Bad actors: YouTube ads have an AI video problem

For the past 40 years, Henry and Margaret Tanner have been crafting leather shoes by hand from their small workshop in Boca Raton, Florida. "No shortcuts, no cheap materials, just honest, top notch craftsmanship," Henry says in a YouTube advertisement for his business Tanner Shoes.

What's even more remarkable?

Henry has been able to do all this despite his mangled, twisted hand. And poor Margaret only has three fingers, as you can see in this photo of the couple from their website.

I discovered Tanner Shoes through a series of YouTube video ads. Having written about men's fashion for years, I was curious about these bespoke leather shoemakers. In a typical YouTube ad for Tanner Shoes, a video of an older man, presumably Henry, is imposed over footage of "handmade" leather shoes, as he wearily intones, "They don’t make them like they used to, but for 40 years we did...Customers say our shoes have a timeless look, and that they’re worth every penny. But now, you won’t have to spend much at all because we’re retiring. For the first and last time, every last pair is 80 percent off."

I suspect the Tanner Shoes "retirement" sale is every bit as real as the photos of Henry and Margaret Tanner. Outside of this advertisement, I have found no online presence for Henry and Margaret Tanner and no evidence of the Tanner Shoes business existing in Boca Raton. I reached out to Tanner Shoes to ask if its namesake owners exist, where the company is located, and if it's really closing soon, but I have not received a response.

Unsurprisingly, Reddit users have spotted nearly identical YouTube video ads for other phony mom-and-pop shops, showing that these misleading ads aren't a one-off. As one Reddit user said, "I've seen ads like this in German with an AI grandma supposedly closing her jewelry store and selling her 'hand-made' goods at a discount." After I asked YouTube about the Tanner Shoes ads, the company suspended the advertiser's account for violating YouTube policies.

These ads are part of a growing trend of YouTube video advertisements featuring AI-generated content. AI video ads exist on Instagram and TikTok too, but as the original and most well-established video platform, I focused my investigation on YouTube, which is owned by Google.

While AI has legitimate uses in advertising, many of the AI video ads I found on YouTube are deceptive, designed to trick the viewer into buying leather shoes or diet pills. While reliable stats on AI scams are hard to find, the FBI warned in 2024 that cybercrime utilizing AI is on the rise. Overall, online scams and phishing have increased 94 percent since 2020, according to a Bolster.ai report.

AI tools can quickly generate lifelike videos, pictures, and audio. Using tools like this, scammers and hustlers can easily create AI "actors," for lack of a better word, to appear in their ads.

In another AI video ad Mashable reviewed, an AI actor pretends to be a financial analyst. I received this advertisement repeatedly over a series of weeks, as did many Reddit and LinkedIn users.

In the video, the nameless financial analyst promises, "I'm probably the only financial advisor who shares all his trades online," and that "I've won 18 of my last 20 trades." Just click the link to join a secret WhatsApp group. Other AI actors promise to help watchers discover an amazing weight loss secret ("I lost 20 pounds using just three ingredients I already had in the back of my fridge!"). And others are just straight-up celebrity deepfakes.

Celebrity deepfakes and deceptive AI video ads

I was surprised to find former Today host Hoda Kotb promoting sketchy weight loss tricks on YouTube, but there she was, casually speaking to the camera.

"Ladies, the new viral recipe for pink salt was featured on the Today show, but for those of you who missed the live show, I'm here to teach you how to do this new 30-second trick that I get so many requests for on social media. As a solo mom of two girls, I barely have time for myself, so I tried the pink salt trick to lose weight faster, only I had to stop, because it was melting too fast."

This fake Kotb promises that even though this weight loss secret sounds too good to be true, it's definitely legit. "This is the same recipe Japanese celebrities use to get thin. When I first learned about this trick, I didn't believe it either. Harvard and Johns Hopkins say it's 12 times more effective than Mounj (sic)...If you don't lose at least four chunks of fat, I'll personally buy you a case of Mounjaro pens."

Click the ad, and you'll be taken to yet another video featuring even more celebrity deepfakes and sketchy customer "testimonials." Spoiler alert: This video culminates not in the promised weight loss recipe, but in a promotion for Exi Shred diet pills. Representatives for Kotb didn't respond to a request for comment, but I found the original video used to create this deepfake. The real video was originally posted on April 28 on Instagram, and it was already being used in AI video ads by May 17.

Kotb is just another victim of AI deepfakes, which are sophisticated enough to slip past YouTube's ad review process.

