Is there an official EU template for Article 50 disclosures?

No. Neither the AI Act nor the Commission's July 2026 guidelines prescribe exact wording. The guidelines set the standard the wording must meet — clear and distinguishable, delivered at the latest at first interaction or exposure, accessible to people with disabilities, and child-appropriate where relevant — and leave the formulation to you. The templates on this page are drafted to meet that standard and are free to adapt.

What should a chatbot AI disclosure say?

Something short and unambiguous, shown before or at the first interaction — for example: 'You are interacting with an AI system. Answers are generated automatically and may be inaccurate.' Per the Commission's guidelines, a single prominent notice before the first interaction usually suffices; riskier or longer contexts warrant a persistent badge or periodic reminders.

Do I need a text label AND machine-readable marking for AI content?

Often yes, but they are different duties on different actors. Article 50(2) requires the provider to mark generative outputs in a machine-readable format and provide a means of detection — a technical solution, not a text snippet. Article 50(4) requires the deployer to label deepfakes and certain AI-generated text in a way humans can perceive. One does not replace the other.

Do the disclosures have to be in the user's language?

Article 50 does not prescribe a language, but the guidelines' requirement that disclosures be clear, distinguishable and effective for the exposed audience points firmly at the language your users actually use. A notice your audience cannot read is unlikely to count as clear.

Article 50 disclosure template: copy-paste texts for the EU AI Act

There is no official EU form for the Article 50 disclosures — the law sets the standard, not the sentence. This page gives you the sentences: adaptable wording for each duty, plus the placement rules from the Commission's July 2026 Article 50 guidelines that decide whether a disclosure actually complies.

This is an information service, not legal advice. For which of these duties applies to you, answer three questions about your AI use — and for the full legal picture, see the Article 50 guide.

The four rules every disclosure must meet

Whatever wording you use, the guidelines require it to be:

Clear and distinguishable — not buried in terms of service or a footer.
Timely — provided at the latest at the first interaction or exposure.
Accessible — usable by people with disabilities (Article 50(5)).
Child-appropriate — easy to understand wherever children may use the system.

Keep a record of where and when each notice is shown — the evidence is the compliance.

Chatbot / interactive AI notice — Article 50(1)

A single prominent notice before the first interaction usually suffices:

You are interacting with an AI system. Answers are generated automatically and may be inaccurate. [Ask for a human / contact us at …]

Minimal variant for constrained UI:

AI assistant — responses are machine-generated.

For riskier or long-running contexts (advice-like conversations, minors, health or finance adjacent), add a persistent "AI" badge and repeat the notice periodically. Not required where it is already obvious to a reasonably attentive person that they are talking to a machine — but the guidelines read that exception restrictively, so when in doubt, disclose.

AI-generated or AI-altered content — Article 50(4)

For images, video, or audio a deployer publishes:

This [image / video / audio] was generated [or altered] using artificial intelligence.

For a deepfake depicting a real person, be explicit about the fabrication:

This content is AI-generated. The person depicted did not actually say or do what is shown.

For evidently artistic, creative or satirical work, a lighter placement is allowed — a credit line such as "Created with the assistance of AI" — so long as it does not spoil the display or enjoyment of the work.

AI-generated public-interest text — Article 50(4)

For published text that informs the public on matters of public interest:

This article was generated [in part] using artificial intelligence.

No label is required where the text has undergone human review and someone holds editorial responsibility for its publication — but document that review; the exemption is for genuine editorial control, not a glance.

Emotion recognition / biometric categorisation — Article 50(3)

Inform the people exposed, before or at first exposure:

This [service / area] uses an AI system that processes biometric data to [infer emotional state / assign people to categories] for [purpose]. For information about your data and your rights, contact [controller / contact details].

This sits alongside your GDPR duties — the notice satisfies the AI Act's transparency requirement; it does not supply a lawful basis for the processing. Note that emotion recognition in workplaces and education is prohibited outright under Article 5, so no notice makes those uses lawful.

What Article 50(2) needs instead of a template

The machine-readable marking duty on providers of generative AI is a technical solution, not a sentence: mark outputs in a machine-readable format (metadata, watermarking, provenance signatures) and provide a means of detection — the guidelines are explicit that one without the other does not comply. If you build on an upstream model, you may rely on its marking where it is compliant. Systems on the market before 2 August 2026 have until 2 December 2026 to comply.

Before you ship: a 60-second checklist

Notice appears before or at the first interaction or exposure — not after, not only in settings.
Wording is in the language of your users.
It is perceivable on every surface the AI appears on (web, app, embedded widget).
Accessibility: readable by screen readers; not conveyed by colour alone.
Child-appropriate variant where minors may use the system.
A record exists of what is shown, where, and since when.

These templates cover the disclosure texts; the timing, exceptions and penalties are in the full Article 50 guide. To be told when the transparency rules change — as they did with the Digital Omnibus — join the waitlist.

The official texts are Regulation (EU) 2024/1689 (Article 50) and the Commission's Article 50 guidelines (20 July 2026). This page is an information service to help you orient — it is not legal advice.