A 22-year-old Indian medical student created an AI-generated conservative influencer called Emily Hart, grew her to 10,000 followers and millions of views in a month, and monetised her through merchandise and an explicit content platform , raising uncomfortable questions about AI literacy, platform responsibility, and the economics of synthetic identity.
The story, reported by WIRED on April 21, has the texture of a hustle story but the implications of a policy problem. Sam, the pseudonymous aspiring orthopaedic surgeon, was not building a business. He was doing what any broke student does: looking for the easiest way to make extra money. He tried YouTube Shorts and selling study materials. Neither worked. Then he asked Google's Gemini for advice. Gemini's response, documented in a transcript Sam provided WIRED, told him not to create a generic attractive woman because the competition was too intense. It recommended the MAGA conservative niche, describing the conservative audience, especially older men in the U.S., as having higher disposable income and being more loyal. Sam followed the advice. He built Emily Hart, a fictional registered nurse resembling Jennifer Lawrence, posting images of her ice fishing, drinking Coors Light, and handling firearms. Within a month she had 10,000 followers and Reels hitting 3 to 5 million views. He was spending 30 to 50 minutes a day. The income flooded in.
The most structurally significant part of the story is not what Sam did. It is what Gemini told him to do. A mainstream commercial AI, built by Google and deployed to millions of users, was asked for a side hustle and recommended targeting a specific political audience based on their psychological loyalty and disposable income profile. That output was not a jailbreak. It was the default response to a direct question. Google's spokesperson said Gemini was asked how to reach an audience with specific political beliefs and responded accordingly. That is a technically accurate but insufficient answer. A tool that recommends building synthetic identities to exploit political psychology is doing something that advertising platforms have been sued and fined for. The AI layer adds one degree of separation and, apparently, significantly reduces moral friction.
For entrepreneurs building AI products, the lesson is that what your model will say is as much a reputational risk as what your platform will host. Gemini gave Sam a roadmap. The liability sat with Sam. That distribution of responsibility will not survive the first serious regulatory case in this space.
Lesson two: platform verification is broken and getting worse
Emily Hart operated across multiple platforms simultaneously, reached millions of views, and generated subscription revenue on Fanvue before being taken down. Fanvue explicitly permits AI-generated content, which is a policy choice. The other platforms involved did not explicitly permit it, but did not catch it either. Sam tried to build a left-wing equivalent and could not get traction: he told WIRED that Democrats know it's AI slop. That asymmetry is telling. It is not that synthetic content is harder to make for one political group. It is that the detection gap is not uniform across audiences. Platforms cannot rely on community detection to identify synthetic accounts when the audience most targeted by that content is least likely to report it.
The EU AI Act's synthetic media disclosure requirements, and the Digital Services Act's provisions on fake accounts, directly apply to exactly this scenario. But enforcement presupposes detection. What Sam built , GPT Image 2-quality faces, AI-generated video, targeted political content , would have been sophisticated disinformation infrastructure two years ago. In April 2026 it is a side hustle a medical student runs for 30 minutes a day.
Lesson three: the economics of synthetic identity are now accessible
Sam estimates he made thousands of dollars monthly at near-zero marginal cost. The capital requirement was Grok for image generation, Gemini for strategy, Fanvue for monetisation, and 30 minutes of daily curation. No human model. No photography. No production budget. The content scaling from that input is not a warning sign about one student. It is a proof of concept that hundreds of thousands of people around the world will read and attempt to replicate. The first-mover advantage Sam had was the idea. That advantage no longer exists.
For the AI industry, the Emily Hart story lands in the same week as confirmed GPT Image 2 disinformation campaigns, the OpenAI astroturfing allegations, and Sam Altman's new five-principle framework pledging to resist misuse. The gap between published principles and deployable misuse infrastructure is now measured in minutes and dollars. The structural question , who is responsible when an AI recommends building a synthetic identity and a user follows the advice , does not have a clean answer yet. It will need one soon.
Also read: The Musk v. Altman trial is the most consequential tech lawsuit in a generation • OpenAI's five new principles reframe its mission from AGI lab to AI infrastructure for humanity • DeepSeek V4 sends Zhipu and MiniMax shares down as China's AI price war deepens
