The fear
The unstated worry is that labelling an ad as AI-generated destroys its performance — that audiences discount it on sight. The fear is understandable and it is doing more damage than the disclosure would, because it drives people toward minimal, grudging compliance that looks exactly as evasive as it is.
Where disclosure actually goes
Requirements say clear and conspicuous, which is a placement standard rather than a size standard. For video that means on-screen and early. In practice, the opening moments are also where you are establishing what the piece is, so the disclosure competes with the hook.
The workable resolution is to place it where the viewer is oriented rather than mid-hook — legible, unhurried, and not designed to be missed. Hiding it in a caption, burying it after the fold, or styling it to blend into the background all fail the standard and are trivially identifiable as attempts to fail it.
Design it rather than bolt it on
A disclosure rendered in the piece's own typography, at a size that reads as intentional, does not feel like an apology. One rendered in default system type at the minimum legible size announces that somebody resented adding it.
This is a genuine craft opportunity. The label is going to be present in every asset you ship; designing it once, properly, is a better use of an hour than negotiating with it repeatedly.
Both disclosures, when both apply
Sponsored AI content needs the partnership disclosed and the AI involvement disclosed. Stacking them badly produces a wall of small print. Handling them deliberately — the platform's sponsorship mechanism for one, a designed on-asset element for the other — keeps both legible without either being buried.
What we do not yet know, honestly
Whether disclosure measurably moves click-through or conversion is an empirical question, and we do not have published data on it. It is one of the things Quantivo's creative benchmark is being built to measure on our own products, where we control the spend and can report honestly.
Until that data exists, treating disclosure as optional on the theory that it hurts performance is a decision made on a hunch, against a requirement, with penalties attached. That is a bad trade regardless of how the hunch turns out.