The first call used to open with a brief, or at least a description of the problem. It now opens with a screenshot. The client has typed a sentence into a prompt-to-mockup tool and received something that looks like a landing page or a logo on a coffee cup. The AI mockup arrives with a line underneath it that every working designer has now read: "we basically have the design, we just need someone to clean it up."
That sentence sounds like a smaller job. It is a spec request with the price already anchored at zero, and the anchoring happened before you were in the room.
The short answer. An AI mockup is a prompt-to-image or prompt-to-layout render that looks finished and contains none of the decisions a finished design contains. Treat it as a mood reference and price the discovery the render skipped. Scope the system behind the screen, because the screen is the cheap part.
What an AI mockup actually is
There are two kinds, and they get confused because both look done. A prompt-to-image render is a picture: pixels arranged to resemble an interface or a package. The button is not a button. A prompt-to-layout render is closer to a working file: a mockup generator drops frames, text boxes and placeholder components into a canvas, sometimes with code behind it. It can be edited, which makes it feel further along than it is.
Both are the surface of a design, produced without the design. The tool has learned what finished work looks like. It has not learned what the client sells, who buys it, or what the legal team will strike, because the person typing the prompt does not know those things yet either.
The definition on this site has not moved: spec work is a deliverable produced before there is an agreement to pay. A generated render is a deliverable-shaped object with no brief behind it. It does not change what spec work is. It changes who made the first move and how convincing that move looks.
Why the client brings one to the first call
Most clients who arrive with an AI design mockup are not trying to underpay anyone. The tool was free, it took ninety seconds, and it answered the question that has always made clients nervous about hiring a designer: what will this look like? Having made something, they feel phase one is done.
What the render actually does is fix a price before the conversation starts. The client has already seen the design at no cost, so the remaining work registers as polish, and polish has a polish-sized budget. This is the anchoring effect that contests and free pitches have always relied on, delivered by software instead of a crowd of unpaid entrants. The wider shift is covered in the free pitch, now AI-generated; this article stays on the mockup itself.
"Just clean this up" is a spec request at zero
Take the request at face value. To clean up a generated pricing page you would need to decide how many plans exist, what the copy says, what happens on a narrow screen, and what the error states look like. That is the job. The render drew a picture of a page where those decisions appear to have been made.
So "clean this up" means "do the design, but bill it as a touch-up." The structure is identical to a contest entry. You produce the finished, usable thing. The client then decides whether to pay for it, take the tidied render to a cheaper vendor, or ship it as is. If the engagement never becomes a project, you have handed over the expensive part for a fee sized to the cheap part, which is why speculation hurts even when nobody intended harm. A hypothetical, for scale: a client sends a generated five-screen onboarding flow and asks for a quick pass at a flat 400. Making those screens real is a week at any defensible rate. The 400 was a price for the picture.
What a generated mockup lacks
Every item on this list is expensive.
- A brief. Nobody wrote down the problem, the audience or the measure of success. The prompt is a wish.
- Constraints. No brand guidelines, no legal copy, no technical limits, no existing component library.
- Decisions. Every element is there because the model has seen it before, and nothing is there because someone chose it against an alternative.
- Licensable provenance. No one can state who made the file, what it was assembled from, or whether it resembles something registered by somebody else.
- Accessibility. Generated interfaces routinely fail contrast, focus order and target size, because the tool optimises for the look of a screenshot. The W3C's WCAG overview lists the checks a render has never been near.
- A system. One screen is not a product. There are no tokens, no responsive rules, nothing for the eleventh screen the client has not thought of yet.
| Attribute | Generated mockup | Designed mockup |
|---|---|---|
| Origin of the layout | Resemblance to training data | Choices made against a brief |
| Content | Placeholder or invented copy | Real copy, with edge cases |
| States and breakpoints | One static view | Error, empty, loading; mobile to desktop |
| Accessibility | Untested, usually failing | Checked against WCAG before handoff |
| Underlying system | None; a single frame | Components, tokens, rules for new screens |
| Authorship record | Prompt text, if anyone saved it | Version history, process files, dated exports |
| Copyright position | Generated material is not registrable | Human authorship, registrable and licensable |
How to price the work when the client has an AI mockup
Do not argue with the render. Arguing tells the client they wasted their time, and they will defend what they made. Reposition it, and price what it skipped.