Sometimes, these AI creations appear real at first, but pay attention, and you'll often find a clear tell. Because the Kotb deepfake used an altered version of a real video, the fake Kotb cycles through the same facial expressions and hand movements repeatedly. Another dead giveaway? These AI impersonators will often inexplicably mispronounce a common word.

The AI financial analyst promises to livestream trades on Twitch, only it mispronounces livestream as "give-stream," not "five-stream." And in AI videos about weight loss, AI actors will trip up over simple phrases like "I lost 35 lbs," awkwardly pronouncing "lbs" as "ell-bees." I've also seen phony Elon Musks pronounce "DOGE" like "doggy" in crypto scams.

However, there isn't always a tell.

Can you tell what's real? Are you sure?

Once I started investigating AI video ads on YouTube, I began to scrutinize every single actor I saw. It's not always easy to tell the difference between a carefully airbrushed model and a glossy AI creation, or to separate bad acting from a digitally altered influencer video.

So, every time YouTube played a new ad, I questioned every little detail — the voice, the clothes, the facial tics, the glasses. What was real? What was fake?

Surely, I thought, that's not Fox News host Dr. Drew Pinsky hawking overpriced supplements, but another deepfake? And is that really Bryan Johnson, the "I want to live forever" viral star, selling "Longevity protein" and extra virgin olive oil? Actually, yes, it turns out they are. Don't forget, plenty of celebrities really do appear in commercials and YouTube ads.

Okay, but what about that shiny bald man with a super secret technique for lowering cholesterol that the pharmaceutical companies don't want you to know about? And is that girl-next-door type in the glasses really selling software to automate my P&L and balance sheets? I genuinely don't know what's real anymore.

Watch enough YouTube video ads, and the overly filtered models and influencers all start to look like artificial people.

To make matters more complicated, most of the AI video ads I found on YouTube didn't feature characters and sets created from scratch.

Rather, the advertisers take real social media videos and change the audio and lip movements to make the subjects say whatever they want. Henry Ajder, an expert on AI deepfakes, told me that these types of AI videos are popular because they’re cheap and easy to make with widely available synthetic lip synchronization and voice cloning tools. These more subtle AI videos are virtually impossible to definitively identify as AI at a glance.

"With just 20 seconds of a person’s voice and a single photograph of them, it is now possible to create a video of them saying or doing anything," Hany Farid, a professor at the University of California Berkeley and an expert in artificial intelligence, said in an email to Mashable.

Ajder told me there are also multiple tools for “the creation of entirely AI-generated influencer style content.” And just this week, TikTok announced new AI-generated influencers that advertisers can use to create AI video ads.

YouTube is supposed to have solutions for deceptive ads. Google's generative AI policies and YouTube's rules against misrepresentation prohibit using AI for "misinformation, misrepresentation, or misleading activities," including for "Frauds, scams, or other deceptive actions." The policies also forbid "Impersonating an individual (living or dead) without explicit disclosure, in order to deceive."

So, what gives?

Consumers deserve clear disclosures for AI-generated content

For viewers who want to know the difference between reality and unreality, clear AI content labels in video advertisements could help.

When scrolling YouTube, you may have noticed that certain videos now carry a tag, which reads "Altered or synthetic content / Sound or visuals were significantly edited or digitally generated." Instead of placing a prominent tag over the video itself, YouTube typically puts this label in the video description.

You might assume that a video advertisement on YouTube generated by AI would be required to use this disclosure, but according to YouTube, that's not actually the case.

Using AI-generated material doesn’t violate YouTube ad policies (in fact, it's encouraged), nor is disclosure required in most cases. In fact, YouTube only requires AI disclosures for ads that use AI-generated content in election-related videos or political content.

In response to Mashable’s questions about AI video ads, Michael Aciman, a Google Policy Communications Manager, provided this statement: “We have clear policies and transparency requirements for the use of AI-generated content in ads, including disclosure requirements for election ads and AI watermarks on ad content created with our own AI tools. We also aggressively enforce our policies to protect people from harmful ads — including scams — regardless of how the ad is created.”

There's another reason why AI video ads that violate YouTube's policies slip through the cracks — the sheer volume of videos and ads uploaded to YouTube each day. How big is the problem? A Google spokesperson told Mashable the company permanently suspended more than 700,000 scam advertiser accounts in 2024 alone. Not 700,000 scam videos, but 700,000 scam advertiser accounts. According to Google's 2024 Ads Safety Report, the company stopped 5.1 billion "bad ads" last year across its expansive ad network, including almost 147 million ads that violated the misrepresentation policy.

