AI-generated books are changing the economics of self-publishing on Amazon. A study of 14,419 randomly selected self-published e-books found that the pressure on human authors may come less from the quality of AI titles and more from the sheer number of them entering the market.
What the study measured
The researchers looked at self-published e-books released between January 2023 and March 2026. For each book, they used daily sales figures from an internal dataset maintained by one of the five major US publishers.
According to the researchers, that dataset tracks about 500,000 Amazon titles and covers roughly 95 percent of all e-books sold daily on the platform. That made it possible to compare catalog growth, sales, and revenue across a large group of books.
The study also differed from earlier work by analyzing full texts rather than short book previews. The researchers used the Pangram v3.3 detector, whose developers report a false-positive rate of 0.04 percent. Pangram 4 has since been released.
Books were grouped into three categories:
- No detected AI text.
- Light AI content, meaning up to 25 percent of the text was flagged.
- Substantial AI content, meaning over 25 percent of the text was flagged.
The catalog is growing faster than revenue
Books with substantial AI content made up 20 percent of the catalog studied. But they produced only 12.1 percent of sales and 11.3 percent of revenue.
Books with no detected AI text accounted for 62.9 percent of the catalog and generated 72.5 percent of revenue. On the surface, that suggests AI-generated books still underperform compared with books where no AI text was found.
The deeper finding is more important. Between Q1 2023 and Q1 2026, the cumulative catalog grew 38.3x. The number of titles selling per quarter grew 19.2x. Quarterly revenue grew only 8.9x.
That means many more books are competing for a revenue pool that is expanding much more slowly. The researchers describe this as "dilution," while stressing that the comparisons are observational and associational rather than experimental proof of causation.
Human-written books are feeling the squeeze
The revenue decline was not limited to AI-heavy titles. In six of eight genres, revenue per book dropped when comparing titles released in 2023 and 2025 over the same post-release window.
For books with no detected AI text, revenue fell in seven of eight genres. That matters because it weakens the idea that averages are falling only because low-selling AI titles are flooding the catalog.
The exception was Fantasy/Supernatural/Horror. In that category, AI text arrived latest and gained the least traction. Revenue per book for titles with no detected AI text rose 35 percent, which the researchers say argues against a broad market trend as the sole explanation.
The study also found stronger patterns in genres with high Kindle Unlimited availability, where readers draw from a shared subscription pool. In those genres, the revenue-share lead of books with no detected AI text was 8.4 percentage points smaller than in genres with low Kindle Unlimited availability. The researchers attribute that gap to genre-specific traits and do not claim a causal link to Kindle Unlimited itself.
AI titles are reaching the top ranks
AI-generated books are not only expanding at the bottom of the catalog. The share of new Top 25 entries with substantial AI content rose from near zero to 31 percent over the study period.
The top ranks also became less stable. The share of books with no detected AI text that stayed in the Top 25 from one quarter to the next dropped as low as about 28 percent at one point before settling around 62 percent by the end of the study.
Output was concentrated among prolific producers. Of 385 author identities that published more titles with substantial AI content after their first AI book, 287 increased monthly output afterward.
The highest-grossing pseudonym earned $1.7 million in gross revenue before platform fees across eight titles. The single highest-grossing book with substantial AI content brought in $643,000 on 80,431 copies sold.
The source also points to a New York Times report on "Coral Hart," who reportedly published over 200 romance titles under 21 pen names in a single year and sold about 50,000 copies. The broader pattern is a market where high-volume publishing can become a major advantage.
Why copyright questions are getting sharper
The study also examined language overlap between successful AI books and existing works. Researchers used the Allen Institute for AI's infini-gram tool with the Google Books index.
They looked for rare expressions that appear in five or fewer Google Books volumes and are absent from a 4.7-trillion-token web snapshot. Among the 50 highest-grossing titles with substantial AI content, those expressions covered 45 percent of the text.
By comparison, the top 50 books with no detected AI text had 37.7 percent coverage, while award-winning or award-nominated fiction had 19.1 percent. Within AI books, overlap rose 7.6 percentage points for every tenfold increase in revenue. No such correlation appeared for books with no detected AI text.
The method does not identify the source of individual passages or prove that a specific book was copied. It measures aggregate language overlap.
The findings connect directly to copyright lawsuits against AI companies. In Kadrey v. Meta, Judge Vince Chhabria ruled in Meta's favor in June 2025 but warned about using copyrighted books to build a product that could generate billions in revenue while producing a potentially endless flood of competing works.
The plaintiffs had not presented empirical evidence of market dilution. This study provides evidence that books with no detected AI text are earning less as AI titles enter the market in large numbers.
There is also a visibility problem for readers. Authors must disclose AI involvement when publishing through Kindle Direct Publishing, but Amazon does not pass that information along to customers. Amazon has mainly responded to AI title problems by capping publications at three per day.