Google expands use of Search data to train Gemini AI models

Google is altering its Search service so that user interactions – clicks, re‑queries, and browsing behavior – are fed into training pipelines for its generative AI models, including Gemini. The company describes the change as part of an "always‑on" AI Search that continuously learns from real‑world usage.

Users can disable the new data collection through the Search Services History setting in their Google Activity page, either turning off further tracking or deleting previously recorded activity. Google notes that data already used for training may be retained for up to four years even after deletion, and the effectiveness of the opt‑out cannot be independently verified.

The shift raises privacy concerns for users and publishers, as more search activity stays inside Google’s ecosystem, potentially reducing traffic to external sites while providing the firm with a richer dataset for AI development.