Two questions now arrive with almost every piece of creative or technical work: "Did you use AI?" and, a beat later, "Then what am I paying for?" Most people answer the first defensively and the second badly. There is a better way. Disclose precisely, from a record you kept from the start, and the disclosure becomes the clearest evidence of the value you added. Here is the format, what to record, and what the rules actually say.

What the rules actually require

Three sources matter for most work, and none of them says "hide it".

  • Google Search says AI use is not against its guidelines; manipulation is. It recommends AI disclosures "where someone might think 'how was this created?'" and advises against listing AI as an author. Bylines should be real people.
  • The US Copyright Office requires applicants to disclose AI-generated material in a registration and to describe the human contribution. Its Copyrightability report, still the operative guidance in 2026, confirms copyright covers only what a human determined: selection, arrangement, modification. Prompts alone do not qualify. A false or missing disclosure can void a registration.
  • Platforms and distributors (music services, stock libraries, app stores, publishers) increasingly require AI disclosure at upload. The specifics vary; the direction does not.

Beyond the rules, there is the client relationship. A client who discovers undisclosed AI use later trusts nothing you said before. A client who was told precisely, up front, what AI did and what you did usually pays the invoice without a question, because the answer to "what am I paying for" is right there.

Record from the start, or you cannot disclose later

You cannot reconstruct an honest disclosure at the end from memory. Capture four things when the work begins and keep them attached to it:

  • What was generated: which outputs, which tool, roughly when.
  • The prompt or brief given to the tool, redacting anything confidential.
  • What a person changed: kept as a delta against the generated version where possible (a diff, a before/after, a change list).
  • Who decided what: approvals, rejections, direction changes, with names and dates.

This is what a job record on After AI Work holds by default. Outside the platform, a folder with dated files and a running change log does the same job.

The disclosure format

Short, specific, structured. Five parts, in this order. It fits in an email, a deliverable's README, a project page, or a copyright application.

AI use in this work

Generated: Four initial logo directions and a colour palette, using [tool], from a brief we wrote, on 12 September.

Selected: Direction B was chosen by [client name] on 14 September for its legibility at small sizes; the palette was discarded.

Human work: The mark was redrawn from scratch as a geometric vector; stroke weights, spacing and small-size variant designed by [designer]. Colour system, typography (licensed typeface), all variations, favicon set and brand rules created by [designer]. No generated pixels remain in the final files.

Verified: Contrast checked to WCAG 2.2 AA; trademark screening run on the final mark; typeface licence on file.

Ownership: Final deliverables are original human authorship and assigned to [client] under the agreement. The generated concepts are retained as reference only.

Adapt the five headings to the medium. For code: generated, reviewed, rewritten, tested, deployed. For an article: drafted, verified, rewritten, approved. The structure is the same: what the machine made, what the person did to it, and who is standing behind the result.

Where to put it

  • In the proposal, one line: "We use AI for first drafts and concepts; all delivered work is reviewed, reworked and approved by a named person, and we provide a written AI-use summary with every deliverable."
  • With the deliverable, the full five-part format above.
  • On the published work, only where a reader would reasonably ask. A byline plus a short line is enough for most content. Google's guidance is exactly this.
  • In a copyright application, as the Office instructs: describe the human authorship in "Author Created" and exclude the AI material in "Limitation of Claim".

Why it helps you

The "then what am I paying for" question is only hard if you have nothing to show. The disclosure format is a list of exactly what you did: the judgement, the craft, the verification, the accountability. It is the single best argument for your rate. It also pre-empts the disputes that come from a client finding out later, gives the client something to hand to their own legal or compliance team, and produces the copyright record they will need if they ever want to register the work. Experts on After AI Work show the before-and-after publicly, with permission, for the same reason: the delta is the portfolio.

Mistakes to avoid

  • Vague disclosure. "AI was used in this project" tells the client nothing and sounds like a confession. Be specific about what and what not.
  • Overclaiming human work. If generated pixels or sentences remain, say so. The Copyright Office point about voided registrations applies.
  • Disclosing at the end only. Clients react to surprise, not to AI. Say it in the proposal.
  • Listing the tool as an author. Google and the Copyright Office both say no. A tool is a tool; a person is the author.
  • Treating it as a risk. It is a receipt for your judgement.

How this was made: written by the After AI Work team from disclosure practice on our own platform. Google's position from "Google Search's guidance about AI-generated content", still Google's current published guidance in 2026. US Copyright Office position from its registration guidance and Copyright and Artificial Intelligence Part 2 (Copyrightability), both operative in 2026. AI tools helped structure the format; every sentence was rewritten and checked by a person.