UGC ads are legal: what the FTC actually requires in 2026
UGC ads are legal, and the line that decides it has nothing to do with the phone-camera look: an ad is lawful when the material connection is disclosed and the person on screen is not claiming an experience they never had.
Every brand that starts running UGC ads asks the same question in the same week: is this actually allowed? The ad looks like a customer filmed it on a kitchen counter, and the customer was paid, or sent a free box, or in 2026 does not exist at all.
Mostly it is not. UGC ads are ordinary advertising and they sit under ordinary advertising law. What moved is where the risk sits. For a decade the compliance conversation was about hashtags: #ad in the caption, the paid-partnership toggle, where the label goes. That half is now well understood and mechanically easy. The half that gets brands fined is one sentence in the script, spoken by someone who never used the product.
We build the AI version of these ads, so this is not a theoretical question here. It is a validator that runs on every script we generate, and it rejects lines our own paying customers typed.
Are UGC ads legal?
Yes: UGC ads are legal in the United States and in the EU, and no rule anywhere requires the person in the ad to be an unpaid stranger who bought the product themselves. Two obligations attach instead, and they are separable.
The first is disclosure. Where there is a material connection between the brand and the person endorsing it, the audience has to be told. Payment counts, and so does a free product, a discount code, an affiliate link, a contest entry or a family tie. The second obligation is truth: the endorsement has to reflect something real. A paid, disclosed endorsement of a product the endorser genuinely likes is lawful advertising. An undisclosed one is a violation. So is a fully disclosed one that invents an experience, and the disclosure does not cure it.
That second half is the part most compliance guides skim, because until recently it rarely bit. Real creators had at least held the box. The framework assumed a human endorser with a memory of the product, so nobody spelled out that the memory had to be real. Once the person on camera is generated, the assumption fails quietly.
None of this is about the phone-camera look. Vertical framing and a face talking to the lens are creative choices, not legal ones. If you are still working out what counts as UGC at all, we set out the definition separately.
What do the FTC endorsement guides require?
The FTC Endorsement Guides (16 CFR Part 255, last revised in June 2023) require any material connection between an advertiser and an endorser to be disclosed clearly and conspicuously, near the claim it qualifies. Clearly and conspicuously has a working definition, and in a fifteen-second vertical video it is demanding.
In practice the disclosure has to be unavoidable for an ordinary viewer. It sits in the video, not in a caption a viewer has to expand and not in a profile bio, and it appears early rather than after the hook has done its work. Where the claim is spoken, saying the disclosure out loud as well as showing it is safer: a muted autoplay and a sound-on watch are different experiences of one file. Shortened forms are the classic failure. "sp", "spon" and "collab" are not disclosures. "Ad", "sponsored" and "paid partnership" are.
The platform toggles help without finishing the job. TikTok's branded content switch and Instagram's paid partnership label put a machine-readable flag on the post, which is what each platform's own policy asks for. The FTC's test is whether a viewer notices, so an on-screen mention still earns its place. Responsibility runs to the advertiser as well as the creator: a brand that hands over a script and never checks how it went out owns the result.
One thing we cannot do for a customer: the toggle lives inside the platform. We render a 9:16 file, you upload it and set the flag. What is inside the file is the part we can control, and that is where the harder rule lives.
What changed with the FTC's fake review rule?
In August 2024 the FTC finalized a rule on consumer reviews and testimonials, 16 CFR Part 465, effective 21 October 2024. It bans reviews and testimonials that misrepresent themselves as coming from someone who does not exist, or from someone who never had experience with the product.
The rule reaches creating, buying, selling and disseminating that content, including where a business should have known it was fake. According to the FTC's announcement of the final rule in August 2024, it carries civil penalties of up to $51,744 per violation. Before it the Commission could still act, but it had to build a deception theory case by case. Now there is a written rule with a number attached.
Read that definition again with AI UGC in mind. A testimonial from someone who does not exist is an exact description of a generated presenter reading a testimonial script. The presenter is not the problem; a synthetic spokesperson has been legal in advertising for as long as animation has existed. The problem is what the synthetic person is made to claim about their own life.
