AI UGC ads: how they are made, where they work for testing and which disclosure rules apply

AI UGC is an ad dressed as a user video

UGC means user-generated content: a phone video where an ordinary person shows a product and says what they think of it. Advertisers like the format because it looks unlike an advertisement and holds attention in a feed. AI UGC copies that look with a presenter who is generated by software and speaks a written script in a synthetic voice. The result is a short vertical video that feels like a person talking to the camera, though no such person made it.

That difference is the whole subject of this story. The format is cheap to repeat, so it is useful for testing. It also sits close to the edge of what readers expect from a testimonial, so it comes with rules about honesty and labelling. Both halves matter before any money goes into distribution.

How a typical AI UGC video is made

The work starts with the script, not the software. A short ad has a hook in the first seconds, a problem the viewer recognises, the product as the answer, and one clear action at the end. A writer drafts several hooks for the same ad, because the opening is the part that decides whether the viewer stays. A hook can start with a question, a complaint or a surprising claim, and each is a separate version to test.

Then the presenter and the voice are chosen from generated options, the script is spoken, the footage of the product is added, captions are placed over it and the clip is cut to the length the platform favours. The product itself should appear as it really is. Software can place a bottle in a generated hand, but it cannot make a product do what it does not do.

Where AI UGC works: testing hooks before a bigger production

The strongest use is testing. A paid campaign is a series of guesses about which message, which opening and which promise will make a stranger stop scrolling. Producing each guess with a real creator takes days and a fee. Producing ten variations of one script with generated presenters takes far less, so a small business can learn which hook gets the clicks before it commits to a shoot.

It also suits products that are easy to show and easy to explain: a household item, an app, a simple accessory, a service with a clear promise. In these cases the viewer needs a clear idea more than a personality, and a generated presenter delivers an idea well enough.

It works less well when the product is the evidence, as with a cream whose effect people want to see, or when the buyer wants to hear from someone like themselves. The ads are a tool for finding what to say, and the winning message can then be reshot with a real person if the budget allows.

The honesty line: a presenter cannot have used what they praise

The first rule is not about technology at all. If the AI presenter says I have used this for three months and nobody has, the sentence is a false testimonial, whoever or whatever speaks it. The FTC’s questions and answers on its rule about consumer reviews and testimonials, which took effect on October 21, 2024, say the rule prohibits writing, creating or selling fake or false consumer reviews or testimonials.

The same FTC page addresses avatars directly. It says the rule has no blanket ban on AI avatars in marketing, and that an avatar testimonial is prohibited only if the underlying testimonial is fake or false. It adds that using a celebrity’s avatar without permission can violate the rule if consumers would reasonably think the celebrity endorsed the product, and that avatar use may also be deceptive under the FTC Act. The page is staff guidance, not a final interpretation.

The practical reading is easy to follow. Let the presenter state facts about the product that are true and can be shown, such as what it is, how it works and what it costs. Do not let the script invent a personal history, a result or a customer who does not exist. And do not borrow the face or voice of a real person without permission. For your own case, ask a lawyer who knows advertising law.

Platform labels: what TikTok, Meta and YouTube say about AI content

TikTok’s ads policy allows significantly edited or AI-generated ads if they carry the AIGC label or the advertiser adds a clear disclaimer, caption, watermark or sticker. It counts content that is completely AI-generated, and real footage that AI has substantially altered, within that rule. Lighting, brightness, background removal and denoising are listed as insignificant edits. If unsure, the policy advises labelling, and undisclosed AI content that TikTok identifies can be rejected or restricted.

TikTok’s Ads Manager help describes an AI-generated content disclaimer that appears at the bottom of an in-feed ad for its whole length, and says it works with both Spark Ads and non-Spark Ads. The same policy page also bars exaggerated product results, including before-and-after comparisons, so a label does not fix a misleading claim.

Meta’s help page says it applies an AI info label to ad images created or materially edited with its own generative AI creative features, and that ads made with third-party tools are not covered by that label. It adds that advertisers must disclose digitally created or altered media in ads about social issues, elections or politics, including media made with third-party AI tools. YouTube’s help page asks creators to disclose realistic content made or meaningfully altered with AI, such as a real person appearing to say something they did not say, and exempts minor edits and non-realistic content. These pages change, so read the current text first.

Testing ten variations without fooling yourself

A useful test changes one thing at a time. Keep the product, the offer and the end action the same and vary the hook across ten versions, or keep the hook and vary the presenter. If two things change together, the result tells you nothing. Give every version the same audience, the same budget and the same period, and decide beforehand what counts as a win: cost per click, cost per order or watch time.

Treat the winner as a lead, not a verdict. A hook that wins among cold viewers may not hold for returning customers, and a small difference between two versions may be chance. Repeat the winning idea once, then build on it. Nobody can tell you in advance which version will win, and a published number from another business says little about yours.

When a real creator is the better choice

Choose a real creator when the product has to be shown working in real hands, when the claim depends on someone actually using it over time, or when trust is the main obstacle: health, children, money, expensive purchases. The FTC’s guidance for influencers also points the same way: when someone has a material connection to a brand, such as a payment or a free product, the disclosure should be in the video itself, not only in the description, in plain words such as ad or sponsored.

A mixed approach is often sensible. Use generated videos to find the hook, the angle and the audience, then film the winning script with a real person who has really used the product. The cheap stage tells you what to say, and the real stage gives it a face.

What to prepare before ordering ten AI UGC videos

You can try this yourself with a script, a generation tool and an evening or two, and it is a fair way to learn the format. Ordering suits you when you want ten consistent versions with several hooks, written and cut for your platform, without learning the tools. Prepare the product facts that may be said aloud, the offer and the end action, two or three customer questions to use as hooks, product footage or photos, the platform where the ads will run and the claims you cannot make.

VITON13 Studio offers this as the 10 AI UGC ad videos package: ten short AI-made videos in the style of user videos, scripts with several hooks for testing, and AI presenters labelled as AI where platform rules require it. Before payment, the studio makes one finished video as a free draft, so you can judge the look and the script on your own product. The fixed price is $73 and the work takes 3–5 working days. Describe the product, and the draft comes first.