Do freelancers and agencies have to legally disclose AI use to clients?

It depends on jurisdiction, work type, and audience. The EU AI Act's Article 50 creates the strongest text-disclosure duty, but it mainly covers AI-generated content published to the public on matters of public interest, and it has a human-review exemption. California's SB 942 targets large AI platforms (1M+ monthly users), not individual freelancers or agencies. There is no comprehensive US federal law requiring disclosure of AI use in client work as of this writing. Even where no law requires it, many client contracts, platform terms, and industry norms expect it.

What is the EU AI Act Article 50 human-review exemption?

Article 50's disclosure duty for AI-generated text primarily applies to content published to inform the public on matters of public interest. Content that has undergone human editorial review, where a natural or legal person holds editorial responsibility, is exempt from the strongest disclosure requirement. The exemption is strongest when you can demonstrate the review happened, not merely assert it.

← DisclosureTrail For agencies & freelancers

Do You Have to Tell Clients You Used AI?

Short answer: it depends on where you are, what you made, and who sees it. Here's what's actually required, what's just expected, and where the two get confused.

The honest answer There is no single law that says "always disclose AI use to clients." What exists is a patchwork: one real EU law with a narrow scope and an exemption, one US state law that targets large platforms rather than freelancers, and a lot of client contracts and platform terms that ask for more than the law does. Most of the anxiety around this question comes from not knowing which bucket your situation falls into.

What's actually legally required

EU: AI Act, Article 50

The EU AI Act's strongest text-disclosure duty applies to AI-generated or AI-manipulated text that is published to inform the public on matters of public interest — think news-adjacent content, not a client's internal slide deck or a product description. And it has a human-review exemption: if a named person held editorial responsibility and reviewed the content, the strict disclosure duty doesn't apply the same way. The exemption is only as strong as your ability to show that review happened. See the full plain-language Article 50 breakdown.

US: California SB 942 (AI Transparency Act)

SB 942 requires AI-content disclosure tools and provenance data from covered generative AI providers — systems with over 1 million monthly users. It targets the platforms building the AI, not the freelancer or agency using one downstream. If you're a solo consultant using ChatGPT for client work, SB 942 isn't the law you need to worry about; it's the one your AI vendor deals with.

US: Colorado, and the rest

Colorado's original AI Act was repealed and replaced — the current version isn't in effect until 2027, and it's aimed at "high-risk" automated decision systems (hiring, lending, healthcare), not general content generation. There's no comprehensive US federal disclosure requirement for freelance or agency AI use as of this writing. Individual states move fast on this though — treat any specific legal claim, including the ones on this page, as a snapshot, not a permanent fact.

The pattern across all of these Every version of this law cares less about whether AI was involved and more about whether a human took responsibility for the result. That's the thread worth remembering even as the specific statutes keep changing.

What's not legally required but expected anyway

Most of the pressure to disclose doesn't come from statute. It comes from:

Client contracts and SOWs that specify deliverables be "original work" or don't address AI at all — silence isn't permission, and a client who finds out later feels misled regardless of what the law says.
Freelance platform terms. Some marketplaces have started requiring AI-use disclosure in listings or deliverable notes; check your platform's current terms rather than assuming.
Industry and professional norms. Legal, journalism, and healthcare-adjacent work carry disclosure expectations well beyond what any AI-specific statute requires, rooted in older professional-conduct rules.
Plain trust. The freelancers who get asked "did you use AI on this?" more than once tend to be the ones who never answered it clearly the first time.

The practical test, regardless of jurisdiction

Whether or not a specific law applies to your situation, the same question keeps coming up if it's ever tested: not "was AI used," but "can you show a named person reviewed this, and when." A verbal assurance doesn't hold up the same way a dated record does — to a client, to a platform, or to a future version of yourself trying to remember what happened on a project from eight months ago.

Situation	What to do
AI did meaningful drafting, not just spellcheck	Disclose — statement or verbal, but say something
Published under the client's name to their audience	Disclose — it's their exposure, not just yours
Client already asked once	Assume others are wondering; get ahead of it
Internal draft, human fully rewrote it	Optional, but a quick record costs nothing

What to actually do

You don't need a lawyer to get this mostly right. A short, factual record beats either extreme — the paragraph of legal boilerplate nobody reads, or saying nothing and hoping it doesn't come up. See what a disclosure statement should actually contain, with a filled-out example, or how to write an internal policy if you're setting rules for a whole team rather than documenting one deliverable.

Skip the research — generate the record

DisclosureTrail fills in tool, reviewer, date, and involvement level, and gives you a downloadable statement with a reference ID. Free, no login, nothing stored on our end.

Generate a disclosure statement →