So the rule does not ban AI presenters. It bans synthetic testimony. That distinction is the entire design space for this category of tool, and from what we can see, most products in it have not drawn the line anywhere. They ship a model, a face and a text box that will happily accept "I have used this for six months".
Can an AI presenter say it has been using the product for months?
No, and we enforce that in code rather than in a help page. Every script our pipeline generates runs through a check called claims_fabricated_experience, and a line that matches is rejected and rewritten before anything is rendered.
The check looks for two families. One is durable usage or purchase history: "I have been using this since last summer", "I have owned one for years", "I bought mine in 2024". The other is measurable personal results: "after two weeks my skin cleared", "it saved me $500", "my sales doubled". Both put a memory in the mouth of someone who has none.
Two implementation details say more about the position than the rule does. The check runs before the line-length check, on purpose, because a complaint about pacing would mask the signal that matters. And it fires in exact-words mode, where the customer typed the sentence and asked us to have the presenter say it verbatim. The comment in our source is blunt about why: the ad is ours to render regardless of who typed the line. We overrule the paying customer there, and we intend to keep doing it.
What stays allowed is a real reaction, because reacting is what the presenter is doing. First person, present tense, opinion, and doubt turning into conviction: "I honestly thought this was nonsense, and here we are." We started stricter than that and narrowed it in August 2026, because a spokesperson forbidden from having an opinion writes ads nobody watches. Two limits we will state plainly rather than imply away: the deterministic check covers English and German, so the other nineteen languages we generate into rely on the prompt contract, which is weaker. And our terms forbid re-cutting the presenter into something that reads as a customer review, which is a rule we can write but cannot police frame by frame.
Do you have to label an AI ad on TikTok or YouTube?
Yes on TikTok, Instagram and YouTube, each through its own setting, and since 2 August 2026 the EU adds a legal duty on top of the platform rules. They are separate obligations that happen to overlap.
YouTube asks creators to use the "AI use" setting in Studio when altered or synthetic media makes realistic content: a real person appearing to say something they did not say, or a realistic scene that never happened. Beauty filters, color correction and using a model to help write a script are exempt. For photorealistic material the label appears on the video itself rather than in the description.
TikTok requires AI-generated content to be labeled and has read Content Credentials since May 2024, when it announced it would become the first video platform to implement the C2PA standard and automatically label incoming AI content carrying that metadata (TikTok Newsroom, 9 May 2024). Meta applies its own AI information labels across Facebook and Instagram.
The EU layer arrived on 2 August 2026, when Article 50 of the AI Act became applicable. Deployers of systems that generate or manipulate image, audio or video deepfakes have to disclose that the content is artificially generated, at the latest at the point of first exposure, and providers have to mark outputs in a machine-readable format. If you run ads into the EU, that duty is yours as the advertiser, not your tool vendor's.
The trap is treating the label as absolution. A correctly labeled ad containing a fabricated testimonial is a correctly labeled fabricated testimonial. Disclosure answers whether something was generated. It never answers whether the claim is true.
Who owns the footage in a UGC ad, and what changes when nobody filmed it?
The creator owns their footage by default, and a brand needs a written license before running it as an ad. Tagging your brand in a post is not a license, and neither is a comment saying go ahead.
The distinction that matters is organic reposting versus paid media. Sharing a customer's video to your own feed with credit is a small exposure. Putting budget behind it, or running it from the creator's handle as a Spark Ad, is a licensing question with a scope, a term and a territory attached. Those windows are priced: we broke down what creators charge, usage rights included in a separate post. Music is a second layer, since a platform's in-app library is licensed for organic posts and usually not for ads.
Generated ads remove one of those layers and none of the others. There is no performer to license and no rights window to renew. The product images stay licensed material, and in our pipeline they come from the brand's own page, which is the customer's to begin with. Music stays licensed, trademarks in frame stay trademarks, and the claims stay the advertiser's responsibility.
Which brings the whole thing back to one place. The legal question people ask about UGC ads is almost always about the format, and the format was never the exposed part. The script is.