Distinguish provider and deployer duties.
Article 50 covers specified AI transparency duties. These include informing people about certain direct AI interactions, provider marking of synthetic outputs and deployer disclosures for particular uses such as deepfakes. Each obligation has its own scope and exceptions.
Do not turn that framework into a blanket statement that all AI-generated text always needs the same label. Public-interest text provisions, for example, include conditions concerning human editorial review and responsibility.
Review the actual point of interaction.
Map where a person first encounters the AI feature or generated content. Ask whether the information is understandable, timely and accessible in that setting. A general statement buried in a policy may not answer the relevant interaction question.
Consider the full journey: embedded widgets, exported content, shared files and downstream channels can change where information is seen. Identify which team owns each presentation surface.
Record the decision and its implementation.
Keep the scope assessment, the approved wording or marking approach and evidence of where it is used. Record any exception relied on and the facts supporting it. If another supplier or team controls part of the experience, make that dependency explicit.
- Interaction or content category reviewed.
- Responsible provider or deployer role.
- Notice, disclosure or marking approach.
- Review outcome, owner and change trigger.
Revisit transparency when the product changes.
A feature that moves from internal drafting to public interaction may need a different review. Changes in content type, editorial responsibility or distribution channel can also affect the assessment. Connect those changes to the release or publication process.
EUAIC can help structure the decisions and supporting evidence. The organisation should obtain appropriate advice on the particular duty and verify the actual implementation rather than relying on a generic policy sentence.