AI on the bookshelf?

Jane Friedman receives article submissions almost daily that include text generated by artificial intelligence. Friedman, a former executive at Writer’s Digest, runs a website where she reports on the publishing industry. She can usually spot AI in minutes, though at least one article made it onto her website before she recognized it included portions of AI-written text.

Like Friedman, book publishing houses have also been inundated with AI writing, and the industry is scrambling to respond. “There is a high level of concern and I think maybe anxiety among editors,” Friedman said. “They’re not sure how to detect it.”

AI-authored books have already flooded the self-publishing space, which by design has a low entry barrier. But at least one major publishing house has released and later recalled a book due to suspected AI use. It’s getting harder to tell the difference between computer-generated and human writing, forcing publishing houses and readers to demand transparency from authors.

Very few authors admit to using AI, making it difficult to estimate how many AI-written books are on the market. Recent publishing statistics might provide some perspective. More than 4 million new books entered the market in 2025, marking a 32% increase from the previous year. An influx of AI-authored books likely drives some of that growth. The AI-detection platform Originality.ai flagged a whopping 63% of paperback religious books on Amazon. In an August study, researchers from Stony Brook University and Columbia University used detection software to analyze over 14,000 self-published genre-fiction books available on Amazon. According to the paper, more than 25% of the books contained “substantial AI text.”

AI-authored books have not yet made it into bookstores en masse. But after publishing Mia Ballard’s Shy Girl in the U.K. last fall, publisher Hachette pulled the book in March ahead of its U.S. release due to suspected AI use. Debut author Jerry Falade recently lost a major book deal for his manuscript “Call Me, I’ll Hide the Body,” reportedly because of accusations of AI use.

As the technology improves, readers are finding it increasingly difficult to spot AI-generated text. In a study that Cambridge University Press published in August, researchers tasked more than 900 participants with reading a mix of human and AI-written excerpts. “We found that, overall, participants were significantly worse than chance at correctly differentiating between stories written by a human author and stories generated by ChatGPT,” the researchers wrote.

While AI-generated text currently has some telltale features, like frequent use of em dashes and clunky metaphors, it may not have such clear giveaways in the near future. “I don’t know that [AI writing is] going to leave such a fingerprint,” Jane Friedman said. “It’s going to be really hard to tell, if you can at all, in a matter of like a few years to one year.”

A survey that BookInsights.ai released in October polled roughly 5,000 readers. It found that about 22% didn’t mind authors using AI at any stage of writing, though 85% of all respondents said authors should disclose how they used AI.

But getting authors to admit to their AI use is easier said than done. According to Authors Guild CEO Mary Rasenberger, publishers already include clauses in book contracts requiring authors to submit original work. “Most publishers interpret that to mean that the author wrote the book and they did not use AI to help them generate it,” Rasenberger said. Now, publishers have to make their AI policies abundantly clear.

In October, the Authors Guild issued guidelines encouraging publishers to provide clear expectations about AI use and let authors know when they plan to use AI detection software.

Some publishers have already spelled out AI expectations. Bloomsbury, an international publishing house, issued guidelines last December about appropriate AI use. The policy clarifies that authors may use AI to research, brainstorm, and edit, but they cannot submit any amount of AI-generated text for publication.

Publishers have more than just their reputation on the line. If it becomes apparent that a book was primarily written with the help of AI, the U.S. Copyright Office could theoretically withhold copyright protection. “So publishers are very justified in saying they don’t want any AI-generated text,” Rasenberger said.

But as the publishing industry tries to rein in AI use, some writers worry they could face false accusations of using AI. Rasenberger said that accusations about AI use mostly come through social media. Anyone can run a book through detection software Pangram and lobby AI accusations online, leading to what Rasenberger calls “a culture of fear.”

But those fears may be somewhat overblown. Jane Friedman says publishers can theoretically reject manuscript submissions if they suspect AI use, but it would take a lot more than suspicion for a publisher to cancel a rubber-stamped book contract. “[Publishers] do ultimately have to point to quality concerns, rather than it just feels like it’s AI-generated,” Friedman said.

Cracking down on AI-written books may actually help human authors stand out amid a swelling sea of slop. While AI books probably haven’t climbed onto bestseller lists, they make it harder for self-published authors to make a name for themselves. In a TikTok video, self-published author Marcia Ercolin explained how it felt to come across an AI-generated book while scrolling through Kindle Unlimited. “It really just puts into perspective how big of a mountain us indie authors have to climb just to get our work out,” she said. “Every time we get halfway up, the mountain just doubles in size with how much is being let out.”

In the near future, Rasenberger hopes that labeling AI and human-authored work will become an industry standard. One organization already partners with publishers to add “organic literature” labels to books written by human authors. Books by People, a U.K.-based group, partners with publishers to add a “Books by People” label on book covers, certifying that the books were written by humans and not AI. Similarly, the Authors Guild has developed a “Human Authored Certification” to verify the author didn’t use AI-generated text.

Rasenberger believes people won’t knowingly reach for a book that includes computer-written text. “AI books just make readers more likely to distrust what they’re getting and what they’re buying,” she said. “I think the market will decrease for [AI-written books]. … But that won’t happen until we have accurate labeling.”

Andrew Steiner, an assistant professor of creative writing at George Fox University, disagrees. Steiner argues that consumers have become accustomed to consuming media tailored to their individual preferences.

“The publishing industry could do the same thing where algorithmically it’s figured out what you are interested in and it will just produce written material that aligns with those interests,” he said. “I could imagine that being incredibly seductive.”

Trope-heavy books like romantasy novels have already gained a massive following. If Steiner is correct, using AI to further pinpoint what readers want could be a natural next step.

Still, Steiner says he wouldn’t read titles that he knew included AI-generated text. “It matters to me that there’s a person on the other side of the book,” he said. Humans reflect the Creator through storytelling.

“I think we should be suspicious of a tool that wants to take that task away from us and do it on our behalf,” Steiner said.