Books by Bots

Librarian Sondra Eklund spends her time stocking books for the public library system she works for in Virginia. One of her patrons recently asked the library to acquire a children’s book about pets other than cats or dogs, so she went looking.

When she came across a book titled Rabbits: Children’s Animal Fact Book from the publisher Bold Kids, it seemed promising. Eklund hadn’t heard of Bold Kids before, but it offers nearly 500 books on Goodreads and Amazon, and its paperbacks aren’t expensive. Though the catalog showed only the book’s cover—not its interior—she put in the order, thinking, “How bad could it be?”

But when the book arrived, Eklund learned the answer: “Unbelievably bad.”

Its pages contained strangely worded sentences, some of them including made-up facts about rabbits (such as the claim that they make their own clothes). Every page of text featured the same clip art of a bunny eating a carrot. Stock photos of rabbits littered the pages, their eyes and noses disappearing into the book’s bleed.

One page inexplicably appeared twice. Another told readers, “If you’ve ever had the pleasure of feeding a rabbit, you’ve probably wondered how they reproduce. The answer is simple: They live in the wild!”

Eklund now suspects the title and others from Bold Kids, which has no website or authors associated with its books, were created by artificial intelligence (AI). To add insult to injury, the book was print-on-demand, making it nonrefundable. “My cataloger was practically purple in the face,” she says.

The experience left Eklund much more cautious when acquiring books for the library. “We’re learning by experience what to look for,” she says. “We don’t want to put our library’s name on something that’s not good.”

At the same time, her quest for quality content has been made more difficult by the surge of AI-generated books on Hoopla, Ingram, OverDrive, and other vendor platforms. While some of those books are relatively simple to sniff out, others are easier to confuse for the real deal.

Worse, some AI-generated books may contain potentially dangerous misinformation. The New York Mycological Society, for example, has warned against purchasing AI-written guides to foraging wild mushrooms: “Please only buy books of known authors and foragers; it can literally mean life or death.”

“AI has gotten good enough for some things,” Eklund says. “But writing books? Our kids deserve better.”

An AI mystery

Eklund is far from the only librarian grappling with AI issues. Last fall, Robin Bradford, a collection development librarian at a public library in Washington, accidentally bought an AI-narrated audiobook on OverDrive. Only after a patron checked it out and complained about a file-corruption issue did Bradford realize the book’s narrator was listed as “Scarlett (synthesized voice).” Looking through her library’s collection, she found more than 100 audiobook titles with the same narrator, all of them thrillers from Lukeman Literary Management.

So the audiobook narrators weren’t human. Were the authors themselves? With monikers such as Blake Pierce, Kate Bold, Molly Black, and Mia Gold, the authors appeared to have many titles to their names, but little to no social-media presence and only bare-bones websites with no substantive author bios—just a list of books written.

In an email to American Libraries, Lukeman Literary Management President Noah Lukeman responds, “Digital narrations have, for quite a while, been ubiquitous and openly accepted and carried by all major libraries and e-tailers, and embraced by major publishers” and adds, “For decades, writers from J. K. Rowling to Stephen King on down have used pseudonyms.”

Bradford also remembers that when she looked up the books’ copyright records, she found that they contained mentions of “text generated by AI.” But she’s not sure exactly what that means. “Was it 10 percent of the book? Was it 70 percent? One hundred percent? I have no idea,” she says.

In the Copyright Public Records System of the US Copyright Office (USCO), a search for “Lukeman Literary Management Ltd.” yields more than 2,000 entries. Many of them contain the notation “Material Excluded: text generated by artificial intelligence.” At least one entry, for Fatal Choice (A Sydney Best Suspense Thriller— Book One), contains a lengthy statement that reads, in part, “Approximately 60% to 80% of the writing is original and human-written. For the portions that are AI-generated, nearly 100% of what was AI-generated has been either revised, rearranged, or rewritten by a human writer.”

In his email to American Libraries, Lukeman writes: “A small portion of our titles have used a hybrid human/AI approach, written by humans yet with some AI assistance in the process to achieve higher quality…. One should not conflate digital narrations or predominantly-human-with AI-assist novels with lower quality. User ratings and reviews ultimately speak loudly for themselves … and they should continue to be allowed to do so.”

For her part, Bradford doesn’t believe that the many patrons who enjoy these books are aware of AI’s involvement in their creation. “When you see a book and it’s written by an author, you automatically think that is a person,” she says. “I don’t think people are tuned into the idea that it’s not.”

She’s continuing to have trouble identifying many AI-generated titles by their vendor catalog entries alone. “I’m just going to look harder at the books I’m buying,” she says, “and making sure that I’m supporting authors who are truly authors.”

AI and copyright

Even when the nature of an AI-generated work is clear, from a librarian’s point of view, there are potential copyright issues to consider. Does a book that may have plagiarized from other books belong in a library?

Sarah Manning, a collection development librarian at Boise (Idaho) Public Library (BPL), pondered that question recently as she fielded a patron request for a book titled The Funny Bear’s Very Scary Journey. While a human author wrote the book, its cover notes that it features illustrations created with Midjourney, a popular AI tool that generates images based on human prompts. Midjourney trains its model on vast copyrighted data sets—and has faced lawsuits as a result.

We don’t want to put our library’s name on something that’s not good.—Sondra Eklund, children’s librarian at a Virginia public library

That didn’t sit right with Manning. “Did [Midjourney] rip off another artist?” she asks. “It’s a gray area.” In the end, she decided to add the book to the library’s collection, particularly since a patron had requested it.

