Google put a generated summary above search results in 2024

A generated summary rolled out on 14 May 2024

Google's May 2024 announcement states that AI Overviews, a generated summary placed above the usual list of results, would 'begin rolling out to everyone in the U.S., with more countries coming soon', built on a custom Gemini model layered onto Search's existing ranking systems, with an ambition to reach 'over a billion people' by the end of the year. Google frames the change in its own favour: 'people use Search more, and are more satisfied with their results' once Overviews are shown, and the company states that 'the links included in AI Overviews get more clicks than if the page had appeared as a traditional web listing for that query', alongside a stated commitment to 'continue to focus on sending valuable traffic to publishers and creators'. These are the company's own measurements, presented without the underlying method, sample size or comparison period a reader would need to judge them independently.

A correction sixteen days later

By 30 May 2024, Google published a follow-up post conceding that 'some odd, inaccurate or unhelpful AI Overviews certainly did show up', citing examples including a satirical article about eating rocks being read as literal advice, and a forum post recommending glue to help cheese stick to pizza being summarised as a real cooking tip. The company said it made 'more than a dozen technical improvements', including better detection of nonsensical queries, limits on drawing from satire and from user-generated forum content for advice, and stricter triggering around health topics. It attributed remaining errors to 'misinterpreting queries, misinterpreting a nuance of language on the web, or not having a lot of great information available', and disclosed a content-policy violation rate of 'less than one in every 7 million unique queries'.

What this changes about starting research with a search

Google's own help-centre page for AI Overviews, describing the feature as retrieved on 16 September 2026, still tells users plainly that 'AI responses may include mistakes' and to 'always check important info in more than one place. Click the links to supporting information'. That is a meaningful shift in what a search result asks of a reader. A ranked list of links puts the synthesis step in the reader's hands; a generated overview performs that synthesis first and asks the reader to verify it afterwards, which is a different, and for an unfamiliar topic often harder, task than comparing a few sources from scratch.

• Is the topic one where a wrong first impression, later corrected, still causes harm, such as health or safety?
• Does the overview cite sources that can be checked directly, and has at least one been opened?
• Would the underlying pages, read directly, actually support the summary given?

Google's public record here is candid about the failure modes it found after launch, which is useful, but it is still the company describing its own product's errors and its own fixes. Research that starts with a generated summary carries that summary's blind spots forward unless the underlying links are actually opened.