Google puts AI Overviews on search

Google announced at I/O 2024 that AI Overviews — synthesised, AI-generated answers displayed above the ordinary list of links — would begin rolling out to all users in the United States that week, with international expansion to follow. The feature used a custom version of Gemini combined with Google’s existing search and ranking systems, and Google said it expected the underlying “AI-organized” search experience to reach over a billion people by the end of the year.

Google positioned AI Overviews as an evolution rather than a replacement of conventional search: results still carried links to source pages, and the company argued that pages cited within an overview received more clicks than the same page would have as a standard listing. Announced alongside it were adjustable summary depth, multi-step reasoning for complex questions, and planned uses such as meal and trip planning — an attempt to move search from a list of links toward direct synthesis, mirroring the direction chatbot products had already taken.

The rollout was Google’s most consequential response yet to the competitive pressure created by ChatGPT and Bing Chat: putting a generative layer directly into the product responsible for the large majority of its advertising revenue, rather than confining it to a separate assistant. That made any errors highly visible by construction — AI Overviews answered queries a huge, unfiltered population of ordinary users typed into Google every day, not a smaller base of people who had opted into a chatbot.

Those errors surfaced within days. Screenshots of overviews recommending eating rocks and putting glue on pizza went viral by 23 May, and Google scaled the feature back at the end of the month — a rapid, public demonstration of the gap between a curated I/O demo and unrestricted deployment at search scale.