How to Write an AI Disclosure for Your Website
An AI disclosure for your website needs different wording for chatbots, generated copy and customer data. Three drafted versions and where each goes.
Most sites write one AI disclosure for your website, park it in the footer, and treat the matter as handled. That single notice usually covers the wrong thing. The obligations differ by what the AI is doing, and a chatbot, a page of generated copy and a form that feeds customer data to a model are three separate disclosures in three separate places.
Here is how to tell which ones you owe and what each should say.
Start with the one rule that is actually binding
Article 50 of the EU AI Act, in force from 2 August 2026, is the provision most small sites will touch. Two parts matter.
Providers of AI systems that interact directly with people must ensure those people are informed they are dealing with an AI system. There is an exemption where it is obvious to a reasonably well-informed, observant and circumspect person in the circumstances.
Deployers of AI systems generating text published to inform the public on matters of public interest must disclose that the text was artificially generated or manipulated. This one carries an exemption that is widely missed: it does not apply where the content has undergone human review or editorial control and a natural or legal person holds editorial responsibility for publication.
Read that second exemption carefully before you label anything. A blog post drafted by a model, then genuinely edited and signed off by a named person who stands behind it, sits outside that disclosure duty. A page generated and published untouched does not. The dividing line is editorial responsibility, not whether a model was involved.
The information has to be given clearly and distinguishably, at the latest at the first interaction or exposure, and must meet accessibility requirements.
The three cases and what each needs
What you are doing | Disclosure owed | Where it goes |
|---|---|---|
Chatbot or assistant users talk to | Yes, unless unmistakably obvious | In the chat window, before the first exchange |
Published text generated with no human editorial control | Yes, where it informs the public on matters of public interest | On the page itself, near the byline |
Published text drafted by AI then edited and owned by a named person | Not under the editorial-control exemption | Optional, and still often worth doing |
Customer data sent to a model | Not Article 50, but privacy law applies | Privacy policy and the point of collection |
The fourth row is the one businesses get wrong in the other direction. Sending customer data to a third-party model is a processing and transparency question under data protection law, not an AI-labelling question, and burying it in an AI notice instead of your privacy policy does not discharge it. Writing a privacy policy for an AI app covers what belongs there.
Drafted wording you can adapt
For a chatbot, shown before the first message rather than after:
> You are chatting with an automated assistant, not a member of our team. It can answer questions about orders, opening hours and our returns policy. Ask for a human at any point and we will pass you to one during business hours.
Three things that wording does: it states the fact, it scopes what the assistant is competent for, and it gives an exit. The exit is not legally required and it is the part that prevents complaints.
For published text with no editorial sign-off:
> This page was generated with AI assistance and has not been individually reviewed by our editorial team. Tell us if you spot an error.
If that sentence makes you uncomfortable, the honest fix is to review the page rather than to soften the sentence.
For text that is edited and owned, where no disclosure is owed but you want to be straightforward:
> Drafted with AI assistance, edited and fact-checked by [name], who is responsible for its accuracy.
Naming a person is what converts this from a hedge into a claim of responsibility. It is also, not coincidentally, the thing the Article 50 exemption turns on.
Where to put the AI disclosure on your website
In the chat window, not in the footer. A footer notice is not given "at the latest at the first interaction" for someone who opens a chat widget on a product page.
On the page, near the byline, not on a separate AI policy page linked from the footer. A reader exposed to the content has to be able to see the notice where they encounter the content.
In your privacy policy for the data question, at the point of collection as well if the processing would surprise someone. A support form that quietly routes messages through a model is the surprising case.
Avoid the generic badge. "This site uses AI" tells a reader nothing about whether the thing in front of them was generated, whether their message is being read by a model, or where their data goes, and it satisfies none of the three obligations properly.
Keep it true as the site changes
The disclosure that was accurate when you launched a chatbot stops being accurate when you widen what it can do. Re-read these notices whenever the underlying system changes, and treat the chatbot one as part of the release checklist rather than as legal boilerplate.
A disclosure also sets an expectation you then have to meet. If your assistant says it can hand over to a human, make sure it can. The failure mode there is covered in what to do when an AI chatbot gives a customer wrong information. Broader transparency duties arriving alongside Article 50 are set out in the EU AI Act transparency rules, and the general risk picture for builders sits in the practical guide to AI risks.
FAQ
Do I legally have to disclose that I used AI to write my website copy?
Under Article 50 the duty bites on text published to inform the public on matters of public interest, and it falls away where the content had human editorial review and a named person holds editorial responsibility. Marketing copy you wrote, edited and stand behind is a weaker case for a mandatory label than an unreviewed generated article.
Does a footer notice count as disclosure for a chatbot?
Not reliably. The information has to reach the person at the latest at their first interaction, and someone who opens a chat widget has not necessarily seen your footer.
What if the AI use is obvious anyway?
Article 50 exempts direct-interaction disclosure where it would be obvious to a reasonably well-informed, observant and circumspect person in the circumstances. That is a narrower test than "our users are tech-savvy", so do not lean on it for a general-audience site.
Should I disclose AI use even when I am not required to?
Often yes, because the cost is one sentence and the alternative is a customer discovering it later. Naming the person responsible turns it into a credibility signal rather than a warning.
How did this land?
About the author
Senior Editor, AI & Product
Cecilia leads the Swarmz editorial desk. She has spent a decade turning complex AI and product topics into writing people actually finish, and she owns the blog's quality bar.
