US v. Google: Google says its controls for publishers to opt out of AI training only cover work by Google DeepMind, not training by its other units for search
Google can train its search-specific AI products, like AI Overviews, on content across the web even when the publishers have chosen …
Context & Ripple Effects
Google had previously presented a robots.txt-based mechanism that let publishers exclude their data from Google AI training while staying in Search. This account draws a narrower boundary: the control applies to DeepMind work, not search-specific AI training.
The distinction helps explain why later coverage focused on service-level choices, including Google's exploration of website controls for AI Overviews and AI Mode and reported requests for broad content rights from Google News AI-feature testers.
First-order effects
- Publishers that opted out of AI training do not gain that protection for Google Search's AI products, including AI Overviews, under Google's stated interpretation.
- Google's search organization can continue using web content for its own AI training even where the publisher believed its AI-training preference had been registered.
Second-order effects
- Publishers face a sharper trade-off between remaining available through Google Search and seeking meaningful limits on AI use, while Google faces pressure to make controls specific to products rather than internal units.
- The gap between a model-training opt-out and search-AI training makes the scope of publisher consent a more central issue in negotiations over access and content rights.
Third-order effects
- If this interpretation persists, AI-content governance may shift from broad company-level opt-outs toward service-by-service permissions, with search distribution determining how much leverage publishers have.
- The later move toward AI Overviews and AI Mode controls suggests regulators and publishers may increasingly test whether meaningful choice requires separating indexing, retrieval, and model training.
The trend: This is one data point in the move from general web-crawling permissions toward granular publisher controls for AI-powered search products.