Disclosure wording
Article 50 AI disclosure wording examples
Disclosure wording must fit the exact trigger, audience and channel. These patterns are drafting prompts, not universal legal text: confirm the current obligation and exceptions before implementation.
Prepared by EU AI Fit editorial team · Published 24 August 2026 · Source review 31 August 2026 · Page updated 24 September 2026
Direct interaction with an AI system
A concise pattern is: ‘You are interacting with an AI system. [Explain its role]. [Explain how to reach a person or challenge the outcome where relevant].’ Present it no later than the first interaction unless the AI nature is obvious to a reasonably well-informed person.
Synthetic or manipulated content
A visible pattern is: ‘This [image/audio/video/text] was generated or materially altered using AI.’ The implementation may also require machine-readable marking. The precise duty, exception and responsible operator depend on the output and publication context.
Deepfake or public-interest text
State clearly that content has been artificially generated or manipulated. Artistic, satirical, creative and editorial contexts have specific qualifications, so preserve the reasoning and obtain review where the presentation could mislead.
Evidence checklist
Retain the trigger analysis, final wording, placement, timing, screenshots, accessibility review, machine-readable marking, approver, release version and monitoring result.
- Test on every relevant channel
- Keep the disclosure close to the interaction or content
- Use language the intended audience can understand
- Reassess when format, audience or model changes
In this scenario: a customer-support chat
A starting point for a direct AI interaction is: ‘You are chatting with an AI assistant. It can make mistakes. To speak to our support team, choose Contact a person.’ Use that wording only if the described handover actually exists. Test placement before the conversation begins, keyboard and screen-reader access, narrow screens and translations. Do not hide the notice in the footer or imply every answer has been checked by a human.
In this scenario: a generated illustration
Where disclosure of generated or manipulated content is appropriate, a starting point might be: ‘This illustration was generated using AI; it is not a photograph of an actual event.’ This describes that example, not a universal Article 50 notice. Identify the actual content category and applicable duties before choosing wording. A visible label does not replace any separate technical marking requirement.
Disclosure implementation record
Complete one record for each interaction or content type. These examples are drafting aids, not legally approved wording for every context.
Download editable worksheet (CSV)No email or account required. Includes response, evidence, owner, review and follow-up columns for your team. This is a working aid, not an official form or proof of compliance.
On smaller screens, scroll the table horizontally or download the worksheet.
| Item | Record | Verification |
|---|---|---|
| Trigger and role | The interaction/content and operator responsibility | Source, scope and exceptions reviewed |
| Wording | Exact message and any promised handover | No claims the service cannot fulfil |
| Placement | Where and when the audience receives it | Screenshots at the relevant point, not just a policy link |
| Accessibility | Languages, assistive technology and narrow screens | Test evidence and outstanding fixes |
| Technical marking | Separate marking/detection needs where relevant | Provider evidence and implementation test |
| Approval and change | Approver, release and reassessment trigger | Dated record tied to the deployed interface |
Recommended next step
Create one disclosure record per trigger and interface, then review the wording and technical marking together before release.
Run the free exposure checkRelated practical guides
Put the guidance into practice