Guides · 7 min read · updated 2026-08-05
The disclosure you owe, and the one you do not
Two obligations, where they sit, and the specific mistakes that turn a technicality into a finding.
Most people running an AI persona are trying to do this properly and get it wrong on placement rather than on intent. The rules are less complicated than they look, and the failures are remarkably consistent.
They are two obligations
One: that the depicted creator is synthetic. Two: that the post is commercial, where it is. An #ad label says nothing about whether the person is real, and an AI label says nothing about whether money changed hands. Satisfying one has never satisfied the other, and nothing in any of the regimes on this site treats them as interchangeable.
Placement is where it actually fails
The consistent test across regulators is whether the audience saw the disclosure before engaging, not whether it existed somewhere on the page. That rules out three popular positions: below the 'more' fold, inside a hashtag block at the end, and in the bio only.
The bio is necessary and not sufficient. Someone arriving on a single reel from a feed has not read your bio, and that is the majority of your audience on every short-form surface.
On video, burn it into the first two seconds. Autoplay is muted and captions are collapsed; text below the video has not reached the person who scrolled past.
Language
The disclosure goes in the language of the post. ASCI is explicit about this and the reasoning generalises everywhere: a disclosure the audience cannot read has not disclosed anything. An English '#ad' on a Hindi reel is a finding against you rather than a defence.
Metadata is not optional
Most platform AI labels are driven by provenance metadata written by the generator. Leaving it intact is the low-effort half of compliance, and stripping it is the one action in this whole area that turns a disclosure question into a deception question — people who have chosen to see fewer AI images are relying on that data.
The one you do not owe
You are not obliged to narrate your workflow. Which model you used, how many re-rolls it took, whether the background is composited — none of that is a disclosure obligation, and treating it as one produces captions that are about the tool rather than about the subject.
The obligation is that the audience knows the creator is not a person and knows when they are being sold to. Everything past that is craft, not compliance.
The line you cannot cross with wording
No disclosure makes a fabricated testimonial acceptable. A synthetic persona describing an experience it did not have, presented as a customer experience, is a fabricated endorsement — and in the United States the 2024 rule on fake reviews reaches exactly that. Labelling the persona as AI does not cure it, because the problem is the claim rather than the face.
In short
- Synthetic label and ad label are separate; you need both
- Front of the caption, above the fold, in the post's language
- Two seconds of on-screen text on any muted-autoplay surface
- Leave the content credentials in the file
- No disclosure rescues a testimonial nobody had