AI-generated content rises as tech firms digitize physical books

Research indicates a significant shift in internet composition, with AI-generated or AI-assisted content making up approximately 35% of newly published websites as of mid-2025. A study involving the Oxford Internet Institute found that while AI text is associated with lower semantic diversity and more positive sentiment, it has not yet shown a statistically significant impact on factual accuracy or stylistic variety.

Data from Cloudflare and HUMAN Security suggest that AI-related traffic and autonomous AI agents are rapidly increasing, with the latter seeing a massive year-on-year growth. To combat ‘model collapse’—a phenomenon where AI models degrade by training on their own synthetic output—companies like Amazon and Anthropic are reportedly seeking high-quality, human-authored data.

This has led to controversial practices where companies purchase physical books to digitize them. Reports indicate that some facilities, such as an Amazon scanning site in Las Vegas, have been observed cutting the spines off books to facilitate rapid scanning, effectively destroying the physical copies to secure human-made text for training large language models.