Disclosure is now a mix of platform policy, advertising code and, in some markets, statute. The practical rule is simple: label to the strictest standard that applies, and put the label where the viewer is looking. A caption nobody expands does nothing.
The commercial fear, that admitting a piece is synthetic destroys its credibility, has largely not survived contact with reality. What destroys credibility is being caught. Brands that label early get to compete on whether the work is good.
Do I have to label AI-generated ads?
On most major platforms, yes, for realistic synthetic depictions of a person, a place or an event that a viewer would take as real. Advertising codes on misleading representation apply independently, and the EU AI Act adds transparency duties for certain content.
Where should the AI label go?
In frame or on the asset itself, as well as in the platform metadata. A metadata flag protects you with the platform; an on-asset label protects you with the audience.