How to Spot a Deepfake or AI-Generated Influencer Endorsement

Fake follower farms used to be the cheapest way to fake influence. In 2026, the cheaper trick is faking the influencer entirely — a cloned voice, a synthetic face, or a fully AI-generated persona putting words in a real creator’s mouth (or replacing them altogether). Here’s how to tell when an endorsement isn’t what it claims to be, before you act on it.

This isn’t a hypothetical — it’s already widespread

According to McAfee’s 2025 Most Dangerous Celebrity: Deepfake Deception survey of 8,600 adults across seven countries, 72% of Americans have seen a fake or AI-generated celebrity endorsement online, and 39% clicked one — often before realizing anything was off. Among those who lost money, the average financial hit was $525. Taylor Swift topped the list of most-impersonated celebrities, followed by Scarlett Johansson and Jenna Ortega.

The pattern isn’t limited to fringe names:

• Tom Hanks publicly warned his Instagram followers in 2023 about a dental-plan ad circulating with “an AI version of me” that he had no involvement in (NBC News).
• MrBeast was deepfaked into a TikTok ad offering iPhone 15s for $2 — an offer he never made (Yahoo Finance).
• Taylor Swift has been the face of a recurring deepfake scam pushing discounted Le Creuset cookware, seeded across TikTok and Facebook (Tech.co; AOL).

None of these creators made these endorsements. Every one of them was manufactured to borrow trust the real person never gave.

Two different problems, both called “AI influencer”

It helps to separate what you’re actually screening for:

1. A real person, deepfaked without consent. A cloned voice or face is stitched onto footage to make it look like a real, known creator or celebrity endorsed something they never touched.
2. A fully synthetic persona. Accounts like Lil Miquela or Aitana Lopez are AI-generated from the start — no real person behind the face, run by a studio, and disclosed (with varying clarity) as virtual. These aren’t inherently scams; the fraud risk is in undisclosed synthetic personas, or in a virtual account’s content being passed off as a real person’s genuine experience with a product.

Both are covered by the same underlying question a brand or reader should ask: does the audience know what they’re looking at?

The tells that catch most fakes

Per warnings from the Better Business Bureau and CBS News, the same handful of tells catch the majority of deepfake endorsements:

Visual:

• Unnatural blinking (too much, or none at all)
• Blurry or doubled edges around the face, hairline, or jaw
• Lighting or shadows on the face that don’t match the background
• Lip movement that doesn’t quite sync with the audio

Audio:

• Choppy sentence rhythm or odd phrasing that doesn’t sound like how the person actually talks
• Flat or oddly-placed inflection, like sentences stitched from different clips

Context (often the fastest tell):

• The “deal” is too good — free or heavily discounted product for shipping cost only, urgency countdowns, “limited spots”
• The video only exists as a paid ad or on a brand-new, low-follower account — not on the creator’s own verified page
• Comments are disabled, or the account posting it isn’t the creator’s real, verified handle

That last one is the highest-leverage check and takes 30 seconds: go to the creator’s own verified account and see if they’ve posted it themselves. A real endorsement exists in more than one ad unit — it shows up on the creator’s actual feed, gets covered by outlets that cover that creator, or is confirmed in their own words. A deepfake usually exists in exactly one place: the ad you’re looking at.

What the law actually says

The FTC isn’t silent on this. Its Trade Regulation Rule on the Use of Consumer Reviews and Testimonials took effect October 21, 2024, on a unanimous 5–0 vote. It explicitly bans creating or selling fake celebrity endorsements that misrepresent who is actually endorsing the product — including AI-generated ones — and separately bans buying or selling fake indicators of social media influence, i.e., purchased followers or engagement. Violators face civil penalties that started at up to $51,744 per violation and, per the FTC’s inflation-adjusted 2025 penalty schedule (unchanged for 2026), now run up to $53,088 per violation.

Separately, the FTC’s revised Endorsement Guides require that any endorsement reflect the endorser’s honest opinion, and that any connection between an endorser and a marketer that consumers wouldn’t expect must be clearly disclosed. A synthetic persona pretending to be an ordinary, unpaid enthusiast — human or not — is exactly the kind of undisclosed connection the rule targets.

In other words: this isn’t just a trust problem for the audience. For a brand running the ad, it’s a compliance problem with real penalties attached.

Why “high engagement” won’t save you

If you’re vetting a creator and their engagement rate looks unusually strong, that alone isn’t proof of a real audience — it can be the opposite signal. As we found digging into a recent creator-economy report, AI-generated creator accounts posted engagement more than double the rate of comparable human creators in the same dataset. Engagement doesn’t distinguish “AI-generated account with real followers who know it’s AI” from “AI-generated content pretending to be a human’s genuine review.” Only disclosure and verification do that.

Verify before you trust it

For any creator making a claim on YouTube or Twitch — the two platforms with real, public data to check against — run the actual account through our free authenticity checker. It won’t catch a deepfaked video ad, but it will confirm whether the channel behind a claim has the growth pattern, engagement quality, and content consistency of a real, organically-grown audience, rather than one built to lend a fake endorsement borrowed credibility. Pair that with the manual checks above — checking the creator’s own page, watching for the too-good-to-be-true framing, and looking for lip-sync or audio artifacts — and you’ll catch the overwhelming majority of fakes before you act on them.

The bottom line

A deepfake or undisclosed AI endorsement only works if you act on it without checking. The fix costs nothing: go to the creator’s own account before you trust the ad, watch for the visual and audio tells, and treat “too good to be true” as the loudest signal of all. Check any creator — or any claim about one — before you trust them.