AI UGC ads: how brands use them and what to watch for

This article is general information, not legal advice, and it contains no performance or cost figures — those depend on your market, format and brief. Advertising rules and platform policies differ by country and change often.

“UGC-style” ads borrow the look of a real person filming themselves: phone held at arm’s length, a kitchen or car in the background, a direct-to-camera voice that sounds like a recommendation rather than a script. The format works because it reads as a person, not a brand.

AI UGC takes that look and generates it. A tool produces a presenter, a room, a voice and a performance from a text brief — no shoot, no booking, no travel. The output can be convincing enough that most viewers would never ask whether the person exists.

That is exactly where the trouble is. The format’s power comes from implied authenticity, and a synthetic version of it is borrowing something it does not have. This article is about the difference between using the format honestly and using it to imply things that are not true.

Three very different things get called “AI UGC”

Before anything else, be precise about which one you are buying. The rules land differently on each.

Fully synthetic creator. A presenter generated from scratch, resembling no particular real person, in a generated setting. Nobody filmed anything.

Licensed likeness of a real person. A real, identifiable adult has agreed that an AI version of them may appear for your brand, within a defined scope. There is a person behind the face, and a record of what they agreed to — this is what an AI likeness licence is applied to the UGC format.

Real footage, AI-processed. A genuine creator filmed genuine video, and AI was used to dub it into another language, change a line, swap a product shot or clean up delivery.

They are not interchangeable. The first cannot hold an opinion. The second can, under conditions. The third can, but the editing can still change what the person is understood to have said. Our comparison of avatars, licensed creators and conventional shoots covers the production trade-offs; this piece is about what you are allowed to imply.

Why brands reach for it

The honest reasons are about production, not persuasion.

Volume and iteration: you can generate many variants of a hook, offer or opening line and test them without rebooking anyone. Markets: one brief can become several language versions. Control: the presenter never goes off-script, never posts something you would rather they had not, and never becomes unavailable. Continuity: a product change does not strand your footage, because the footage can be regenerated.

Those are real operational advantages. None of them is a reason to misrepresent who is speaking.

The testimonial line is the hard limit

This is the part most AI UGC briefs get wrong, and it is not a grey area.

The FTC’s endorsement guidance is blunt: “You can’t talk about your experience with a product if you haven’t tried it.” A generated presenter has not tried anything. There is nobody there to have an experience, so there is no honest “I use this every morning”, no “it cleared up my skin in a week”, and no “I was sceptical at first”.

The Commission’s rule on consumer reviews and testimonials made this explicit rather than merely implied. Announced in August 2024, it prohibits “reviews and testimonials that misrepresent that they are by someone who does not exist, such as AI-generated fake reviews”, along with testimonials by people lacking genuine experience with the product, and it reaches businesses that create, buy, sell or disseminate them (FTC). A synthetic presenter delivering a first-person product experience is close to the centre of that description.

The position is similar elsewhere. In the UK, the CAP Code requires marketers to “hold documentary evidence that a testimonial or endorsement used in a marketing communication is genuine” and to keep contact details for the person who gave it (ASA). You cannot hold contact details for someone a model generated. In Australia, the ACCC ran a sweep of influencer advertising specifically to establish “the nature and extent of advertising practices which raise concerns under the Australian Consumer Law”, and encouraged businesses and influencers to consider changes to their practices “to ensure that consumers are not misled” (ACCC).

So the practical rule for synthetic presenters: they can explain, demonstrate and state claims you can substantiate. They cannot vouch. Write the script first, then ask whether the route you have chosen can lawfully say those words — see how to write a script for an AI spokesperson ad.

A licensed likeness can carry a genuine endorsement, but only if two separate things are true: the person has actually used the product, and they have specifically agreed to that testimonial. A likeness licence on its own is permission to appear, not permission to recommend.

Ad recognition: the second problem

Even a scrupulously factual AI UGC ad can fail a different test, because the format is designed to look like something other than advertising.

The CAP Code requires that “marketing communications must be obviously identifiable as such”, and that they “must not falsely claim or imply that the marketer is acting as a consumer or for purposes outside its trade, business, craft or profession” (ASA). A generated presenter who appears to be an ordinary customer sharing a find, made and placed by the brand, is the advertiser posing as a consumer. The paid-media label helps, but the creative itself should not work against it.

The same logic applies to the ongoing “creator” persona. A recurring fictional character fronting a brand is a known advertising device and is fine when the audience can tell. An account presented as a real person’s account, posting as though they were an independent customer, is not the same thing.

Disclosure: assume you need a label

Both the fully synthetic and the licensed-likeness routes attract transparency duties. Plan them into the creative rather than bolting them on.

YouTube requires creators to disclose content that “makes a real person appear to say or do something they didn’t do”, “alters footage of a real event or place”, or “generates a realistic scene that didn’t actually occur”, while exempting “non-realistic content that’s made with AI, or edits to realistic content that are minor” such as beauty filters and colour adjustment (YouTube Help). Note that the third limb catches a fully synthetic presenter in a realistic kitchen even though no real person is depicted.

Meta says it requires people to “disclose, using our AI-disclosure tool, whenever they post organic content with photorealistic video or realistic-sounding audio that was digitally created or altered” (Meta).

TikTok’s advertising policy allows significantly edited media and AI-generated content where advertisers “apply the AIGC label, or by adding a clear disclaimer, caption, watermark, or sticker of your own”, and treats as significant anything “completely AI-generated”, anything “showing the primary subject doing something they didn’t actually do”, and “making the primary subject say something they didn’t actually say, using AI voice-cloning”. It also prohibits content that “uses or promotes the misuse of the likeness (audio or visual) of a public figure without their permission, particularly in misleading contexts” (TikTok Ads Policy).

Our platform-by-platform disclosure guide sets out where each switch lives. In the EU, Article 50 of the AI Act adds machine-readable marking duties on providers and a disclosure duty on deployers of deep fakes (Article 50) — see our Article 50 guide. Carrying content credentials in the file helps the label survive a re-upload.

Whose face is that, actually?

“It’s synthetic” is not an answer to “who is that?”.

Generated presenters are built from training data, and some tools build library avatars from real people who sat for a capture session. Ask what those people agreed to: advertising use, your product category, your countries, and how long. A face that merely resembles someone recognisable can still attract a personality-rights claim, and the newer US digital replica statutes raise the stakes — see digital replica laws in the US.

Also ask what happens afterwards: whether the presenter is exclusive to you or shared with competitors, whether it can be withdrawn from the library mid-campaign, and what indemnity you actually hold if a claim arrives.

Before anything goes live

  • Which of the three routes is this, documented per asset
  • No first-person product experience in the script unless a real user agreed to it
  • Every factual claim substantiated independently of the presenter
  • Material connections disclosed where a real person is involved
  • AI disclosure applied on each platform, and the creative designed around the label
  • Obviously identifiable as advertising, and not posing as an ordinary consumer
  • Consent record for any real face, voice, name or mannerism used
  • Territory, category and expiry dates for every presenter in a calendar
  • A named person who can pull the asset quickly if consent is withdrawn

Where Likeness fits in

Likeness is being built for the middle route: real, verified adults who agree to specific brands and uses, with a licence record a business can point to, labelled output, and a payment each time a licence is issued or renewed. It is in development and not yet open — see for businesses and trust and consent, or email [email protected].


General information only, not legal advice, and no claim that any format or workflow is compliant with any rule. Advertising codes, consumer law and platform policies differ by country and change often — check the current versions for your markets before you run an AI UGC campaign.

Sources

Previous post
Next post