In March, the book publisher Hachette yanked the horror title “Shy Girl” from its scheduled U.S. release after hefty allegations that the novel’s author Mia Ballard used AI to write the book.

While Ballard denied using AI for the novel, Hachette pulled the book from its website, Amazon, and U.K. shelves, but the incident became a moment of reckoning for the book industry: What should booksellers do about AI-generated books, particularly as they proliferate across marketplaces?

Today, the urgency to answer that question is growing. Recent research shows the magnitude of AI-generated books in online marketplaces presents a problem for human authors, and has the potential to change the landscape of bookselling more broadly.

An Originality.AI report published last month found that of more than 2,000 books on religion published on Amazon over the last 12 months, 1,272 titles, or 63%, were flagged as being likely written by large language models. A Stony Brook University-led study found that of 14,419 self-published genre-fiction books sold on Amazon between 2023 and 2026 and run through AI-detection software Pangram, 2,880 (about 20%) had “substantial” AI text, meaning more than 25% of the book’s prose was flagged as AI-generated. Additionally, 2,168 had more than 50% of its prose AI-generated. Nearly 1,000 novels were more than 90% AI-generated.

This study, which has yet to be peer-reviewed, found that more AI books came at the expense of revenue. Over the period of three years, the number of books with observed sales in a quarter increased 19.2-fold, but quarterly revenue increased only 8.9-fold, indicating decreased revenue per book sold. The effect, according to the researchers, was that the sheer magnitude of books flooding the marketplace made it less likely that an Amazon customer would select a given book written by a human, leading to fewer sales of any given title, and less money in the pockets of human authors.

“The crowding-out effect generated by these books—most of which are low quality, but some of which are still good enough that they lead to higher ranks—due to this increased competition, they’re likely to get squeezed, and their revenues are reduced,” Paramveer Dhillon, an associate professor of information at the University of Michigan and the study’s coauthor, told Fortune.

Amazon requires self-published authors to disclose if they’ve used AI to generate texts, and limits the number of titles they can upload to the site to three per day.

“We are committed to a bookstore that readers love and publishers and authors trust, with multiple safeguards we continuously strengthen to protect their experience,” an Amazon spokesperson told Fortune in a statement. “Our content guidelines govern which books can be listed for sale, and we take proactive and reactive measures to prevent, detect, and remove content that violates those guidelines, whether AI-generated or not. We continually monitor our store and enhance these protections as technology and the publishing industry evolve.”

How human authors are impacted by AI flooding the book market

Tuhin Chakrabarty, an assistant professor of computer science at Stony Brook University and the study’s coauthor, said the “algorithmic bubble” of online marketplaces is in part to blame for humans being squeezed out of the market. More AI books on Amazon also increase the likelihood of customers buying and clicking on AI-generated titles, which means these titles are more likely to be prioritized on the website’s pages. As AI technology improves, so too, does the quality of the AI-generated books, which may also increase their appeal among readers.

“If there is an author who is not a very big name, very famous, and they have completely written a book with their you know sweat and tears, now that book might not even show in the related books listing because there is so many AI-generated books that are surfacing based on the keyword and the search terms associated with them,” Chakrabarty told Fortune.

Authors, for their part, have gone after AI companies over copyright concerns, suing the likes of OpenAI, Meta, Anthropic, and xAI over allegedly using their books to train AI models while removing copyright notices and other copyright-related information. In July, a federal judge approved a $1.5 billion copyright settlement, ordering Anthropic to pay thousands of authors after the AI company downloaded and stored millions of pirated books to train its chatbot Claude.

The changing landscape of bookselling

Some booksellers are still figuring out how to engage with AI-generated content. Barnes & Noble CEO James Daunt said earlier this year that he would not outright ban the sale of AI-generated books, provided that it is clear the books were created with the technology. But he noted that major publishers would be unlikely to release an AI-generated book and therefore more unlikely that the retailer would stock them.

“Our position is that we do not sell AI books, as far as we are aware; we take active measures to exclude all AI-generated books from our online catalogue and never knowingly order any for stocking in our stores; and we demand that publishers label any books that are AI-generated,” Daunt told Fortune in May. “This is a straightforward rejection of AI books.”

But the CEO added that engaging in a conversation around not selling AI-generated books could get hairy, as it would veer into conversations around outright banning titles. Would Barnes & Noble sell books that were 25% AI-generated? Would it refuse to sell books more than 50% AI-generated?

“This would take a days-long conference to explore, and suggests further to us that our position is one of common sense,” Daunt said.

Dhillon and Chakrabarty believe the increase of AI-generated content will change how authors, particularly those up-and-coming, approach their work. They imagine a bifurcated marketplace that will both cater to an audience of readers that have no qualms with reading AI-generated works, and simultaneously create higher demand for genres of literary fiction that reflect a quality of writing AI cannot yet achieve.

“There will be some literature which won’t be encroached by AI, and maybe that will just become more popular,” Chakrabarty said. “There could also be this kind of market which is very formulaic AI-generated works, which will be very profitable. So I guess human authors will find which clan they want to belong to.”