Identifying AI-Generated Books

Submitted by Alexa Martin, Molly Moore, and Kim Tomblin

Image Credit: Forsyth County Public Library using Google Gemini's AI image generation features

Why are AI-generated books a problem?

Have you tried browsing for a book on Amazon lately? If you type in a vague search (think “motorcycle books for third grader”), you will probably encounter AI-generated books within the first several search results. 

This was the second result on Amazon for "motorcycle books for third grader."

There are people who run AI book mills that produce tons of titles, which are written, formatted, and illustrated entirely by AI with no human editing. They often upload their works to Amazon’s Kindle Direct Publishing program, which allows individual authors to submit up to 10 new titles per week. The titles are then printed on demand to fill customer orders. 

These books can wind up as top sellers on Amazon due to clever marketing and algorithm manipulation, but they may have low-quality writing and inaccurate information. Imagine the potentially disastrous consequences for people who turn to these books for medical or legal knowledge. It’s possible AI-generated books could improve in the future, but they’re just not great right now.

How can I identify books written by AI?

Look at a few key factors. Some AI-generated books may have many of these; some may only have a few.

• Publication Date: 2022 or later
• Book cover:
• Features AI-generated art
• Similar to an existing bestseller
• Title:
• Long and wordy
• Makes grand claims
• Seems primed for search engine optimization
• For example: The Ultimate Home Gardening and Yard Care Bible: A Detailed Guide for Homesteading Beginners (3 Books in 1) (I made this title up, but you get the idea)
• Publisher:
• Self-published or publisher does not have a legitimate website
• Author:
• No social media presence or website
• Published a large number of titles in a short time frame
• Published titles covering a wide array of topics, no expertise in any area (ex. May have travel guides, coloring books, youth biographies, and adult fiction thrillers) 
• No details that can be fact-checked in bio—Where do they live? What do they do for work? What are their credentials?
• Weird copyright statements
• Text of the book:
• Vague and generic, no humor or personal experiences
• Awkwardly written 
• Lacks citations or citation sources do not exist when searched
• Product Reviews:
• 1 and 2 star-reviews say it appears AI-generated (or note some of the above factors)
• Don’t rely on 5-star reviews alone; they may be fake

What are Collection Development staff doing to avoid purchasing AI-generated books?

We don’t use Amazon to browse potential books for the collection. But we still have to be on guard. Ingram carries independently published titles, some of which are AI-generated. They can junk up our searches, but we can filter by things like publisher and professional reviews. 

On an interesting note, Kim recently found a pirated AI-generated title on OverDrive. She noticed the author’s name appeared on the cover twice, it had a duration of only 1 hour and 14 minutes, and the narrator was listed as Digital Voice Madison G. We notified OverDrive and they promptly removed the title from their marketplace. While OverDrive carries some legitimate titles with AI narration, this was a unique situation because the title was already published by a real publisher.

This was the AI-generated title.

This is the legitimate version.

What can Public Services staff do to help?

Consider using NoveList Plus and NoveList K-8 Plus as additional book browsing resources. NoveList is available to all staff and patrons to help identify titles not in the Library’s collection that may be good candidates for ILL or the Suggestion for Purchase form.

It’s only going to get harder to identify AI-generated books. If you encounter a book in the collection that you think was substantially created by AI, please send it to the appropriate Materials staff member with a note. We’ll take a hard look to see if it meets FCPL’s Collection Development Policy.