A new study shows that AI-generated books flooding Amazon's catalog are reducing sales and revenue for human authors across almost every genre, even as the total book market grows far more slowly than the number of titles.
The analysis found that books with substantial AI content (20% of the catalog) generate only 12.1% of sales, while the catalog expanded 38.3× between early 2023 and early 2026 but quarterly revenue grew just 8.9×.
Notably, revenue per book declined even for human-authored titles in seven of eight genres, indicating that market dilution, not just low-quality content at the bottom, is the problem—a finding that could strengthen copyright lawsuits against AI companies seeking evidence of market harm.
What happened
A study analyzing Amazon's full book catalog found that titles with substantial AI content (over 25% AI-detected text) make up 20 percent of books but generate only 12.1 percent of sales and 11.3 percent of revenue. Between Q1 2023 and Q1 2026, the catalog grew 38.3× while quarterly revenue grew only 8.9×. Revenue per book fell in seven of eight genres even for human-authored titles, and AI books now account for up to 31 percent of new Top 25 bestseller entries.
Why it matters
The influx of AI-generated books is diluting the market for all authors, not just displacing low-quality content. Human-written books that earned less in 2025 than in 2023 suggest systematic revenue loss as the total catalog balloons faster than reader spending grows. The findings provide empirical evidence for ongoing copyright lawsuits against AI companies—judges had previously noted the absence of concrete proof that AI-generated works harm the market for copyrighted human-authored works.
What to watch
A small number of prolific AI publishers are capturing outsized revenue; the highest-grossing pseudonym earned $1.7 million in gross revenue across eight titles, and one AI book brought in $643,000 on 80,431 copies sold. Top-selling AI titles draw 45 percent of their text from rare language patterns that appear in only a handful of published books, compared to 19.1 percent for award-winning fiction, suggesting potential unauthorized copying of copyrighted material.
Researchers at the Allen Institute for AI and collaborating institutions conducted the most comprehensive analysis to date of AI-generated books on Amazon, examining the full text of titles in the Kindle catalog rather than short previews alone. Using the Pangram v3.3 detector (with a reported false-positive rate of 0.04 percent), they classified each book into three categories: no detected AI text, light AI content (up to 25 percent), and substantial AI content (over 25 percent). The headline finding challenges the narrative that AI books remain a marginal phenomenon: books with substantial AI content comprise 20 percent of the studied catalog but generate only 12.1 percent of sales and 11.3 percent of revenue, compared to human-authored titles, which represent 62.9 percent of the catalog and generate 72.5 percent of revenue.
The deeper issue, however, is market dilution rather than simple quality stratification. Between Q1 2023 and Q1 2026, the cumulative catalog grew 38.3× while the number of titles selling per quarter grew 19.2×—yet quarterly revenue grew only 8.9×. This mismatch has compressed earnings across the board. In six of eight genres, revenue per book fell when comparing titles released in 2023 and 2025 over the same post-release period. Critically, when the researchers looked only at books with no detected AI text, revenue per book fell in seven of eight genres. The sole exception is Fantasy/Supernatural/Horror, where AI text arrived latest and gained the least traction; there, revenue per book for human-authored titles rose 35 percent. The researchers describe their finding as "dilution" but stress that their comparisons are observational and associational, not experimental proof of causation. They note that Kindle Unlimited availability correlates with a smaller revenue advantage for human-authored books, attributing the gap to genre-specific traits rather than drawing a direct causal link to the subscription service itself.
AI content is also breaking into the bestseller ranks at an accelerating pace. The share of new Top 25 entries with substantial AI content rose from near zero to 31 percent over the study period, and the churn rate among bestsellers increased: the share of books with no detected AI text that remained in the Top 25 from one quarter to the next dropped to about 28 percent at one point before settling around 62 percent by the study's end. Output is highly concentrated among a small number of prolific producers. Of 385 author identities (pseudonyms) that published more titles with substantial AI content after their first AI book, 287 increased their 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 study echoes reporting on "Coral Hart," a figure who reportedly published over 200 romance titles under 21 pen names in a single year and sold about 50,000 copies.
A striking finding concerns the language composition of top-selling AI books. Using the Allen Institute for AI's infini-gram tool and the Google Books index, researchers identified rare expressions—phrases appearing in five or fewer Google Books volumes and completely absent from a 4.7-trillion-token web snapshot—to measure overlap with published works. Among the 50 highest-grossing AI titles with substantial content, these rare expressions covered 45 percent of the text, compared to 37.7 percent for the top 50 human-authored titles and just 19.1 percent for award-winning or award-nominated fiction. Notably, within AI books, overlap rose 7.6 percentage points for every tenfold increase in revenue, a correlation entirely absent for human-authored titles. Researcher Chakrabarty explained to the Allen Institute that AI detectors return only a probabilistic estimate without identifying the source of flagged passages; however, when suspect text also contains rare expressions from a small number of published books and absent from the web, "one can say with some confidence that it was taken from books." These findings have direct bearing on copyright litigation. In Kadrey v. Meta, Judge Vince Chhabria ruled in Meta's favor in June 2025 but warned that it was hard to imagine using copyrighted books to build a multi-billion-dollar product while flooding the market with competing works would qualify as fair use—yet the plaintiffs had presented no empirical evidence of this market dilution. The current study provides exactly that evidence. Complicating enforcement is the fact that Amazon does not publicly disclose AI involvement to customers, despite requiring authors to disclose it during Kindle Direct Publishing; the platform has mainly responded to AI-generated books that hijack well-known author names by capping publications at three per day.
The study reveals a market dynamic that differs fundamentally from the common assumption that AI books are simply low-quality "slop" at the bottom of the sales ladder. While books with substantial AI content do underperform relative to their catalog presence (20% of titles but only 12.1% of sales), the real problem is systemic: the entire market is being diluted. The catalog expanded 38.3× between Q1 2023 and Q1 2026, but quarterly revenue grew only 8.9×—far too slowly to support the explosion of new titles. This mismatch has compressed revenue per book across almost all genres, affecting even human-authored works. The exception—Fantasy/Supernatural/Horror, where AI text arrived latest and remains least prevalent—saw revenue per book for human titles rise 35 percent, suggesting that the genre-specific timing of AI adoption, not a broader market trend, explains the pattern. The researchers note that Kindle Unlimited genres (where readers draw from a shared subscription pool) show a smaller revenue advantage for human-authored books, pointing to platform mechanics as a secondary factor.
The concentration of revenue among a handful of prolific AI publishers underscores the market's structural shift. Of 385 author identities that published more AI books after their first, 287 increased monthly output. The highest-earning pseudonym earned $1.7 million gross across eight titles—evidence that AI-generated books are not marginal to the market but actively capturing significant reader attention and spending. Top-selling AI books also show a striking pattern: they overlap heavily with rare language from existing published works. Among the 50 highest-grossing AI titles, rare expressions (phrases appearing in five or fewer Google Books volumes and absent from a 4.7-trillion-token web snapshot) covered 45 percent of text, compared to only 19.1 percent for award-winning fiction. Within AI books, this overlap increased 7.6 percentage points for every tenfold increase in revenue—a correlation absent for human-authored titles. While the study does not trace the origin of individual passages, the researchers describe this pattern as circumstantial evidence that high-grossing AI books may be drawing heavily from copyrighted sources, a finding with direct implications for ongoing copyright litigation.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
The AI news that matters, in one minute each morning.
Sign up free