Amazon study finds AI ebooks are crowding out no-AI books

Books with substantial AI-written text have spread quickly on Amazon and appear to be taking a larger share of the market from titles with no detected AI text, according to a new study of 14,419 self-published genre-fiction ebooks sold between 2023 and 2026. The researchers, writing on arXiv, matched the books to daily sales records through June 2026 and classified them by the share of text detected as AI. They defined “substantial AI text” as more than 25% of a book’s full text, “light AI text” as anything above zero and up to 25%, and “no AI text” as no detected AI at all. The books in the sample did not disclose whether they contained AI-produced content. By their measure, books with substantial AI text accounted for 20% of all books in the sample, but only 12.1% of sales and 11.3% of revenue. Books with no AI text made up 62.9% of the catalogue, yet captured 71.7% of sales and 72.5% of revenue. Titles with light AI text accounted for the remaining 17.1% of books and 16.2% of both sales and revenue. The imbalance was starker at the top of the market. Books with no AI text held 63% of launch-window positions and 73% of the best-selling top 5%. Books with substantial AI text, by contrast, fell from 20% of all books to 10% of the top 5%. Over the study period, the number of books with observed sales in a quarter rose 19.2-fold, while quarterly revenue increased 8.9-fold and quarterly sales 7.3-fold. The paper says the market added selling books faster than it added revenue, so revenue per selling book fell across most genres. The share of observed sales held by books with substantial AI text rose from near zero in early 2023 to roughly 20% by the second quarter of 2026. Their share of constructed Top-25 rank slots also grew, while the retention of books with no AI text in the Top-25 fell to a trough near 28% before recovering unevenly to about 62%. The paper says books with substantial AI text spread across every genre it tracked, and that books with no AI text lost the most ground in genres with high AI diffusion and in areas where Kindle Unlimited availability was high. It also says that, among top-selling books, those with substantial AI text draw on more distinctive language from existing books than books with no AI text, and that overlap rises with revenue in that group. The authors argue that the pattern is consistent with market dilution, where low-cost generative AI allows hundreds of producers to draft and list books faster than sales, revenue and ranking slots can absorb them. They say that many AI books reach readers without disclosure and that short previews can make it hard to tell whether a title is AI-generated. The study frames those findings as relevant to the market-effect question in copyright law. The U.S. Copyright Office’s Part 3 report on generative AI training, published in pre-publication form, examined whether using copyrighted works to develop AI systems requires consent or compensation, and noted the issue sits amid dozens of U.S. lawsuits. Fairly Trained chief executive Ed Newton-Rex wrote on X that the research leaves AI companies’ argument that there is no financial harm “well and truly dead.”