Charge a discovery fee first
Every engagement that begins with a generated file should begin with paid discovery: a fixed fee for the brief, the constraints and an audit of what in the render survives contact with reality. Credit it against the project if it proceeds. The rates, contracts and scope pillar covers how to size it.
Treat the render as a mood reference
A mood reference is useful: it tells you what the client responds to. File it with the other references and say so in the proposal: "the generated concept supplied on 12 September is treated as reference material and is not a deliverable or a specification." That sentence removes the anchor without insulting anyone.
Scope the system rather than the screen
Quote for the component set, the responsive rules, the content model and the handoff, and list the screens as outputs of that system. Once the proposal describes a system, the screen the client brought looks like one of forty artefacts instead of most of the job. If you need to state how generated material will and will not be used, the AI disclosure clause belongs in the same proposal.
When the email says "we already have a mockup, we just need it finished," this is the reply:
Thanks for sending the concept. I can see what you are going for and it is a useful reference. To be straightforward about scope: a generated mockup gives us a look, and the work I am pricing is the part underneath it, the decisions, the content, the states and the system the screens come from, so the result can be built, maintained and owned. The first step is a paid discovery phase at [fee], credited to the project if we proceed. I will send a scope for that today.
Send it on the first reply, before any pricing conversation, and attach nothing.
The provenance problem: nobody can say who made it
This part of the AI mockup problem outlasts the pricing conversation. Ask who made the render and the honest answer is: a client typed a sentence, and a tool assembled an image from a training set nobody involved can itemise. Ask whether it can be registered and the answer, in the United States, is no, at least for the generated portion.
The U.S. Copyright Office's 2023 registration guidance states that material produced by a generative tool without human creative control is not the product of human authorship and will not be registered, and that applicants must identify and disclaim such content in a work that contains it. A client who "owns" a render owns a file with no registrable core. If you build on it, the registrable part is your contribution, and only if you can show what that contribution was.
That turns your process files into the asset. Version history and dated exports separate the design you made from the picture the client brought; how to prove a design is yours covers what to keep. Where your tools support it, Content Credentials can attach a signed record of edits to the exported file; the C2PA specification defines the format, and it can record that a source asset was generated as well as that a human edited it afterward. If the render is a logo, the trademark side is covered in who owns an AI-generated logo.
None of this is a reason to refuse clients who arrive with a render. It is a reason to write down, in the contract, what was supplied, what was made, and who can claim which.
Frequently asked questions
Is it spec work if the client made the mockup?
Yes, if the ask is for you to produce finished work before a fee is agreed. Who made the starting file is irrelevant; the test is whether a usable deliverable leaves your hands ahead of the agreement. A generated render only makes the finished state look closer than it is.
Can I use the client's AI mockup in the final design?
Only with the tool's usage terms in hand, a contract note that the client supplied generated material, and a record of what you changed. Most designers rebuild from the reference instead, because the file rarely holds up and the authorship position is cleaner when every layer is yours.
How much should a discovery fee be?
Enough to cover the brief, the constraints and an audit of the render, and small enough that a serious client does not hesitate. Many studios price it between a half day and three days at their standard rate and credit it to the project. The rate range tool gives a starting day rate.
What if the client refuses to pay for discovery?
Then the budget was never real, and the render was doing the job a contest used to do: extracting finished work while keeping the option to walk away. Decline politely, leave the door open for a scoped project, and do not send a "quick version" to prove a point. The quick version is the spec work.
What to do on Monday
Put the reply script above into your templates so it goes out on the first exchange. Add one sentence to your proposal template that names supplied generated material as reference and nothing more. Then fix your discovery fee in advance, so the next AI mockup in your inbox lands on a price you chose rather than the one the tool set for you.
The longer project is the record. Start with the Proof of Work hub, then read how to prove a design is yours before the next client asks which part of the finished file was theirs.