AI-generated books are flooding online marketplaces, and researchers say the result is plain: human authors are getting squeezed.
A recent study found that as more AI-written titles appear on Amazon, revenue per book is falling for writers publishing without automation. The research looked at 14,419 self-published genre-fiction books sold on Amazon between 2023 and 2026 and found 2,880, or about 20%, had 'substantial' AI text. It also found 2,168 titles with more than 50% AI-generated prose, and nearly 1,000 novels that were more than 90% AI-generated.
The study, which has not yet been peer-reviewed, said the volume effect matters as much as the technology itself. Over three years, the number of books with observed sales in a quarter rose 19.2-fold, while quarterly revenue rose only 8.9-fold. That gap points to lower revenue per book sold.
Paramveer Dhillon, an associate professor of information at the University of Michigan and a coauthor of the study, said the crowding-out effect is hitting human writers directly. 'They’re likely to get squeezed,' Dhillon told Fortune, adding that revenues are reduced because of the increased competition.
Tuhin Chakrabarty, an assistant professor of computer science at Stony Brook University and another coauthor, said the 'algorithmic bubble' of online marketplaces is helping push human authors out of view. More AI books on Amazon increase the chance that customers click and buy AI-generated titles, which can push those titles higher on the site and make it harder for smaller authors to surface in related-book listings.
Amazon says self-published authors must disclose if they use AI to generate text, and it limits uploads to three titles per day. An Amazon spokesperson said the company is 'committed to a bookstore that readers love and publishers and authors trust' and said it uses 'multiple safeguards' to protect the marketplace.
The pressure on authors is not limited to marketplace ranking. Writers have also sued OpenAI, Meta, Anthropic and xAI over allegations that their books were used to train AI models while copyright notices and other information were removed. In July, a federal judge approved a $1.5 billion copyright settlement involving Anthropic after the company downloaded and stored millions of pirated books to train its chatbot Claude.
For free enterprise, the lesson is not anti-technology. It is that markets need clear rules, honest disclosure and real enforcement when new tools start distorting incentives. If the digital shelves are crowded with low-cost machine output, the taxpayer is not the only one who loses; so do working authors whose paychecks depend on being found, clicked and sold.
