AI Overviews

For many searches, Google now answers the question itself before the list of results begins. A model from the Gemini family, customized for Search, condenses information from several web pages into a short text and links to the pages it relies on. Google calls these summaries AI Overviews. According to Google, they appear when its systems judge them to be especially helpful, so not on every search.

They grew out of an experiment. In 2023, Google tested the feature as the Search Generative Experience in Search Labs. In May 2024, it launched as AI Overviews for all users in the US. Since March 26, 2025, it has also appeared in Germany and other European countries, initially for signed-in users aged 18 and over. In October 2025, Google expanded AI Mode to more than 35 additional languages. AI Mode is a separate search experience for longer questions and follow-ups in a conversation.

For companies, AI Overviews change the path from a search to their own website. People who find the answer in the overview click through less often; sites linked as sources sit in a prominent spot. Rankings alone therefore no longer describe a company’s visibility in search. What also counts is whether its pages show up as sources in AI answers, and what those answers say about the company.

How an AI Overview is built.

For AI Overviews, Google combines a language model with its search systems. The principle resembles retrieval-augmented generation (RAG): the model writes the answer based on pages from the search index instead of relying only on what it learned in training. The pages the answer draws on appear as links in and next to the text.

According to Google Search Central documentation, AI Overviews and AI Mode may use a technique called query fan-out. The system runs several related searches across subtopics and data sources for a single query. Google says this lets it show a wider and more diverse set of links than a classic list of results. A page can therefore be cited for one aspect of a question even if it does not rank high for the original query.

Google decides query by query whether an overview appears at all. Like any language model, this one makes mistakes, for example summarizing a source incorrectly or wrongly combining statements from different pages. Google’s own help page says that AI Overviews can and will make mistakes, and recommends checking important information against the linked sources.

AI Overviews, AI Mode and featured snippets compared.

Three features sit above or in place of the classic results in Google Search, and they are often confused. A featured snippet is a highlighted excerpt from exactly one page; the text is taken verbatim from there. An AI Overview, by contrast, is generated text that pulls together information from several sources and links to them. Even when an overview relies on a single source, the text is still generated.

AI Mode is a separate area of Search with its own tab. Google recommends it for queries that call for further exploration, reasoning or complex comparisons. Users can ask follow-up questions in a conversation, and the answers are longer and more detailed. Both can use query fan-out, but according to Google’s documentation they may rely on different models and techniques, so their answers and links differ.

Answers from ChatGPT, Perplexity, Microsoft Copilot or the Gemini app are not AI Overviews. These services use their own search sources and their own rules for choosing what to cite. Anyone assessing their visibility in AI answers has to look at each separately and cannot infer the picture elsewhere from Google’s overviews.

Which pages get cited as sources.

Google’s documentation names a single technical requirement: a page must be indexed and eligible to be shown in Google Search with a snippet. There are no additional requirements or special optimizations, it says. Google explicitly states that new machine-readable files, AI text files, extra markup or special schema.org structured data are not needed. By that account, an llms.txt file plays no role in AI Overviews.

What counts are the familiar basics: pages Google is allowed to crawl, internal links that lead to them, important content as text rather than only in images or scripts, a good page experience, and structured data that matches the visible content. For the content itself, Google points to its guidance on helpful, reliable content, whose quality yardstick is known as E-E-A-T.

In practice, a page has the best chance if it answers a specific question or sub-question clearly and early, contributes first-hand experience or original data, and makes clear who wrote it and when. A spot as a source cannot be bought. Ads are a separate area with their own rules.

What is known about clicks.

How much AI Overviews change clicks to websites is disputed, and solid numbers are scarce. In July 2025, Pew Research Center published an analysis of the browsing data of 900 US adults, covering 68,879 Google searches from March 2025. About 18 percent of the searches produced an AI summary. Users who saw one clicked a traditional search result in 8 percent of visits, compared with 15 percent when there was no summary. They clicked a link inside the summary in 1 percent of visits.

Google disagrees. Total organic click volume from Google Search to websites has been relatively stable year over year, Head of Search Liz Reid wrote in August 2025, and click quality has gone up. She also acknowledged that traffic is shifting between sites: some lose, others gain. The post gives no concrete figures.

Both statements describe averages. How a particular website fares depends on its topics and queries. Explanatory content whose answer fits into a few sentences is affected differently from pages where people want to buy, book or compare. The only reliable guide is measuring your own numbers over several months.

Measuring and controlling through Search Console.

Google Search Console counts clicks and impressions from AI Overviews and AI Mode in the Performance report under the “Web” search type, together with the other search results. In June 2026, Google also announced a dedicated report on generative AI features. It shows impressions and which pages appear in AI answers in which countries. According to Google, it has been available to all websites worldwide since August 31, 2026 (as of November 2026).

The same rollout added a setting that lets site owners remove their site from AI Overviews, AI Mode and AI Overviews in Discover. Google says this does not affect rankings in classic search. Sites that opt out, however, also receive no traffic and no impressions from these features.

Finer control comes from the familiar snippet rules: nosnippet, data-nosnippet, max-snippet and noindex. They apply to classic search results as well. The Google-Extended control limits training and use in other Google systems, according to Google; it does not affect inclusion in Google Search. Before opting out, check the new report first: it shows how many impressions would be affected.

Where the term stops working

AI Overviews are a feature of Google Search; ChatGPT, Perplexity, Copilot and the Gemini app choose their sources by their own rules. The term describes how results are presented, not a method. The broader work on visibility in AI answers is called GEO, a featured snippet is a verbatim excerpt from one page, and AI Mode is the separate conversational area of Search. Whether and which source appears is decided by Google for each query.