Brands are already using AI influencers to sell real products, and the disclosure rules are still chasing the feed.
The fake person in the ad is no longer a novelty act. The Guardian reported on June 21, 2026, that brands are using AI-generated influencers in social posts that look like ordinary customer videos, often without telling viewers the person on screen doesn't exist. If you're buying from a recommendation in your feed, that distinction isn't cosmetic. It tells you whether you're hearing from a person with a point of view or from an ad asset built to look like one.
The examples are not hidden in some experimental corner of the internet. The Guardian found promotional content for Once, a photo app used for event pictures, in which a crying bride appeared to praise the product on Instagram. Reality Defender, a cybersecurity company that works on deepfake detection, told the paper the brand had likely used AI-generated influencers. Once didn't respond to the Guardian's request for comment. Maket, an AI home-design app, said AI-generated influencers had been one of several ways it tested marketing hooks at a small scale.
That is the useful detail here. Brands are not only building glossy fictional celebrities like Lil Miquela, the CGI character whose creator Isaac Bratzel later founded AvatarOS. They're making synthetic versions of the most ordinary sales pitch on social media: the happy customer showing you the thing she supposedly just found. That is where the trust problem lives.
Startups can see the money. Business Insider reported in March 2025 that AvatarOS raised a $7 million seed round led by M13, with participation from Andreessen Horowitz Games Fund and others, to build digital avatars for social media, games and apps. The company was founded by Bratzel, who helped create Lil Miquela while at Brud. Meanwhile, Synthesia, the London AI avatar company used for corporate video, raised $200 million at a $4 billion valuation in January 2026, according to The Wall Street Journal and The Guardian. Synthesia is not an influencer shop in the narrow sense, but its growth tells you why investors like synthetic faces: businesses want cheaper, repeatable video without booking people, studios or travel.
Frankly, the brand argument is easy to understand. A synthetic influencer doesn't miss a shoot, ask for more money, get tired, post something embarrassing at 2 a.m. or object to a script. It can be translated, resized, edited and tested like any other campaign asset. Clarissa Mansbridge, who creates AI influencer images through Mia Metaverse, told The Guardian that brands don't want to pay $20,000 to $70,000 for a traditional photoshoot. You can see why a budget holder listens.
You should also see why disclosure cannot be optional.
The disclosure problem is the product problem
The current rules are not built cleanly for this. In the UK, the Advertising Standards Authority told The Guardian there is nothing in its rules that explicitly bans brands from posting AI-generated promotional content without an AI label, though the ad still must not mislead consumers. In the EU, the AI Act will start applying transparency obligations for AI-generated and manipulated content in August 2026, including machine-readable marking for synthetic outputs. That is a stronger framework, but it still has to survive the messy reality of Instagram, TikTok edits, reposts, captions, affiliate pages and content stripped of metadata.
The United States is no cleaner. The FTC's 2023 Endorsement Guides require clear disclosure of paid endorsements and material connections. They also warn against deceptive endorsements. But a fully synthetic persona creates a different problem from a human influencer forgetting to write #ad. The issue is not only whether money changed hands. It is whether the endorser exists at all.
This is why vague labels won't do much. A tiny "AI-assisted" caption under a video that looks like a real bride crying over a wedding app does not give the viewer the central fact. The plain disclosure should be just as direct as the deception is subtle: this person is AI-generated, this is a paid promotion, and no real customer gave this testimonial. If that kills the ad, the ad was living on confusion.
There is a broader cost, too. Which? told The Guardian that its recent deepfake investigation found 70% of people could not correctly identify all the real and fake videos they were shown. That is not a consumer failing. It is the predictable result of tools designed to make fake faces look normal. You cannot put the burden entirely on viewers and then act surprised when they believe what the platform serves them.
Brands using these tools now are making a bet. They are betting the cost savings arrive before the enforcement does, and that consumers will forgive the trick once the practice is common enough. Some will. Many won't. The first major brand caught selling through an undisclosed synthetic customer will not be remembered for efficiency. It will be remembered for making people feel foolish for trusting what they saw.
Also read: Kevin Warsh just killed the Fed's safety blanket and now everyone has to price their own risk and Defense tech startups have pulled in $14.6 billion this year and Silicon Valley is not turning back and John Ternus inherits an Apple where design answers to operations and plans to change that
