Google can train its search-specific AI products on web content across the internet even when publishers have opted out of AI training, according to testimony from a Google vice president in court on Friday. Eli Collins, vice president of product at Google DeepMind, explained that the company’s opt-out controls for AI training specifically cover Google DeepMind’s work. However, other teams within Google can further train the models for their own products, effectively bypassing the opt-out mechanism for search-related AI features like AI Overviews.
This distinction in how Google handles AI training data raises questions about content ownership and the company’s control over user data. The ability of Google to utilize web content for training its search AI, regardless of publisher preferences, highlights the complex relationship between tech companies and content creators in the digital age.
The testimony was given in court on Friday, as reported by Davey Alba for Bloomberg. The original article was updated on May 3, 2025, providing further insight into Google’s AI training practices and their implications for content publishers and users alike.
