AI disclosure in ads sounds simple until you start counting what qualifies: a generated background, a synthesized soundtrack, a digital human, a product demo that never happened. That nuance matters because labeling isn’t just a legal checkbox—it can be a performance tax. Research from NYU Stern and Emory University suggests AI disclosure can cut ad effectiveness by as much as 31.5%, which helps explain why brands are suddenly very interested in where the disclosure “line” gets drawn (and who’s holding the pen).
The timing is getting real: the EU’s AI transparency rules hit an August compliance deadline, and New York state requires disclosure of AI-generated humans in marketing starting June. The World Federation of Advertisers’ guidance tries to keep this manageable: existing ad self-reg rules still apply (don’t mislead, don’t manufacture claims), and disclosure should kick in when AI is central to the commercial message—i.e., it materially changes what someone believes about the product. The IAB lands in roughly the same place (don’t label everything; synthetic humans are the clearest trigger), but gets more specific: label prompt-generated images/videos and digital twins of living people in fabricated events; disclose AI renderings of deceased individuals; and attach C2PA metadata to every ad so platforms can see what AI was used and label accordingly.
Meanwhile, some brands are choosing the opposite of “AI, but responsibly” and going with “No AI” as a trust signal. Aerie put Pamela Anderson in an ad that teases AI-generated models before revealing the people are real—tying it to the brand’s long-running anti-retouching stance. Coterie says it won’t use AI-generated images in social at all, positioning it as a credibility play with skeptical parents. Le Creuset has been proactively clarifying (in comments, before anyone asks) that certain eye-catching social videos involved no AI—an interesting pivot for an industry that’s spent years selling AI’s inevitability, and is now trying to avoid “AI” reading as “inauthentic” or “manipulative.”
Read more at Digiday.