Marketing in 3 · Issue 21 · Week of 1 June 2026
They punish you for it. Across three American studies, more than half of viewers worked out on their own that the presenter in a customer-style video advert was an AI avatar, and those who did scored the ad 2.83 out of 7 against 4.59 among the people who never noticed. Being told afterwards was a bigger surprise but a far gentler one, because the technology still read as clever. A short briefing beforehand made spotting it markedly more likely, which is the warning in the whole paper.
The cheapest performer in advertising right now is an invented one, filmed talking about a product nobody ever bought, in the style of a customer who cannot wait to tell you about it. Jing Yang and Susanna Lee, both at Boston University, went looking for what happens at the moment the illusion fails.
They built two of these ads on a commercial platform, using the same AI avatar, a woman in her twenties or thirties. One was a 15 second vertical video about customisable family portraits, loose enough to pass as lifestyle content. The other was a 30 second piece for a real deodorant brand, built like a straight product review. Both were deliberately good rather than flawless, which a separate check confirmed.
Then they sorted viewers by how they found out. Some never twigged. Some worked it out on their own, which the authors call natural recognition. Some were told afterwards. A third study asked whether a short briefing beforehand changes how many people spot it, which is the closest thing here to a forecast.
Start with how leaky this format already is. A clear majority of the first study's viewers worked out that the presenter was synthetic with no prompt at all. Not a minority of experts, ordinary people watching a short video.
What happened to those people is the finding. Their opinion of the ad sat at 2.83 out of 7, while viewers who took the presenter at face value gave 4.59, and intention to buy dropped alongside it. The gap in attitude was the larger of the two, which fits the authors' account of the mechanism: the ad broke a promise rather than failing to sell.
That promise is specific to the format. A polished brand film is expected to be staged, so nothing is violated when it looks it. A person talking into a phone about a product they bought is supposed to be a person, and viewers who discover otherwise report feeling lied to, not merely unimpressed.
| Group | Value (out of 7) |
|---|---|
| Attitude: did not notice | 4.59 |
| Attitude: spotted it | 2.83 |
| Would buy: did not notice | 3.52 |
| Would buy: spotted it | 2.81 |
Study 1: 187 US adults, of whom 103 identified the presenter as synthetic without being prompted.
The second study is the counterintuitive one. Viewers who missed it entirely and were then informed registered a bigger jolt, rating the revelation as further from what they expected, 4.08 against 3.57. By the logic of surprise alone they should have reacted worse. They did not.
The reason sits in a second measure. People who were told found the whole thing considerably more novel, and novelty pulled the sting out of the surprise. Told about it, the reaction is closer to "I had no idea they could do that". Catching it yourself produces something nearer to "they thought I would not notice".
The practical reading is uncomfortable for anyone planning a campaign around this. The gentler path depends on the audience finding the technique impressive, which is a wasting asset. Novelty by definition does not survive becoming normal.
The third study is the one to show your head of brand. Half the viewers read a short description of how advertisers make these videos, plus a post from someone who had made one, before they watched anything. That briefing alone took the share who spotted the avatar from 46% to 58%.
Nothing about the ad changed. Only what the audience knew. The authors treat this as a stand in for the real world, where the same knowledge arrives for free through news coverage, platform labels and people posting about it.
So the economics of the tactic point one way. The pool of viewers who will take your synthetic customer at face value is the largest it will ever be today, and the penalty you pay when the rest of them catch on is already measurable.
| Group | Value (%) |
|---|---|
| Briefed beforehand | 58.3 |
| No briefing | 45.8 |
Study 3: 238 US adults, half of whom read a short description of how brands make these ads.
The damage is concentrated in formats that promise a real human experience. Testimonials, reviews, unboxing, influencer endorsements, before and after stories. Those are precisely the places where AI avatars are cheapest to use and most tempting, and they are the places where being caught costs the most.
Stretching the finding past what was tested, the same logic should hold wherever authenticity is the product rather than the packaging. A founder's note that was never written by the founder, a customer service reply signed by a name with nobody behind it, a case study built from a composite client. None of these were studied here, and all of them carry the same structure: a promise of a person, and a reveal waiting to happen.
The authors' own recommendation is blunt and worth repeating. If you are going to use synthetic people in a format that implies a real one, say so before the audience finds out, because the version where they tell you is the expensive one.
Marketing in 3 steps
Most of them did here. Watching a short customer-style video, a clear majority of participants identified the presenter as machine-made without being asked to look for it (Yang and Lee, 2026, Journal of Retailing and Consumer Services).
Attitude towards the ad fell by more than a point and a half on a seven-point scale, and willingness to buy dropped with it, although by a smaller margin.
Disclose. Viewers who were informed afterwards reacted far less harshly than those who caught it themselves, because the technique still struck them as novel rather than sneaky.
Because the format promises a real person. Spotting the deception yourself triggers a sense of betrayal, while a disclosure lets people process the same fact as a piece of clever technology.
The evidence points the other way. Simply explaining the practice beforehand made people substantially better at catching it, and that same awareness is now spreading through news coverage and platform labels without anyone's help.
Psychology & Marketing · 2026
Le, K., Khan, H., Li, F., & Kunz, W.
Four studies on what a disclosure should actually contain. Spelling out the role AI played in making the ad raised the credibility of the production process, and that lifted attitudes towards the ad. The effect was stronger when the disclosure looked like a response to rules rather than a boast about innovation.
Journal of Interactive Marketing · 2026
Dagogo-Jack, S., Nakos, K., & Kaju, A.
Five studies find that people who feel they are being fake on social media become more willing to delete the account entirely. The mechanism is self-threat, with deletion as the way to resolve it. A useful counterpoint for platforms that measure authenticity only as something audiences demand of brands.
Yang, J., & Lee, S. S. (2026). Caught in the act: Natural recognition of deepfake UGC ad, expectancy violation and consumer responses. Journal of Retailing and Consumer Services, 88, 104535. https://doi.org/10.1016/j.jretconser.2025.104535
This brief is written for practitioners from the published article. Numbers are as reported by the authors; read the article for full methods.
More issues on AI · Advertising · UGC
Posts carrying TikTok's AI label collected 7% to 8% fewer likes at the same reach. Not because people distrust AI. Because the label says the creator did not try.
Handing customers a ready-made post lifted sharing from 0.4% to 2.1% in a real gym campaign, and quietly stopped unhappy customers posting at all. How to use it, and where it gets uncomfortable.
A plain "this ad is deepfaked" label changed nothing. Adding one sentence about what that means cut perceived truthfulness from 3.33 to 2.41, without costing a single sale.
Bruno Schivinski is Associate Professor of Marketing at Gdańsk University of Technology. He studies how consumers engage with brands in digital and social media.
