How AI disclosure works
There are two separate obligations hiding under one phrase, and confusing them causes most of the trouble. The first is a platform declaration: a box you tick when uploading, telling the platform that the material is synthetic. YouTube asks publishers to declare realistic altered or synthetic content at upload, and Meta requires advertisers to declare digitally created or altered photorealistic imagery in ads about social issues, elections and politics. These are rules with enforcement behind them, and they can change, so check the current policy in the platform’s own help pages before you assume.
The second is an editorial disclosure: a line you choose to publish telling readers how something was made. Nobody compels it, and it is a positioning decision rather than a compliance one. Alongside both sits invisible labelling — provenance metadata carried in a file — which platforms increasingly read automatically and may act on whether or not you declare anything.
Why AI disclosure matters
The platform side matters because the penalty for getting it wrong is not a telling-off. Undeclared synthetic media can mean removed posts, rejected ads or restrictions on the account, and an advertising account restriction is expensive in a way an apology cannot fix.
The audience side matters for a simpler reason. People mind far less about AI being used than about being misled, and the damage comes from discovery rather than from disclosure. A realistic image of a product, a place or a person that never existed is where trust breaks, especially if a customer only finds out when the real thing arrives.
Where AI disclosure goes wrong
Two opposite errors. The first is over-labelling: stamping a notice on anything an assistant touched, including spelling checks, outline help and routine editing. That tells readers nothing useful and slowly trains them to discount your work. Disclosure is about synthetic media and machine-written claims, not about which tools sat on the desk.
The second is assuming Google demands a visible AI label before it will rank a page. It does not. Google’s stated position is that it rewards helpful, original content however it was produced, and acts against material mass-produced mainly to manipulate rankings. A label neither rescues a weak page nor is required by a good one — although saying who is responsible for a page, and why they can be trusted, genuinely helps.
Getting it right
Write the rule down before you need it. Decide what your business treats as disclosable — photorealistic imagery of people, places or products; synthetic voice or video; anything presented as first-hand experience — and state that in a short editorial policy on the site. A visible policy answers the question once, for every page.
Then declare where the upload flow asks, keep a named human accountable for what goes out, and keep the wording plain. Consistency is the point: a single sentence used the same way everywhere reads as confidence, whereas an occasional label reads as an admission. The same discipline underpins credible AI-generated content generally, and it belongs in the editorial standards behind your content marketing.