AI slop floods uploads, but audiences stall fast

A new analysis tied to a creative-economy book suggests AI-generated content is soaring on some platforms yet barely reaches viewers—hinting at a “slop ceiling” driven more by discovery systems than audience tastes.

For a growing slice of the internet. the experience has started to feel like a landfill problem: content piles up faster than anyone can genuinely find what they want. And early signals suggest the public isn’t simply refusing AI work—it’s hitting a limit on how much of the low-cost. disposable stuff gets through.

A new analysis of the creative economy, published in the online repository arXiv and part of a new book, points to that stark mismatch. In some platforms, AI-generated material may now account for 44% of uploads—yet it draws only 1% to 3% of streams.

The authors call it a “slop ceiling,” and they trace the pattern to more than just taste. Peter Woodbridge and John O’Hare. co-authors of Dream Machine. argue that audiences appear to have a limit on how much AI-generated work they’re willing to consume. But they are not fully certain why that ceiling exists.

Woodbridge, one of the researchers behind the analysis, leans toward a platform-level problem rather than a straightforward consumer-level rejection. The challenge, he says, isn’t that people won’t take in creative work—it’s discovery.

“You can make all the good things you like, but the hard thing in music, or any creative medium, is always discovery,” Woodbridge says.

The low share of streams for AI content, he argues, doesn’t automatically mean the content is bad. It may indicate that recommendation systems and fan communities—places where human endorsement helps decide what rises—are still outcompeting AI uploads, at least for now.

That dynamic is especially visible in areas where volume is exploding. Woodbridge points to mobile games, where he says 186,000 titles have been released in the last six months alone. Many of those games, he says, are made by humans using AI tools or through AI-heavy workflows.

“Is it getting watched and seen? I don’t know,” Woodbridge says. “It raises the volume, but does it raise the quality?”

If the bottleneck is discovery, then one solution becomes less glamorous than the AI tools themselves: better curation. Woodbridge points to “the human curation. it’s the curation layer.” He argues that the internet has always relied on filters—from Google search to influencers. journalists. and fan communities.

But as AI search rises and the web becomes “a lot of noise and not a lot of signal,” Woodbridge suggests trusted communities may matter more, not less.

The relationship between label and reaction adds another complication. Woodbridge says research often finds that people struggle to distinguish between human-made and AI-made work. But “as soon as they’re told it’s AI-made. that really changes things.” That means the story can’t be reduced to a simple culture-war split between human and machine.

And with AI already used by hundreds of millions of people. the next phase of the creative economy is unlikely to be an either/or showdown. Woodbridge describes it as more likely to be “human plus AI.” Still. he worries that the “human plus AI” future could create a quieter problem: a “hidden and unauthorized” layer of AI use. where creators rely on the tools for an edge but feel discouraged from admitting it.

AI slop generative AI creative economy arXiv Dream Machine uploads streams mobile games curation recommendation systems