You have scrolled past dozens of them this week without knowing it. A warm, ordinary-looking person talks straight into their phone camera, in a kitchen or a car or a bedroom with slightly messy shelves behind them, and tells you why they love a product. They sound like a real customer. They sound like a scrappy creator who just had to share this. And they do not exist. The face was generated. The voice was generated. The room was generated. Nobody was ever there.
The technology to do this got good, fast, and cheap almost overnight, and it put every small-business owner in front of the same two questions. Should I be doing this? And if I do, will people resent me when they figure out the person was never real? Both are fair questions, and the honest answer to both comes down to a single distinction that most of the people rushing into AI UGC have not stopped to draw. Get that distinction right and this is one of the most useful new tools you have. Get it wrong and it does something worse than waste your money.
Yes, This Is a Real Unlock, and I Am Not Going to Tell You to Avoid It
Let me start by defending the thing, because the reflexive take in a lot of marketing circles is snobbery, and snobbery will cost you.
Creator-style video, the loose handheld testimonial format that has quietly become the highest-converting kind of ad on Meta and TikTok, used to be genuinely hard for a small business to make at volume. You needed a real person willing to be on camera, or a creator who charged a real fee, or an actor and a half-day shoot. Every fresh angle you wanted to test meant more money and more calendar. So most small businesses tested two or three creator-style ads a quarter, if that, and called it a program.
AI collapses that cost to almost nothing. You can now produce a dozen presenter-style videos, each with a different hook, a different opening line, a different framing of the same offer, in an afternoon, for the price of a subscription. For a business that could never previously afford this format at any real cadence, that is not a gimmick. That is a genuine expansion of what you can test and how fast you can learn. The teams treating AI presenters as just another way to generate more shots on goal, faster, are onto something real. I would never tell you to leave that on the table.
The problem is not the tool. The problem is what most people do with it the moment they get their hands on it.
Most of It Is Slop, and Slop Does Not Just Fail
Here is what actually happens on the average account. Someone discovers one of these tools, types in a rough idea, and lets it generate a generic person reading generic hype in a generic room. Then they run it. And two bad things happen at once.
The first is boring and expected: it does not convert. The person says nothing a real person would say, in the flat over-polished cadence these tools default to, about a product they obviously have no relationship with. It is an ad that smells like an ad, which is exactly the thing the creator format was supposed to escape. That failure is survivable. Most ads fail. That is why you test your way to a winner instead of betting on one.
The second thing is the one that should worry you. At some point during that flat little monologue, a chunk of your audience clocks it. The eyes are slightly dead. The mouth moves a half-beat wrong. The skin is too smooth, the delivery too even. And the viewer realizes, consciously or not, that the person is fake. That realization does not just make them scroll. It makes them feel, for a second, like someone tried to put one over on them. And that feeling does not stay contained to the ad. It reaches back and colors the whole business behind it. If the person is fake, what else is fake? The reviews? The guarantee? You reached for a shortcut to trust and spent trust to pay for it, which is the worst trade in marketing. It is the same mechanism that makes a fake countdown timer so expensive: the damage is not the failed tactic, it is the suspicion it spreads to everything around it.
The Line: Directed Versus Generated
So what separates the AI presenter that quietly works from the one that quietly poisons the account? It is not the tool. The same tool produces both. The difference is whether a human directed it or simply generated it.
Generated means you typed a rough prompt, took whatever the model handed back, and shipped it. No one decided who this person should be, what they should actually say, or why anyone would believe them. The machine made every choice, and the machine has no idea who your customer is.
Directed means the tool is just the camera, and you are still the one making the ad. You decided who the person is and why they fit your audience. You wrote what they say. You chose the tone. The AI executed a plan you built. That is the entire difference between the two outcomes, and it maps exactly onto the older truth that anyone can make an ad now, but almost nobody knows why one actually sells. AI made the person cheap. It did nothing at all to make the thinking cheap.
What Directing an AI Presenter Actually Means
Directing here is not mystical. It is four concrete decisions the tool will never make for you.
Casting comes first. The face is not decoration; it is a signal about who this product is for. A believable presenter for a premium skincare line and a believable presenter for a contractor-tools brand are not interchangeable, and the model's default good-looking stranger is usually right for neither. Choose a face that reads as your customer, not as a stock human.
Script comes second, and it is where most AI UGC dies. The tool will happily generate adjective soup, the best, the most amazing, you have to try this. Real people do not talk like that. Write one clear message, one concrete benefit, in the plain specific language an actual customer would use. One true, specific sentence beats a paragraph of superlatives every time.
Restraint is third. These tools default to a hyped-up, over-articulated delivery that screams commercial. Push the other way. Calmer, slower, more matter-of-fact almost always reads as more human, and more human is the entire point of the format.
Specificity is fourth. Generic is the tell. The details that make a presenter believable are the small ones a real brand would actually know: the specific situation the product is for, the exact objection it answers, the particular before-and-after. Feed those in. They are what the model cannot invent and what your audience is actually listening for.
The One Line You Cannot Cross
Everything above is about quality. This next part is about honesty, and it is not optional, because it is also the law.
There is a clean line between two uses of an AI person that look similar and are not. Using an AI presenter to explain your offer is fine. It is a spokesperson, no more deceptive than a voiceover, an illustrated mascot, or a hired actor reading a script in a normal commercial. Nobody is being told that character is a real customer, so nobody is being fooled.
Fabricating an AI customer who reviews your product is a different thing entirely. The moment your AI person says in effect I bought this and it changed my life, you have manufactured a fake testimonial from someone who does not exist. That is not an edgy growth tactic. Since October 2024, the FTC's Consumer Reviews and Testimonials Rule has made it a violation that carries civil penalties, and the rule names AI-generated fake reviews specifically. The FTC has been careful and clear about where the line sits: it says plainly that there is no blanket prohibition on using AI-generated avatars in marketing, and that an AI presenter only becomes illegal when the underlying testimonial is fake or false. Which is the same line common sense would draw. A fictional spokesperson is fine. A fictional customer is fraud.
Stay on the presenter side of that line and you are on solid ground, legally and ethically. Cross it and you are one screenshot away from a trust collapse that no conversion rate is worth.
When It Fits, and When to Skip It
Even done well and kept honest, AI UGC is not right for every job. Match the tool to the task.
It fits best at the top of the funnel, where you are buying attention and testing angles: explainer-style videos, hook testing at volume, quick variations on how you frame the offer. This is where cheap and fast genuinely compound, because you can test your way to what works at a cadence you could never afford with live shoots.
It fits worst anywhere authenticity is the whole point. If your edge is that you are a real local person people can walk in and meet, a synthetic stranger undercuts the one thing you have. If you are making claims in a regulated category, health, finance, anything with legal exposure, an AI mouth does not lower your burden of proof. And if the honest version of the ad requires a real customer's real experience, then use a real customer. There is no shortcut around that one.
The Tool Changed the Cost. It Did Not Change the Craft.
Strip away the novelty and here is what actually happened. AI made one expensive input, a person on camera, nearly free. That is a real and useful change, and pretending otherwise is just posturing.
But it left every hard part exactly where it was. Who should this person be. What should they honestly say. Where is the line between a spokesperson and a lie. Those were always the questions that separated creative that works from creative that just exists, and no model answers them for you. The businesses that win with AI presenters will not be the ones who generate the most. They will be the ones who direct, the way the tool making it cheap to produce a thousand ads still cannot tell you which one is worth making. That judgment, who to be and what to say and where the honest line runs, is the whole job. It is what we do when we build creative for our clients, whether the camera is real or not.
