Who Actually Wrote That Book?
The AI-book panic isn’t about quality. It’s about who gets to be seen.
There’s a stat rattling around right now that’s easy to skim past: the rise of large language models roughly tripled the number of new book releases between 2022 and 2025, and more than half of everything published in 2025 now contains AI-generated text somewhere in it. Average quality, measured by how much people actually use what they buy, went down over that same period. Researchers Imke Reimers and Joel Waldfogel tracked this in a recent NBER working paper, and the causal story they tell is narrower than “AI makes books worse.” It’s this: the books that mix in AI content score lower on average, and since those books now make up the majority of new releases, they drag the average down. Strip out the AI-containing books, and the human-only average barely moved.
That’s a strange, ambiguous fact to build a moral panic on. So naturally, we built one anyway — and the shape the panic has taken tells you more about labor than about literature.
Here’s the sentence doing the most work in this whole debate, in one form or another: prompting isn’t real labor, so it shouldn’t be rewarded like real labor.










































