Back to feed
News Story
APriority78
THE DECODER
1 sources

AI-generated books flood Amazon, hurting human authors' sales

A new study finds that AI-generated books make up 20% of Amazon's self-published catalog but only 12% of sales, and revenue per book is dropping for human authors in seven of eight genres. This data could provide market-harm evidence for copyright lawsuits against AI companies.

SynthePulse Insight · AI deep readingMembers

AI Books Flood Amazon: The Truth Behind Declining Sales and Market Dilution

Version 1 · 1 source

A study of over 14,000 self-published e-books shows that AI-generated content is squeezing out human authors through quantity rather than quality, with even books without detected AI text seeing declining per-book revenue.

  • The study analyzed 14,419 self-published e-books and found that books with high AI content made up 20% of the catalog but contributed only 12.1% of sales and 11.3% of revenue.
  • From Q1 2023 to Q1 2026, the catalog grew 38.3-fold, but quarterly revenue only grew 8.9-fold, leading to a general decline in per-book revenue.
  • In seven of eight genres, per-book revenue for books without AI text declined; the only exception was Fantasy/Supernatural/Horror, where revenue rose 35%.
Open section navigationResearch Background and Methodology

Research Background and Methodology

Researchers analyzed 14,419 randomly selected self-published e-books from January 2023 to March 2026, and obtained an internal dataset from one of the top five U.S. publishers, tracking about 500,000 Amazon books and covering approximately 95% of daily e-book sales.

Unlike previous studies that only detected AI text in previews, this study used the Pangram v3.3 detector to classify full texts, with a reported false positive rate of 0.04%. Books were categorized into three tiers based on the percentage of AI-marked text: none, light (≤25%), and heavy (>25%).

Free for now

Read the full analysis

3 more sections of analysis, plus the full takeaway

Loading

Credibility boundary

This study relies on internal datasets and a detector, but the detector is only an estimate, and the study is observational, not causal. Some data comes from secondary reports, such as the 'Coral Hart' case.

Primary report

THE DECODER

Primary source