Signal Audit found 45% of a Google training dataset restricted by site terms of service

Summary

This signal measures how much training data is now covered by restrictions from the websites it came from. A recent audit found that 45% of the data in one Google training dataset falls under website terms of service that limit its use. The finding comes as more sites and creators try to stop AI companies from using their content, including by changing their terms. The report cautions that such changes can be hard to enforce and do not undo scraping that has already happened. Creators are also turning to lawsuits and technical tools that block crawlers or poison data. The figure suggests the supply of freely usable web data for AI training is shrinking.