The Deepfake Ad Playbook: How to Spot a Fabricated Celebrity Endorsement

A doctor recommending a specific supplement on video looks like strong evidence — until you learn how cheap that video has become to fabricate. Over the past two years, independent fact-checkers and journalists have documented real cases of physicians, researchers, and public figures whose actual footage — from conferences, TV appearances, or parliamentary testimony — was altered with AI to make it look like they were endorsing products they'd never heard of. This isn't a hypothetical risk. It's a pattern with names, dates, and documented investigations attached.

Documented cases: real people, real footage, altered without consent

A Full Fact investigation (a UK-based independent fact-checking organization) traced a cluster of TikTok accounts using AI-altered footage of real doctors and academics to promote a supplement brand called Wellness Nest, primarily shilajit-based products marketed for menopause symptoms. Genuine video — pulled from real conferences, UK parliamentary select-committee testimony, and podcast appearances — was re-edited so the speakers appeared to endorse the products. Professor David Taylor-Robinson, a professor of children's public health who has no connection to menopause research, found himself appearing in a deepfake video discussing an invented symptom the video called "thermometer leg." That single video reached more than 365,000 views, 7,691 likes, and nearly 3,000 bookmarks before removal. Full Fact identified more than a dozen other real doctors and researchers targeted the same way, and TikTok itself only banned the source account after direct contact from investigators, at one point acknowledging the initial non-removal was "a moderating error."^[1]

Separately, Dr. Maurice Sholas, a pediatric rehabilitation specialist in New Orleans, discovered that AI-altered video of his face had been used to promote a vitamin supplement (marketed as "K2D3") he had never recorded any content for or endorsed — footage that circulated on TikTok and X and specifically targeted Black consumers, according to reporting on the case.^[2] Other physicians named in reporting on this pattern, including Dr. Robert Lustig and Dr. Gemma Newman, have described discovering AI-cloned versions of themselves promoting weight-loss drugs and vitamin products they never endorsed.^[2] These aren't isolated incidents — they're specific, named, reported cases of the same underlying tactic: take real public-appearance footage of a credentialed person, alter it with AI, and attach it to a product.

This sits inside a larger surge. The UK's Advertising Standards Authority logged 177 scam-ad alerts involving deepfaked public figures (including senior politicians and broadcasters) in 2024 alone, and separate reporting has tracked AI-deepfake celebrity ad scams — using the likenesses of figures like Taylor Swift, Kim Kardashian, and others — spreading across platforms including TikTok.^[3]

Why this works on viewers

The technique is effective precisely because the starting material is real. These aren't cartoonish AI avatars — they're a real, recognizable person's real face and often their real voice, lifted from an actual public appearance, with only the words and framing altered. That makes the usual "does this look fake" instinct much less reliable than it used to be, because the visual baseline is genuine footage of a genuine expert.

Practical signals for spotting a likely-fabricated endorsement

• Check the person's own verified channels first. A real endorsement from a well-known physician or public figure will typically also appear on their own official website, verified social accounts, or be covered by mainstream press. If a supplement endorsement exists only as a single video circulating on TikTok or in a Facebook ad, with no matching statement anywhere the person actually controls, treat it as unverified.
• Watch the mouth and listen for audio drift. Fact-checkers point to specific tells: lip movement that doesn't quite match hard consonant sounds like "p," "b," and "m" (which require a full lip closure that AI lip-sync often misses), flat or oddly paced vocal delivery, missing natural breath sounds, and background audio that doesn't stay consistent through the clip.^[4]
• Look for the same "expert" endorsing many unrelated products. The documented cases above involved the same handful of altered clips being reused across multiple different supplement brands and claims. A real expert endorsing a rotating cast of unrelated products, often outside their actual specialty (a pediatric specialist pitching vitamins, a children's-health professor discussing menopause), is a strong signal the "endorsement" was manufactured rather than given.
• Try a reverse video or image search. Taking a still frame from the video and running it through a reverse-image search tool can sometimes surface the original, unaltered footage — the real conference talk or interview the clip was built from — which will not mention the product at all.
• Be skeptical of urgency and discount-code framing. Several of the documented cases paired the fabricated endorsement with a limited-time discount code or claims that mainstream medicine was suppressing the product — a combination of manufactured urgency and conspiratorial framing that legitimate medical endorsements essentially never use.

What this doesn't mean

None of this means every product video featuring a real expert is fake, or that video evidence is now worthless. It means the bar for treating a video endorsement as reliable evidence has to include actually checking whether the person said what the video shows them saying, through a source you don't have to take the ad's word for. That verification step used to be optional. Based on the documented cases above, it no longer is.