back.

A working version:

"Ownership. Upon full payment, Client owns all right, title, and interest in the final, approved deliverables, including content and code produced with AI assistance. Contractor retains ownership of its prompts, templates, and general AI-assisted workflows, and may reuse them for other clients provided no Client-confidential material is included. Because the IP status of AI-generated output is unsettled in many jurisdictions, Contractor makes no warranty that AI-assisted deliverables are free of third-party claims, and Client accepts this risk as a condition of AI-assisted services under this SOW."

If a client needs a stronger IP warranty for one asset, like a logo going into a trademark filing, scope it out and produce it without AI assistance rather than letting a blanket clause cover something that needs its own terms.

Revision Limits: Where Unpaid Work Hides

"Just ask the AI again" sounds harmless until you've generated the ninth variant of a hero image for a client who keeps saying "close, try again." Each attempt is cheap for you to produce, so clients under-price how much time it consumes: prompting, reviewing, adjusting, re-explaining brand guidelines the model keeps missing.

Set a specific count per deliverable, not a vague "reasonable number of revisions." Define what counts as one round, and set the rate for revisions past the limit before the client hits it, not after.

Example wording:

"Revision Limits. Each deliverable in Section [X] includes up to [2] revision rounds. One revision round consists of Client providing consolidated written feedback and Contractor delivering one updated set of options in response. Requests submitted individually and reviewed separately count as separate rounds. Revisions beyond the included rounds are billed at [$X]/hour, invoiced before work begins."

The "consolidated feedback" line matters specifically for AI work. Without it, a client can send five one-line notes over a week, each triggering a new generation cycle, and argue that was one round.

Reviewing What the AI Drafts

Treat the AI's first draft as a scaffold, not a contract. Check that dollar amounts and dates are real, that the revision and change control numbers are ones you'll actually enforce rather than a cap you know you'll cave on, and that the ownership clause doesn't promise a warranty you can't stand behind.

This is general business practice guidance, not legal advice. A lawyer should review any contract before you use it on a real engagement, especially once fees or IP stakes get large.

Pricing and contract structure are covered more broadly in our AI monetization strategies guide. If the engagement includes ongoing support after delivery, writing an SLA for an AI product covers the clauses that matter once the SOW is signed.

The revision-limits problem above is closely tied to how you price the engagement in the first place: see fixed-scope vs hourly pricing for an AI project for that decision. And if the SOW includes a features list, prompting AI to write user stories is a fast way to turn it into something a developer can actually scope.

FAQ

What should a statement of work for an AI project include that a normal SOW doesn't?

A change control clause that treats new prompts or reworked approaches as scope changes, an ownership clause addressing AI-generated code specifically, and a fixed revision limit with a rate for going over it. A generic template usually assumes none of these are separate risks.

Can I just have AI write my entire statement of work?

You can use AI for a strong first draft quickly, but you still need to fill in real numbers, adjust clauses to your risk tolerance, and read the whole thing before sending it. AI is good at structure, not at knowing what you're willing to eat the cost of.

Who owns code or content generated by AI during a client project?

In most arrangements, the client owns the final paid deliverable regardless of whether AI assisted in producing it, while the freelancer keeps their own prompts and workflows. Because the legal status of AI-generated output is unsettled in many places, state it plainly in the SOW rather than assuming standard language covers it.

How many revision rounds should I include in an AI-assisted project?

There's no universal number, but two rounds per deliverable is a common starting point, since each round can include several AI-generated options. Define what counts as one round, and set the rate for extra rounds up front so it isn't a mid-project negotiation.

What counts as a change request versus a normal revision in an AI project?

A revision refines a deliverable within the approach already approved, like adjusting tone in an approved draft. A change request alters the scope or approach itself, like a different content strategy or a new set of prompts. AI makes the two cheap to blur, so the SOW should define the line rather than leave it to interpretation.