Algorithms profit from fake celebrity photos more than real ones

In the months between Taylor Swift and Travis Kelce becoming public and their first officially photographed appearance together, the internet filled with something that looked like news. AI-generated images of the two of them at the Super Bowl, in the stands, holding hands, kissing, clean and plausible and moving through Twitter and TikTok with the velocity of genuine documentation, though they were fabricated and they worked.

The prevailing explanation is intuitive. Fans and gossip consumers want content, there is no content yet, so they accept the synthetic. But this inverts what actually happened — there was no gap that needed filling, no demand that preceded the supply.

The images created the appetite they pretended to satisfy. Content farms with 500,000 followers posting AI images with captions like "JUST IN: Travis spotted at Chiefs practice" saw engagement spike into six figures per post.

Twitter's algorithm didn't distinguish between real and fabricated photographs. It rewarded velocity and volume. A legitimate paparazzi image might generate 50,000 engagements over a week. An AI image generated for 40 seconds of computational cost could match that in two hours. For accounts run as pure engagement operations, the math was immediate.

The platforms knew this was happening and have the infrastructure to detect AI images, or at least to label them. They chose friction-free amplification instead. Their revenue depends on engagement velocity, not accuracy. A user scrolling through a timeline full of authentic celebrity photographs might stop after five posts, but a timeline thick with speculation, fabrication. The constant possibility of "new" content keeps them scrolling. This wasn't about human appetite overriding skepticism. It was about economic incentive structures designed to make falsification more profitable than verification.