FTC, NY, and Article 50 are three labels

FTC, NY, and Article 50 are three labels. A paid integration needs a clear endorsement disclosure. An ad reaching New York consumers with a generated human who is not a recognisable real person needs the NY synthetic-performer sentence, in force 9 June 2026. A deepfake published to an EU audience needs an Article 50 mark at first exposure, applied 2 August 2026. One platform sparkle does not do those jobs. Stack the ones that apply. Skip the ones that do not. This is a production map, not legal advice.

Three questions a viewer can still get wrong

Hold the finished unit and ask all three.

1. Would this person reasonably think nobody paid for this? That is the FTC endorsement disclosure. The Endorsement Guides (16 CFR Part 255) care whether a reasonable consumer can tell there is a material connection — payment, free product, affiliate, employment — and whether the endorsement reflects an honest experience. The FTC's creator guidance wants the disclosure clear and conspicuous: hard to miss, in the same language as the content, close to the claim. A buried #ad in a hashtag pile has repeatedly failed that standard. Since 2023 the FTC has said this applies whether the on-screen endorser is a human, an AI avatar, or a cloned voice. The tool that produced the video changes nothing about the commercial fact.

2. Is this an advertisement, reaching New York consumers, with a generated human who is nobody in particular? That is N.Y. GBL §396-b. A synthetic performer is a digitally created asset, made with generative AI or another algorithm, intended as a human performance by someone not recognisable as an identifiable natural person. Advertisers with actual knowledge must conspicuously disclose in the ad. HQ does not matter. Reach does. Penalties $1,000 first / $5,000 subsequent — a statute, not a Versely price. AP and firm notes (McDermott / Crowell / Manatt, June 2026) are the public recap layer. Commentators point enforcement at the NY AG; no private right of action in those recaps. Audio-only ads are exempt. AI used solely for language translation of a human performer is exempt. Ads for expressive works (film, TV, games) can be exempt when the synthetic use matches the work. Animals, objects, and non-human imagery are out. Digital twins of identifiable real people are a different legal object. This section does not cover them.

3. Is this authentic-looking synthetic media (or covered public-interest text) published to an EU audience? That is Article 50 of the EU AI Act. Guidelines 20 July 2026. Providers machine-mark synthetic audio, image, video, and text; grace to 2 December 2026 for systems already on the market before 2 August. Deployers who publish a deepfake owe a visible or audible label at first exposure. Artistic, satirical, or fictional work may label without spoiling. Assistive editing is exempt. Public-interest AI text is labelled unless a human exercised editorial control over the claims. Three official EU icons exist. Fines can reach €15 million or 3% of worldwide turnover. A platform AI toggle is not automatically this duty.

A file can hit three, two, one, or none. Clearing one does not clear the others.

Where they overlap and where they split

A photoreal generated UGC presenter in a paid unit that runs in the US and the EU:

• FTC: if there is a material connection (the creator was paid, the "customer" is a hired presenter, the brand gifted the SKU), say it is an ad, on the creative, in language a muted three-second viewer can read. An AI info tag does not say that.
• NY: likely in, if the person is not a recognisable real individual, the producer knows it is synthetic, and NY consumers can see the ad. Conspicuous disclose in the ad.
• Article 50: likely in for EU viewers, because a realistic person who was never filmed is a deepfake-shaped file. First-exposure label. Provider should have machine-marked.

A real creator, paid, on camera, no generate:

• FTC: in. Material connection.
• NY: out. Not a synthetic performer.
• Article 50: out of the deepfake deployer duty for that face, unless you also published a separate synthetic.

A photoreal generate in organic social, not an ad:

• FTC: usually out, unless there is still a material connection (employee, gifted SKU, affiliate). Organic is not an exemption from endorsement rules if the connection exists. It is an exemption from NY's ad-only statute.
• NY: out. This is an ad rule.
• Article 50: still in if it is a deepfake to an EU audience.

A generated background or product turn with no fake human:

• FTC: only if someone is still endorsing with a hidden connection.
• NY: out. Not a synthetic performer.
• Article 50: maybe, if the place or event would appear authentic. A clearly impossible scene is a weaker case.

YouTube's 27 May 2026 Blog post: photoreal / meaningfully AI-altered content gets a Shorts overlay on the video; long-form sits below the player, above the description. Labels alone do not impact recommendations or monetization eligibility. That overlay is not an FTC #ad, not a NY sentence, and not an Article 50 deployer mark. Mashable, 21 July 2026, separately recapped YouTube's inauthentic buckets — including synthetic personas presenting as human authorities on healthcare, finance, law, or politics. Monetisation eligibility is a fourth system. Do not restack it onto these three.

Placement is the production problem

Three labels are three graphics. Count them as one budget with the captions and the CTA.

Speech captions transcribe. They are not a legal row. Do not hide a mark in a subtitle cue the viewer can miss, or that a mute autoplay feed never enables. Burn legal lines into the picture at first exposure when the duty is first exposure. Article 50 is explicit about first exposure. NY wants the disclose in the ad. FTC wants conspicuous, near the claim.

Safe area still applies. A Shorts AI overlay occupies pixels. A NY sentence plus an #ad plus two-line speech plus a shop card is four objects in the bottom fifth. Sequence them. First frame: the paid mark and, if required, the synthetic mark, readable on mute. Then the hook. Then the proof. Then the ask. Captions are not the CTA card. They are also not the legal stack.

Platform toggles: Meta paid partnership, YouTube paid promotion, TikTok branded content. Use them where offered. They satisfy the platform. They do not automatically satisfy the FTC's clear-and-conspicuous standard, NY's on-ad sentence, or Article 50's first-exposure test. Check both boxes.

The copy-paste labels live in the Article 50 disclosure template. The NY-only explainer is New York synthetic performer ad disclosure. The two-way compare is NY synthetic performer vs Article 50. This page is only the three-way: paid, NY human, EU deepfake.

Decide per file, then typeset

Log at generate time whether a human on screen is a real identifiable person, a synthetic nobody, or a replica of someone named. Actual knowledge is easy to prove if you pressed generate. It is messy if you only notice at delivery.

Then ask: is it an ad, and who can see it. US paid social still runs the FTC question. NY reach still runs §396-b. EU publication still runs Article 50. Organic, audio-only, translation-only, expressive-work trailers — those are real exemptions. "It looks cinematic" is not.

Burn the marks that apply. Keep them out of chrome. Keep them off the transcript. The AI caption generator will not write the legal row. Author it. 21 BASIC styles at 1× are for speech. A static legal card is overlay.

FTC, NY, and Article 50 are three labels. Paid, fake nobody in an ad, EU deepfake. One badge is not three.

Stack only the duties the file actually owes, in Versely.

FAQ

Does YouTube's AI overlay cover NY and Article 50?

No. The 27 May 2026 YouTube label is a platform mark on photoreal or meaningfully altered content. NY wants an on-ad sentence about a synthetic performer. Article 50 wants a deployer label at first exposure for EU deepfakes. Use the overlay. Do not stop there if the other tests fire.

If I already wrote #ad, do I still need a synthetic disclose?

If the on-screen person is a generated human who is not a real identifiable person, and the unit is an ad that can reach New York, yes — NY is a different sentence. If EU viewers will see an authentic-looking synthetic, Article 50 is a third. #ad answers money. It does not answer "this human is not a person."

Are audio ads in the NY rule?

The recaps we are using treat audio-only ads as exempt, and AI used solely to translate a human performer as exempt. A talking generated face in a video ad is not audio-only. Picture is the usual trigger.