AI Books Flood Amazon and Slash Human Authors' Revenue
A study of over 14,000 Amazon e-books shows that a massive influx of AI-generated titles is diluting the self-publishing market, driving down revenues even for human authors who avoid AI.
Researchers analyzed 14,419 randomly selected self-published Amazon e-books released between January 2023 and March 2026. Using daily sales data from a publisher's dataset of 500,000 titles—covering 95 percent of daily e-book sales—they classified texts using the Pangram v3.3 detector, though Pangram 4 has since been released. The detector has a 0.04 percent false-positive rate. Books were grouped into three bands: no AI, light AI (up to 25 percent), and substantial AI (over 25 percent).
Books with substantial AI represent 20 percent of the catalog but generate only 12.1 percent of sales and 11.3 percent of revenue. Meanwhile, books with no AI make up 62.9 percent of the catalog and bring in 72.5 percent of revenue. However, massive volume is diluting the market. Between the first quarter of 2023 and the first quarter of 2026, the catalog grew 38.3-fold, while quarterly revenue only grew 8.9-fold. Consequently, revenue per book fell in seven of eight genres for titles with no detected AI. The only exception was the Fantasy, Supernatural, and Horror genre, where human-authored revenue rose 35 percent.
AI-generated content is also capturing top sales spots. The share of new Top 25 entries with substantial AI rose from near zero to 31 percent, while the sales share of human-written books fell from nearly 100 percent to about 60 percent by the second quarter of 2026. This output is concentrated among a few creators. Out of 385 author identities, 287 boosted their output after their first AI book. One pseudonym generated 1.7 million dollars across eight titles, and the highest-grossing AI book earned 643,000 dollars on 80,431 copies. This mirrors reports of romance writer Coral Hart, who published over 200 titles under 21 pen names in a year.
Using the Allen Institute for AI's infini-gram tool and the Google Books index, researchers found that the top 50 AI-heavy books had a 45 percent overlap with rare expressions from existing books, compared to 37.7 percent for human books and 19.1 percent for award-winning literature. This overlap increased by 7.6 percentage points for every tenfold increase in revenue. These findings provide empirical evidence of market dilution that could influence copyright cases like Kadrey v. Meta, where Judge Vince Chhabria warned about the impact of AI-generated competition.
This is our own summary of reporting by The Decoder
