As we all know, the Interactive Advertising Bureau (IAB) published Version 2 of its AI Transparency and Disclosure Framework on August 18. The rule is risk-based: label AI when it could mislead someone about authenticity, identity, or representation. Do not badge every caption, colour grade, or headline just because a model touched the file.
High-risk cases that need a consumer-facing label include photorealistic images and video generated from prompts, some synthetic voices, synthetic avatars, digital twins of deceased people, living people placed in events that never happened, and chatbots that could be mistaken for staff. What does not automatically need a badge is routine post-production, internal workflows, clearly fantastical imagery, authorised commercial voice clones, generic voiceovers, background music, standard audio cleanup, and ordinary copy.
The timing is not academic. Article 50 of the EU AI Act applied from 2 August 2026. California’s SB 942 labelling rules landed the same day. New York’s synthetic-performer law has been in force since June. IAB’s framework does not replace those statutes. It gives brands one test so they are not inventing a different policy for every market.
Performance is why “do not label everything” matters. Research IAB cites from NYU Stern found that telling consumers an ad was made with generative AI cut click-through rate by about 31.5 percent. IAB’s work with Sonata Insights still found that more than half of consumers want disclosure when an ad is fully AI-generated or uses AI imagery or video, and 73 percent of Gen Z and Millennials said a clear label would increase or not change purchase likelihood. Over-labelling trains people to ignore badges and burns media budget. Under-labelling creates legal and platform risk.
What a small ad account should change this week
If you run Meta or Google ads yourself, the cost of a sloppy rule is immediate. Stamp “AI” on every ChatGPT headline and Midjourney product shot and you can give up roughly a third of the clicks you already paid for. Skip the label on a synthetic spokesperson or a human-like chatbot in the EU and you pick up Article 50 risk without a lawyer on retainer.
Use one rule in the brief: disclose when identity or authenticity could fool a customer, i.e. a face that is not real, a voice that sounds like someone who never said those words, a bot that pretends to be staff, a video of an event that did not happen. Leave the badge off AI-drafted headlines, product descriptions, colour correction, upscaling, and obvious stylised graphics. In the US, IAB accepts a sparkle icon or plain text. In the EU, Article 50 requires disclosure of AI-generated content and deepfakes but does not mandate a specific icon. That is a one-page policy a founder or freelancer can apply before the next campaign, not after a platform flag.