Google Says It Won’t Pay to Train AI on Open‑Web Content

Overview

• In Lords testimony, Google’s Roxanne Carter said training LLMs on unpaywalled pages is statistical analysis that yields new outputs, so no licence is needed.
• She said payments apply to controlled access such as archives, specialized datasets or APIs, reflected in recent extended display and delivery deals with major publishers.
• Google Extended lets sites remain in Search while blocking training crawlers, yet it does not prevent use in AI Overviews.
• Publishers report lower click‑throughs from AI Overviews and warn that deploying NOSNIPPET to avoid summaries can depress traffic, with Google previously estimating about a 45% drop.
• The UK CMA has identified Google’s search and AI features as holding strategic market status and is consulting on possible interventions, keeping opt‑out and compensation questions open.