FULL STORY

Over Half of Amazon Self-Published Books Suspected AI-Generated

A new study quantifies the flood of AI-generated books on Amazon, revealing that over half of sampled self-published genre novels are suspected to be AI-generated, severely impacting human authors' market share.

2026-07-28 ~ 2026-07-31 · 2 episodes · 33 posts

Episode 1 · New Paper Quantifies AI Book Flood: Amazon Self-Publishing Revenue Distribution Rewritten (2026-07-28, 31 posts)

A new paper on AI and copyright advances the debate over 'AI-generated book flooding' into quantifiable market analysis. Researchers randomly sampled 14,419 Amazon self-published genre fiction titles, used Pangram for full-text AI detection, and matched sales data, finding about 20% of samples are heavy AI text. The conclusion: AI books are not more successful per title, but generate significant revenue through low cost, high frequency, and massive volume, substantially squeezing human authors' per-book income in most genres.

Confirmed

  • The study covers 2023 to March 2026. Researchers classified samples into heavy AI (>25%), light AI (<25%), and no AI (0%) based on full-book AI proportion. They emphasize this quantifies different degrees of AI text, not labeling all AI-assisted writing as 'slop'.
  • About 20% of samples contain heavy AI text; nearly one-third of new bestsellers show AI traces across romance, thriller, sci-fi, etc. AI books' total sales share has surged from near 0% in early 2023.
  • AI books typically sell fewer copies per title, but supply structure has changed due to low cost and rapid release. Over 3 years, catalog size grew 38x while quarterly revenue only grew 9x; in 7 of 8 genres, human authors' per-book income is lower than in 2023. Fantasy and horror have least AI penetration and still see human income above 2023.
  • Among authors who continued writing after adopting AI, 3/4 update faster than before, some producing dozens of books per year. One pseudonymous author reportedly earned $1.7 million from 8 books in 12 months.
  • Successful AI books reuse rare phrases from published works more often, and the correlation between commercial success and 'rare expression coverage' is stronger for AI books than non-AI books.

Unconfirmed

  • The paper links the 'rare phrase reuse' phenomenon to the ongoing Kadrey v. Meta lawsuit, suggesting it could support 'market dilution' harm claims; this is a legal inference by the paper, not a court-confirmed ruling.
  • Another report claims over 50% of sampled Amazon books in 2025 contain AI-generated 'slop' text, but the post lacks methodological details and may not be directly comparable to the paper's heavy/light AI classification.

Why it matters

  • This research moves the copyright debate from micro-level ('does this book look AI-written?') to macro market level: even if AI books are not individually competitive, their sheer volume can dilute catalog and discovery mechanisms, rewriting revenue distribution rules in publishing.
  • For copyright and training data discussions, it provides quantitative evidence on supply surge, income crowding out, and potential 'market dilution', not just individual plagiarism cases.

11 more related posts →

Episode 2 · AI-Generated Books Flood Amazon, Squeezing Human Authors (2026-07-30, 2 posts)

A recent study reveals that over 50% of new self-published books on Amazon are AI-generated, with some containing up to 90% AI text. This flood of automated content is generating significant revenue while severely squeezing the living space of human authors.