YouTube’s solution to deceptive AI content on YouTube? More AI, of course. While human reviewers are still used for some videos, YouTube has invested heavily in automated systems using LLM technology to review ad content. “To address the rise of public figure impersonation scams over the last year, we quickly assembled a dedicated team of over 100 experts to analyze these scams and develop effective countermeasures, such as updating our Misrepresentation policy to suspend the advertisers that promote these scams," a Google representative told Mashable.

After I asked the company about specific AI videos described in this article, YouTube suspended at least two advertiser accounts; users can also report deceptive ads for review.

However, while celebrity deepfakes are a clear violation of YouTube's ad policies (and federal law), the rules governing AI-generated actors and ads in general are far less clear.

AI video isn't going away

If YouTube fills up with AI-generated videos, you won't have to look far for an explanation. The call is very much coming from inside the house. At Google I/O 2025, Google introduced Veo 3, a breakthrough new model for creating AI video and dialogue. Veo 3 is an impressive leap forward in AI video creation, as I've reported previously for Mashable.

To be clear, Veo 3 was released too recently to be behind any of the deceptive videos described in this story. On top of that, Google includes a hidden watermark in all Veo 3 videos for identification (a visual watermark was recently introduced as well). However, with so many AI tools now available to the public, the volume of fake videos on the web is certain to grow.

One of the first Veo 3 viral videos I saw was a mock pharmaceutical ad. While the faux commercial was meant to be humorous, I wasn't laughing. What happens when a pharmaceutical company uses an AI actor to portray a pharmacist or doctor?

Deepfake expert Henry Ajder says AI content in ads is forcing us to confront the deception that already exists in advertising.

"One of the big things that it's done is it's held up a looking glass for society, as kind of how the sausage is already being made, which is like, 'Oh, I don't like this. AI is involved. This feels not very trustworthy. The feels deceptive.' And then, 'Oh, wait, actually, that person in the white lab coat was just some random person they hired from an agency in the first place, right?'"

In the United States, TV commercials and other advertisements have to abide by consumer protection laws and are subject to Federal Trade Commission regulations. In 2024, the FTC passed a rule banning the use of AI to impersonate government and business agencies, and Congress recently passed a law criminalizing deepfakes, the "Take It Down" Act. However, many AI-generated videos fall into a legal grey area with no explicit rules.

It's a tricky question: If an entire commercial is made with AI actors and no clear disclosure, is that advertisement definitionally deceptive? And is it any more deceptive than hiring actors to portray faux pharmacists, paying influencers to promote products, or using Photoshop to airbrush a model?

These are no longer hypothetical questions. YouTube already promotes using Google AI technology to create advertising materials, including video ads for YouTube, to "save time and resources." In a blog post, Google promotes how its "AI-powered advertising solutions can assist you with the creation and adaptation of videos for YouTube’s wide range of ad formats." And based on the success of Google Veo 3, it seems inevitable that platforms like YouTube will soon allow advertisers to generate full-length ads using AI. Indeed, TikTok recently announced exactly this.

The FTC says that whether or not a company must disclose that it's using "AI actors" depends on the context, and that many FTC regulations are "technology neutral."

"Generally speaking, any disclosures that an advertiser would have to make about human actors (e.g., that they are only an actor and not a medical professional) would also be required for an AI-generated persona in an analogous situation," an FTC representative with the Bureau of Consumer Protection told Mashable by email.

The same is true for an AI creation providing a "testimonial" in an advertisement. "If the AI-generated individual is providing a testimonial (which would necessarily be fake) or claiming to have specific expertise (such as a medical degree or license or financial experience) that affects consumers’ perception of the speaker’s credibility, that may be deceptive," the representative said.

The FTC Act, a comprehensive statute that governs issues such as consumer reviews, prohibits the creation of fake testimonials. And in October 2024, the FTC regulation titled "Rule on the Use of Consumer Reviews and Testimonials" specifically banned fake celebrity testimonials.

However, some experts on deepfakes and artificial intelligence believe new legislation is urgently needed to protect consumers.

"The current U.S. laws on the use of another person’s likeness are — at best — outdated and were not designed for the age of generative AI," Professor Farid said.

Again, the sheer volume of AI videos, and the ease of making them, will make enforcement of existing rules extremely difficult.

"I would go further and say that in addition to needing federal regulation around this issue, YouTube, TikTok, Facebook, and the others have to step up their enforcement to stop these types of fraudulent and misleading videos," Farid said.

And without clear, mandatory labels for AI content, deceptive AI video ads could soon become a fact of life.