Librarians concerned about AI-related copyright issues have a lot to keep up with. Generative-AI companies such as ChatGPT and Midjourney maintain that these technologies should be considered fair use, but many authors and voice actors, along with organizations like The New York Times, argue instead that using their work to train AI violates copyright.

In January, USCO stated in a press release announcing the release of part two of its Copyright and Artificial Intelligence report that the output of generative AI can be protected by copyright “where a human author has determined sufficient expressive elements.” Typing a prompt into ChatGPT doesn’t count. However, as the report itself states, if a human selects, arranges, or modifies AI-generated output in a “sufficiently creative way,” the result amounts to an original work. But how much human input is required? The USCO report states that “the courts will provide further guidance.”

In May, USCO released a pre-publication version of part three of the report, which stated that whether AI training can be considered fair use hinges on “what works were used, from what source, for what purpose, and with what controls on the outputs.” It also stated that “making commercial use of vast troves of copyrighted works to produce expressive content that competes with them in existing markets … goes beyond established fair use boundaries.” Subsequent to the report’s issuance, the White House dismissed USCO Register of Copyrights Shira Perlmutter, who has filed a lawsuit in response. In July, President Trump said copyright posed limitations to the development of AI.

While they wait for further developments, librarians like Manning continue to wonder whether AI-generated material belongs in a library at all. “I personally lean toward no,” she says. “It is actually coming from someone else’s writing, which is a form of plagiarism. I don’t know that everyone [who works] in libraries would feel the same way, but that’s how I see it.”

Pining for a policy

Some libraries have developed policies regarding staff and patron use of AI. These policies typically address the appropriate use of AI in the workplace and how to manage or restrict the data these tools collect.

But collection development policies regarding AI are harder to come by. During ALA’s 2025 Annual Conference and Exhibition in Philadelphia in June, a resolution to create an artificial intelligence cross-divisional working group passed. The working group will be charged with developing a unified, critically informed ALA position on AI and libraries for Council consideration.

In the meantime, patron requests for AI-created materials have pushed BPL’s collection development staff to vet purchase requests more thoroughly, particularly with independent or self-published titles. “We have had a lot of discussions about what to look out for on materials to see if we can try to determine if it’s AI. But I’m sure things have slipped through,” Manning says.

Nick Tanzi, library technology consultant and assistant director of South Huntington (N.Y.) Public Library, says that if an AI-generated book does slip into a library’s collection against policy, it’s because of a lack of disclosure from publishers and vendors. Without that disclosure, it’s hard to consistently enforce AI policies on collection development, he says.

“The worst policy is policy you can’t enforce, or that you enforce very unevenly,” he adds. “A lot of that comes down to the ability for an aggregator or distributor to say, ‘These are AI’ or not.”

While ebook catalogs like OverDrive let libraries choose the content they make available to patrons, that’s not the case with Hoopla. Instead, libraries typically opt into Hoopla’s entire catalog and pay afterward for what their patrons borrow, says Clayton Cheever, director of Morrill Memorial Library in Norwood, Massachusetts.

That means patrons sometimes check out AI-generated books in Hoopla without realizing it—a fact that frustrates him. “You look at the book, and you can tell it was made by some early-gen AI bot, because it can’t even spell monetization correctly,” he says.

Research and advocacy organization Library Futures is calling on Hoopla to filter out low-quality, AI-generated ebooks from its collections.

“Every single company that offers AI-generated material should have some sort of transparency standards put in place so that people know what they’re getting, just like [with] many other consumer products that have labeling on them,” says Jennie Rose Halperin, Library Futures’ executive director.

Hoopla relies on accurate metadata from its vendor partners to flag low-quality, AI-generated content. Trouble is, the industry standards necessary for this flagging don’t exist, says Jeff Jankowski, president of Hoopla Digital, who responded in writing to questions from American Libraries. “We are actively collaborating with our digital aggregation partners to ensure they provide metadata noting AI-generated ebooks, so we can offer transparency to our library customers,” he writes.

It would technically be possible for librarians to opt out of AI content from Hoopla’s collection by making lists of blocked books or by working with Hoopla itself. “You can talk to a rep and say, ‘Hey, we don’t want this,’” Cheever says. But, he adds, removing all possible AI material in the catalog would be incredibly time-consuming for librarians: “There is way too much content out there.”

How to Spot AI-Generated Books

Once an AI-generated book has made it to your library, it will likely give itself away with telltale signs such as jumbled, repetitive, or contradicting sentences; glaring grammatical errors or false statements; or digital art that looks too smooth around the corners.

Of course, if you can get a digital sneak-peek inside a book before ordering, all the better. But if not, how can you head off AI content so it never arrives on your desk? The following tips can help.

• Look into who the author is and how “real” they seem, says Robin Bradford, a collection development librarian at a public library in Washington. An author with no digital footprint is a red flag, especially if they are credited with a slew of titles each year. Also a red flag: a book with no author listed at all.
• Exercise caution regarding self-published books, small presses, or platforms such as Amazon, which filters out less AI-generated content than other vendors do.
• Think about whether the book is capitalizing on the chance that a reader will confuse it with another, more popular book, says Jane Stimpson, a library instruction and educational technology consultant for the Massachusetts Library System. Does it have a cover similar to that of an existing bestseller? Just as animated Disney movies get imitated by low-budget knockoffs, popular titles get imitated by AI-generated books.
• Check if there is mention of AI use in the Library of Congress record associated with the book, says Sarah Manning, a collection development librarian at Boise (Idaho) Public Library (BPL). If the book has been registered with the US Copyright Office, its record may mention AI.