AI companies are increasingly turning to printed books published before 2022 for training material, as web content becomes more likely to contain AI-generated text. ISBNdb, a company that helps AI companies source books in bulk, states that pre-2022 books are free from AI-generated content and less vulnerable to techniques designed to interfere with AI model training. ISBNdb explains that "Print books from the pre-LLM era are structurally guaranteed to be free of this contamination," and "physical books published before this date [pre-2022] are structurally clean of modern poisoning tools." The company notes that AI-generated content on the internet raises the risk of "model collapse," where AI systems trained on AI-generated data become less reliable. ISBNdb helps AI companies acquire between 1,000 and one million printed books per order. Interest in large-scale book purchases increased after court filings in a copyright lawsuit against Anthropic revealed plans to obtain, scan and destroy millions of printed books for AI training. ISBNdb signs confidentiality agreements with customers, stating "Strict NDA on every engagement," and acknowledges concerns about book destruction, noting "The optics problem is real."