Guide

AI use disclosure to clients: what agencies should tell them

Before work starts, send each client one page on how you use AI. Besides a purpose line and an owner with a review date, it covers five things: which AI tools and accounts you use and for what, what client data never goes into them, who reviews the work before it's delivered, when you'll say a piece was made with AI, and what the client owns. The last point matters: the US Copyright Office's January 2025 report says prompts alone don't make you the author of AI output, while human-authored parts and the creative selection and arrangement of material can be protected, case by case. Put the note in the contract or statement of work, and check each client's contract for its own AI clause first.

Short answer

  • Tools: named, with the account and its training setting.
  • Data: what never goes in, with examples.
  • Review: a named person checks everything before delivery.
  • Disclosure and ownership: when you say AI was used; what the client owns.

The one-page note

  1. Purpose, in two sentences: you use AI to work faster, and you protect the client's data and the quality of the work.
  2. Approved tools: each tool, the plan, what you use it for, and whether the account trains on your data. Anything not listed isn't used on client work.
  3. Client data: what never goes in (passwords, personal data of their customers, unreleased results, anything the contract restricts), what goes only into approved accounts, and what's fine anywhere.
  4. Review: a person checks every deliverable; facts, numbers and quotes are checked against a source.
  5. Disclosure: when you'll tell the client, or their audience, that AI was used.
  6. Ownership: what you assign to the client, written with your lawyer.
  7. Owner and review date: who keeps the note current and approves new tools, how fast you'll tell the client about a data problem, and when the note is next reviewed.

A worked example

Illustration: a made-up business. A 6-person content agency writes the note for a new retail client. In part:

From the note

Purpose: We use AI tools to research and draft faster. Your data stays in accounts that don't train on it, and a person on our team checks everything we send you.

Approved tools: Claude Team, for research and first drafts; our business plan doesn't train on our data by default [check: vendor terms]. Nothing else is used on your work.

Your data: Never entered: customer lists, sales figures before you publish them, launch dates, anything marked confidential. Approved account only: your brand guide, product details, past campaigns. Any tool: your public website and published posts.

Review: The account lead reads every deliverable before it reaches you and checks each fact and figure against a source.

Disclosure: We'll tell you if a deliverable is mainly AI-generated, and follow any platform or legal rule that requires a label [check: legal].

Ownership: On payment we assign to you our rights in the deliverables [check: legal: AI-generated parts may not be protected by copyright].

Owner: Maya Chen, managing director, approves any new tool before it touches your work. If anything goes wrong with your data, we tell you within one business day · Next review: in 6 months

The client's procurement team asks one question: whether drafts of their unreleased product launch can go into the tool. The note already answers it: product details, yes, in the approved account; launch dates and sales figures, no.

Before you send it

  1. Read the client's contract for an AI clause; theirs wins over your note.
  2. Check each tool's data terms on the vendor's page, for the plan you actually pay for.
  3. Use business accounts for client data; personal plans may train on chats unless you turn it off.
  4. Have a lawyer read the ownership and disclosure lines.
  5. Review the note every six months, because tools and terms change.

A summary, not legal advice. Copyright rules differ by country; the Copyright Office's report covers the US.

Next step: deliver your method as skills

Once clients trust how you use AI, the next step is giving them your method to run themselves. Delivering your method to clients as AI skills covers that, and the AI usage policy template covers the rules inside your own team.

Skills for this

All skills

Questions

Should agencies tell clients they use AI?

Yes, before work starts and in writing. Say which tools and accounts you use and for what, what client data never goes in, that a person reviews everything before delivery, and when you'll say a piece was made with AI. Many client contracts now ask, so check yours first.

What should an AI disclosure to clients include?

Five things on one page: approved tools with their data settings, the client data that never goes in, who reviews the work, when AI use is disclosed, and what the client owns. Mark anything that depends on law or the contract for legal review.

Does the client own AI-generated work?

It depends on how much a person contributed. The US Copyright Office's January 2025 report says prompts alone don't make you the author of AI output, while human-authored parts and the creative selection and arrangement of material can be protected, case by case. Say in the contract what's assigned, and have a lawyer check it.

AI skill

Have your AI draft your company's AI use rules

Put this skill in AI Skill Pod and everyone on your team gets it in their Claude, Cursor or ChatGPT (Pro, Business, Enterprise): one invite link, free to start. Company AI use policy draft has your AI draft your client AI note from your tools and data: approved tools with their training settings, three data classes with examples, review and disclosure rules, an owner, and every legal point marked for review.

Use this skill in your AI

14 days free · No card needed

Sources

  1. US Copyright Office: Copyright and Artificial Intelligence, Part 2: Copyrightability (January 2025) — prompts alone do not provide sufficient human control to make users the authors; human-authored elements and creative selection, coordination or arrangement can be protected; case by case (read 2026-11-05)
  2. Anthropic Privacy Center: commercial products — Claude for Work not used for training by default (read 2026-11-05)
  3. ChatGPT pricing — Business: no training on business data by default (read 2026-11-05)
  4. OpenAI Help: Data controls — personal plans train on chats by default unless turned off; read through search-result text (read 2026-11-05)