Answer it once, properly
Do you have to tell clients you used AI? Sometimes by law, more often by contract, and almost always eventually.
No card. No time limit.
Do I Have to Tell Clients I Used AI?
The legal answer is quite boxed in. The commercial answer is not so much.
Three different things get collapsed into whether you have to tell clients you used AI, and they have three different answers. We pulled them apart so its easier to tell.
When you have to tell clients you used AI by law
Under Article 50 of the EU AI Act, which has applied since 2 August 2026, the disclosure duties are rather specific than general. They cover:
- Systems that interact directly with people. A chatbot has to make clear it is one, unless that is already obvious to a reasonably observant person. The Act writes this duty for whoever builds the system rather than whoever installs it.
- Deepfakes. Synthetic image, audio or video that would falsely appear authentic has to be disclosed as artificial. This duty sits with the deployer, which is likely you if you had the image made by ChatGPT for example. Where the content is part of an evidently artistic, creative, satirical or fictional work, it narrows to disclosing that such content exists, in a way that does not spoil the work.
- AI-generated text published to inform the public on a matter of public interest. Also a deployer duty, and it falls away where the content went through human review or editorial control and a person or a company holds editorial responsibility for publishing it.
Source: the three duties are in Article 50(1) and 50(4). “Deep fake” is defined in Article 3(60), and Recital 134 sets out how the disclosure should be made and why artistic work is treated differently.
However, there is no general duty to tell a client that AI helped draft their brochure, their campaign copy or their report. If the work is not published to inform the public on a matter of public interest, and it is not a deepfake, and it is not a chatbot, the Act does not require you to tell clients you used AI.
The law governs what you owe the public. Your contract governs what you owe the client.
What your contract requires
This is where it actually bites. More and more client agreements now carry an AI clause, and they come in three flavours.
Notify. You must tell the client where AI was used. Common, and easy to comply with if you keep a record.
Consent. You must get sign-off before using AI on their work. Harder, because it means asking per project.
Prohibit. No AI at all, or none on certain deliverables. Rare, but it exists, usually where the client has their own regulator.
A notify or consent clause makes the question of whether you have to tell clients you used AI contractual rather than regulatory, and a contractual breach is far more likely to cost you a relationship than a regulatory one is to cost you a fine.
What the questionnaire requires
Even without a clause, the question arrives. When we asked 53 companies for The Human Behind It, 59% had already been asked about their AI use by a client, a customer or a partner. It turns up in procurement packs, security reviews and renewal paperwork, usually as a short form with a return date near the top.
The questions are consistent: what AI do you use, who is responsible for it, does a person check the output, and can you show it. We broke down the eight that keep coming up.
At that point the honest answer is not optional, because an inaccurate answer on a procurement document is a worse problem than the AI use ever was.
Do I have to tell clients I used AI if they never ask
Practically, consider what happens if they find out later.
The downside is asymmetric. Telling a client you used AI to accelerate a first draft, which a named person then reviewed, costs you very little and most clients shrug. Being discovered to have used it after saying nothing reads as concealment, whatever the contract said.
The companies in our study that felt most comfortable could describe exactly where AI sat in their process and who checked the output. Avoiding it had nothing to do with it.
How to say it without undermining the work
Two things make the difference.
Lead with the human, not the tool. “Drafted with AI assistance and reviewed in full by our Head of Content” lands very differently from “we used ChatGPT for this”, even though they describe the same thing.
Be specific about where it did not go. Saying what AI does not touch, such as pricing, strategy or anything involving their customers’ data, is more reassuring than any assurance about what it does.
The part that takes the work out of it
All three versions of this question, the legal one, the contractual one and the questionnaire one, need the same underlying thing: a record of what you used and who checked it, made at the time.
Half the companies we asked keep no written record of where AI is used at all, and 12 of the 43 who answered said they could not work out how to prove the human check happened. Both become a scramble the moment a client asks, and the scramble is what costs the week, not the answer.
That is what Art50 is for: an inventory of the AI you use with a named person against each line, and sealed records of review that anyone can verify from a link. The free plan covers twenty records a month, no card.
Sources
- Regulation (EU) 2024/1689, the EU AI Act as published in the Official Journal. The binding text.
- Article 50, transparency obligations for providers and deployers of certain AI systems.
- Article 3(60), the definition of a deep fake.
- Article 113, which sets 2 August 2026 as the date Article 50 began to apply.
- Recital 134, on how disclosure should be made, and the narrowing for evidently artistic, creative, satirical and fictional work.
- The Human Behind It, Art50, October 2026. A study of 53 European companies, and the source of every figure on this page.
Figures come from The Human Behind It, our October 2026 study of 53 companies. Art50 provides evidence, not legal advice. See the EU AI Act timeline for which deadlines have passed, and the full text of Article 50.
