Selling Books Grew 19x. The Money Grew 9x. Somebody Finally Measured What AI Does to a Knowledge Market.

Four researchers did the thing this industry has been arguing about for three years without evidence. They counted.

The team ran full-text AI detection across 14,419 self-published genre-fiction books sold on Amazon from 2023 to 2026, matched to daily sales records through June 2026. None of those books disclosed whether or not they contained AI-produced content.

Then comes the number that should stop a course creator mid-scroll.

Over the period studied, the number of books with observed sales in a quarter grew 19.2-fold, while quarterly revenue grew only 8.9-fold. The market added selling books faster than it added revenue, and revenue per selling book fell across most genres.

More product. Same wallet. Everyone gets a thinner slice.

The slop was not commercially irrelevant

The comforting story about AI content has always been that buyers can smell it and will ignore it. The paper was built specifically to test that assumption, and it didn't survive.

Books with substantial detected AI text — more than 25% — make up a large share of the catalog and a smaller share of sales. Even so, they reach commercial scale, winning a growing share of sales over time and taking more of the scarce top-rank positions once held by books with no detected AI text.

The researchers' own framing, via the MIT Initiative on the Digital Economy, is that generative AI "reshaped the book market without competing on quality at all".

That is the part that translates directly to courses. Nobody needs to make a better course than yours. They need to make four hundred adequate ones and let the algorithm sort it out.

Where the damage concentrated

This is the detail course sellers should read twice.

Books with no AI text lost the most ground in genres with high AI diffusion, and most of all where Kindle Unlimited availability was high.

Kindle Unlimited is an all-you-can-eat subscription with a shared payout pool. The closest analogue in this business is the all-access course subscription — the bundled libraries, the unlimited-seat memberships, the marketplace pools where a fixed amount of money gets divided by however much content shows up.

Flood a pool like that and the math is not subtle.

There's a second finding with teeth. Among top-selling books, those with substantial AI text drew on more distinctive language from existing books than books with no AI text did — and for those titles, overlap rose with revenue, a pattern the researchers did not detect for books with no AI text.

Translation: the best-performing synthetic products were the ones that most closely echoed what already worked.

"Likely to get squeezed"

Coauthor Paramveer Dhillon, an associate professor of information at the University of Michigan, described the mechanism to Fortune as a crowding-out effect driven by competition, where human authors "are likely to get squeezed, and their revenues are reduced."

Coauthor Tuhin Chakrabarty expects the market to split rather than collapse — "There will be some literature which won't be encroached by AI, and maybe that will just become more popular," he told Fortune, alongside a far more formulaic and profitable AI tier. "So I guess human authors will find which clan they want to belong to."

Swap "authors" for "creators" and that sentence is a 2027 positioning memo.

The paper is also a fight

Read it with the context attached. It is a working paper, not peer-reviewed, and OpenAI has moved to strike it and related expert testimony in the MDL copyright litigation against the company, pointing to a disclosure on Chakrabarty's CV of a $100,000 "gift" from Susman Godfrey — the plaintiffs' firm that paid the Stony Brook professor to research how AI-generated books dilute the market for human authors.

And there's a real counterargument on the numbers. a16z's Moses Sternstein noted that by the authors' own calculation, self-published revenue still rose nearly 9x in three years, and asked whether a flood that grows the market faster than it otherwise would have grown is necessarily bad for the humans in it.

Both things can be true. The pie got much bigger. The per-product slice got smaller.

There is no detector running across Kajabi, Teachable, Thinkific or Udemy that anyone has published. There is no disclosure standard. The metric worth watching isn't how many courses exist in your category — it's revenue per selling course, and almost nobody is measuring it.