No US law makes you label an AI ad, and New York's new one is narrower than the alerts say

01Does a US ad have to be labeled as AI?

No. There is no federal AI advertising label, and no state requires a blanket "made with AI" mark on commercial advertising. New York requires something narrower: a conspicuous disclosure when a synthetic performer appears in the ad. Everything else US law polices is the claim the ad makes and who the ad says is speaking.

I am reading the public text here as the person who builds this creative, not as a lawyer; the statute is linked below and takes four minutes to read yourself. Two questions keep getting answered as one. A generated product shot, a generated background, AI-written copy, an upscaled photo, an AI voiceover in a radio spot: none of those triggers a disclosure duty anywhere in the United States today. A generated person who seems to recommend your product triggers two regimes at once, and only one is about AI.

Read down the first column and the shape comes clear. Four of the ten rows care whether a real, identifiable human being is depicted. Two care whether the ad is political. One asks whether a viewer thinks they are watching a customer. Only one row asks whether a machine made the picture, and that row is a platform policy rather than a law. Google also carries an AI label setting it points at New York by name, and the platform side of this is a separate post, because the four platforms disagree more than the states do.

02What New York's law says, in its own words

It amends General Business Law 396-b, the 1965 advertising section, rather than creating a new chapter. Subdivision 3 is the whole duty: a person who for a commercial purpose produces or creates an advertisement shall "conspicuously disclose in such advertisement that a synthetic performer is in such advertisement, where such person has actual knowledge." Three phrases carry the whole duty: conspicuously, synthetic performer, and actual knowledge.

Notice what the statute declines to do. It prescribes no wording, no font size, no placement and no duration. That leaves a creative team to borrow a floor from somewhere else, and the nearest available one is the FTC's: a disclosure "difficult to miss (i.e., easily noticeable) and easily understandable by ordinary consumers," and unavoidable in an interactive electronic medium. The consumer reviews rule is blunter still, saying a disclosure fails if a consumer "must take any action, such as clicking on a hyperlink or hovering over an icon, to see it."

Two corrections worth carrying into a meeting. Some secondary write-ups describe a five-day written-notice grace mechanism for advertising media; that language is not in the enacted text, and repeating it will cost you credibility with whoever checks. And subdivision 2, the 1965 text, contains the words "in this state" while the new subdivision 3 does not. That is an observation about the text rather than a conclusion about reach, and it is the first thing I would put in front of counsel if your buying is national.

New York tells you to disclose conspicuously, and then stops. No wording, no font, no placement, no duration.

GBL 396-b(3), read on its own terms

03Does this ad need a disclosure?

Three questions decide it, and the first one is not about AI. Run the cut you are about to upload through the tree below. It answers for the ad as it stands, not for the workflow that produced it. Most compliance conversations get that backwards.

04Four states price a cloned voice far above New York's thousand dollars

New York's law covers a performer nobody can identify, and prices it at a thousand dollars. Four states cover the opposite case and price it much higher. Tennessee made voice a protected property right in July 2024, defining it to include "a simulation of the voice of the individual." Illinois bars distributing a work containing an unauthorized digital replica. Washington added a forged digital likeness to its Personality Rights Act in June 2026. California's advertising hook for a deceased personality has sat in Civil Code 3344.1 for decades.

The risky brief is the one that names a real person. Make me this creator. Make me our founder, saying the line he never got around to recording. A voice clone of somebody who signed nothing is the cheapest lawsuit a marketing budget can buy, and the craft answer is to build a character who belongs to nobody and hold that same person across every shot so you never need a real face to anchor the campaign.

05Do Utah, Colorado, Texas or Illinois label an ad?

None of the four. Utah's disclosure duty attaches to conversational AI and runs only when a customer makes a clear and unambiguous request to know. Colorado's replacement act names advertising and marketing as excluded and does not start until January 1, 2027. Texas puts its disclosure duty on governmental agencies and, by extension, health-care providers. The Illinois AI act is employment law.

One line in Utah's chapter does reach an advertiser, and it is the sentence to memorize: it is no defense that generative AI made the violative statement or was used in furtherance of the violation. If the model wrote a claim you cannot substantiate, the claim is yours. That principle runs through every layer of this subject, and it is why the fake reviews rule matters more than any label for a brand running generated testimonials.

NEW YORK, GBL 396-b(3)          no wording prescribed.
  The duty:  "conspicuously disclose in such advertisement that a
              synthetic performer is in such advertisement"

CALIFORNIA, SB 1050 (PENDING)   would prescribe, substantially:
              "this performance features a synthetic performer"
              "no human performer is depicted"

GOOGLE ADS (ELECTION LANE)      its own published examples:
              "Altered or synthetic content"
              "This audio was computer generated"
              "This video content was synthetically generated"

FTC, 16 CFR 255.2(c)            the one federal wording, and it is
                                about customers rather than machines:
              disclose that "the persons in such advertisements are
              not actual consumers of the advertised product"

PLACEMENT FLOOR, FTC STAFF      in the video, not in the description.
                                Not behind a tap, a hover or a MORE.

06What changes in the file

This is a paper decision, which costs almost nothing early and a re-render late. On a written direction board the synthetic performer question gets answered before any frame exists: a generated human is on screen or one is not, and if one is, the disclosure gets a layout slot as serious as the price. Deciding it in writing is also what makes buying creative without an RFP survive a legal review later. Two questions stay off this page: whether the finished ad is copyrightable, and what the EU AI Act and the UK's ASA require of an AI ad. The sister journal answers both at their own sources.

Every alert about this law leads with the penalty, and the penalty is a thousand dollars. The definition is what deserves your attention. New York decided the fact worth telling a viewer is that nobody was there. Once you accept that, the disclosure stops being a compliance line at the end of the process and becomes a casting decision at the start of it, made on paper, before anything renders.