The ISBNdb AI training books page is gone. After 404 Media reported that the book database company had advertised a service to source printed books for AI training datasets, ISBNdb removed the offer and said the page had been only “a test of market interest.”
ISBNdb posted a note to its homepage and news page on July 30, nine days after 404 Media’s report. The company said it had “never purchased, scanned, or sold a book,” whether for AI training or any other purpose, and said it does not train AI models.
The company also told 404 Media the same thing in response to a request for comment after the pages came down. According to 404 Media, ISBNdb had not responded to earlier requests before the removal.
Why did ISBNdb remove its AI training books page?
ISBNdb removed a landing page titled “Printed Books Sourcing for Your AI LLMs Dataset Needs” on July 28. In its update, the company said the page was part of an effort to gauge demand and that it had decided to move away from that idea.
The company said its regular business, a book metadata API built around ISBN records, remains unchanged. In its July 30 note, ISBNdb described itself as a cataloging service that provides data about books rather than the books themselves.
The removed pitch matters because printed books have become attractive to AI companies looking for training text that predates the flood of machine-generated writing online. A large language model learns patterns from large collections of text; for a plain-English explanation of the mechanism, see Kernel’s guide to how LLMs answer prompts.
404 Media reported that one deleted ISBNdb article argued books published before 2022 were useful for training because they were less likely to contain AI-generated material. The same article said ingesting those books could help avoid model collapse, and suggested that authors opposed to AI use of their work could write with the goal of manipulating or sabotaging AI systems.
404 Media also spoke with booksellers and reviewed online reports describing a recent jump in unusual book orders. One bookseller told 404 Media the orders were notable for both volume and odd selection, with buyers showing little concern for price.
In another removed page, ISBNdb argued that buying used physical books in bulk did not take money from authors because the books had already completed their original commercial sale. That framing is likely to irritate authors who see their work being converted into training material without a new license, regardless of whether the copies came from the secondary market.
The issue is already visible in litigation around AI training. 404 Media cited a January copyright lawsuit brought by book authors against Anthropic, which revealed internal documents showing that Anthropic planned to acquire, scan and destroy millions of printed books. The Washington Post later reported that Anthropic was buying books through Better World Books, a marketplace used by libraries, retailers and individuals.
ISBNdb’s own removed marketing materials acknowledged the reputational risk of destroying books during scanning. One archived page said the “optics problem” was real and that a headline about an AI company destroying two million books would not win much sympathy.
This story draws on original reporting from 404 Media.
