Do you have to disclose AI-generated ads?

01Do you have to disclose AI-generated ads?

Usually not, and the reason matters more than the answer. No regulator I could find asks whether a tool touched your ad. They ask whether the ad claims something untrue. A generated backdrop behind your own product makes no claim about a person or an event, so it sits outside almost everything below. A synthetic woman saying the cream worked for her makes a claim about a person who does not exist.

That single distinction sorts most cases. The law is aimed at deception and at impersonation, not at rendering. Which is why the harder question is rarely legal at all, and is covered separately in why AI ads look fake: whether the audience clocks it before any regulator does.

02What the law says, by jurisdiction

Four regimes, kept apart because they answer different questions and move on different clocks. Each panel names what I read and the date I read it. Where I read an unofficial reproduction because the official one would not load, the panel says so rather than pretending otherwise.

03What do the platforms actually require?

Platform policy is the rule that will actually stop your ad, and it is not the same rule as the statute. Every platform below treats political and social issue advertising separately and more strictly. None of them, on the pages I could read, requires a label on an ordinary AI-assisted product ad.

One nuance worth carrying away. Meta's requirement on ordinary posts is aimed at people who upload photorealistic AI video or realistic-sounding audio, and YouTube's is aimed at content that makes a real person appear to say or do something they did not do. Both are about depicting reality, not about the tool. Google's carve-out is the clearest version: resizing, cropping, color correction and background edits are inconsequential and need no disclosure.

04Who owns the copyright in an AI-generated ad?

In the United States, nobody owns the purely generated part. The Copyright Office has held since March 2023 that copyright protects only material that is the product of human creativity, that it will not register works produced by a machine, and that an applicant has a duty to disclose AI-generated content and disclaim anything beyond minimal use of it.

Prompts alone, however, at this stage are unlikely to satisfy those requirements.

US Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability, January 2025

The same report concludes that existing legal doctrines are adequate and appropriate to resolve questions of copyrightability, and that outputs will often be copyrightable in whole or in part where AI is used as a tool and a human determined the expressive elements. For an ad that means the edit, the selection, the arrangement and any real footage are yours. A frame that came out of a prompt is not.

05What I could not verify, and the page that would settle it

This is the part most pages leave out, and it is the part I would want if I were you. Three things on this page are weaker than the rest, and here is exactly how weak and exactly which page would fix it.

The three gaps, plainly. TikTok: unverified, because the connection was refused. The pages that would settle it are the Integrity and Authenticity section of the Community Guidelines and TikTok's AI-generated content help article. EUR-Lex: the authoritative Article 50 text would not render in full, so the wording here came from an unofficial reproduction and the dates from the Commission's own page.

Everywhere outside the EU, US and UK: not checked at all. China, India and the individual US states each have their own instruments and their own clocks, and I would rather leave a hole in this page than fill it from memory. If your spend sits in one of those markets, that hole is the most important thing on the page for you.

06Provenance is a story, not a badge

Compliance is a floor, and clearing it is not a position. Ours is narrower and older than any of these rules: no invented statistic, no clinical claim, every claim checkable in public sources on the day the ad runs. It is the hardest line in what this studio makes and how, and it disposes of most of the real risk before a regulator is involved.

The interesting question is not whether to put a badge on the ad. It is whether the truth about how it was made is worth telling. Usually it is, and not as a label. Put the real product photography at the spine of the film and let generated coverage sit around it, and the ad has told the viewer what it is without a word of disclosure text.

There is one more reason to be careful here that has nothing to do with law, and it is the only place anyone has put a number on the commercial cost. A Journal of Consumer Research analysis of 1,135,817 TikTok posts measured a roughly 7 to 8% engagement penalty on disclosed-AI content, within-creator and holding views constant, so the gap is not explained by who posted or how far the post traveled. The mechanism the authors identify is a drop in parasocial connection, driven by the audience reading disclosed work as less effort. The penalty runs larger on emotional content than on informational content, and posts labeled AI-enhanced are penalized about half as much as posts labeled AI-generated.

Take the method rather than the answers. Every claim above names the page it came from and the day I read it, and the three I could not verify are marked as holes instead of filled from memory. Apply that standard to whoever you hire next, which is most of what nine questions to ask an AI ad studio is about, and to claim-heavy categories like supplement ad creative.