On 2 August 2026 the transparency obligations in Article 50 of the European AI Act started to apply. For anyone in advertising, the summary fits in one line: if a piece was generated or retouched by an AI and could pass for real, you have to say so.
What it actually requires
- Whoever uses an AI to create images, audio or video that look authentic (“deepfakes”) has to disclose that the content is artificial. For clearly creative or satirical work the obligation is lighter: it is enough to disclose without spoiling the piece.
- Whoever develops the system has to mark what it generates so that a machine can detect it.
- If an AI talks to people (a customer service chatbot, for example), it has to be clear that it is not a person.
What really interests me: disclosing works
The usual pitch is fear: comply so you don’t get fined. It is a weak argument and, besides, not the right one. In my final degree project, a systematic review of AI-driven hyper-personalised advertising, something else keeps coming up: what the law requires and what makes an ad work tend to be the same action.
- Hiding that it is AI triggers rejection when the audience finds out, because they feel deceived (Bakpayev et al., 2022).
- Hyper-personalising without explaining why makes users feel watched and vulnerable (Peter et al., 2025).
- Targeting without human oversight ends in biases that discriminate against part of the audience (Häglund and Björklund, 2024).
How I apply it
- A visible, honest label on the piece, not hidden in the small print.
- Human review of everything published with AI, with someone who signs it off.
- Personalisation with judgement: explaining why you see what you see.
Compliance is not what slows the campaign down: it is part of why it works.
This is not legal advice. For a specific case, read the Regulation or ask a lawyer.