A Google AI Overview is a short answer written by a language model that appears at the top of a Google search results page, above the usual list of links. Google generates it from pages in its own search index and attaches links to the sources it used. The regular results still appear below it, so the searcher sees a written summary and a ranked list on the same page.
AI Overviews are Google's version of AI search inside its classic search product. They grew out of an experiment Google previously called the Search Generative Experience, and they run on a version of Gemini, Google's family of large language models, adapted for search.
How does a Google AI Overview work?
An AI Overview is a form of retrieval-augmented generation: the system first retrieves text from the web, then a model writes an answer from that text. Each time an Overview is shown, the process runs in a few stages.
- Decide whether to show one. Google's systems judge whether a generated summary is likely to help for this query, and many queries never get one.
- Search the index. Google runs the query, and often several related queries, against its existing search index. The same crawling and indexing that power normal results supply the candidate pages.
- Pick passages. Ranking systems score sections of those pages against the question and keep the ones that answer it most directly.
- Write the summary. The model writes a few sentences or a short list from the selected passages.
- Attach links. Each part of the summary is linked to the pages that support it, shown as small link cards or inline link icons.
Because the input comes from the search index, an Overview reflects pages Google has already crawled, and a page missing from the index cannot appear in one.
When does Google show an AI Overview?
Google shows AI Overviews for some queries and not others, and the share changes as Google adjusts the feature. They appear most often for informational questions, where a summary of several sources is useful, such as "how does a heat pump work" or "difference between an LLC and a sole proprietorship".
They appear less often for navigational queries (a brand name typed to reach its website, for example) and for many simple transactional queries. Google also applies stricter quality thresholds to topics where a wrong answer could cause harm, such as health and finance, so Overviews on those topics may be shorter or absent. The same query can show an Overview one day and not the next, and the text can change between runs.
How is an AI Overview different from a featured snippet?
Before AI Overviews, Google already showed answer boxes called featured snippets: a quoted passage from one page placed above the results. The two look similar, but they are built in different ways.
| Featured snippet | AI Overview | |
|---|---|---|
| Text | Copied from one page | Written by a model from several pages |
| Sources | One link | Several links, often to different sites |
| Wording | Matches the source exactly | Paraphrased and combined |
| Stability | Usually the same for a query | Can vary between searches |
In practice, a featured snippet has one winner, while an AI Overview can cite several sites at once. A site that ranks lower in the regular list can still be one of the linked sources if one of its passages answers part of the question well.
How does a page get cited in an AI Overview?
A page has to clear the same basic requirements as any Google result, then pass a passage-level selection. How AI search chooses which sources to cite describes that selection in general terms. For AI Overviews the requirements look like this:
- The page is crawlable by Googlebot, Google's main crawler, and is indexed.
- It is eligible to show a snippet in normal search results.
- Somewhere on the page, a passage answers the question or a part of it in a few sentences that make sense on their own.
- Google's ranking systems trust it enough to include it among the candidates for the query.
Google has said there is no special markup or separate submission for AI Overviews. The pages that tend to be cited already rank for the query or for closely related queries, and they state facts plainly, for example with a definition in the first sentence, a list of steps or a table of real values.
Can a website opt out of AI Overviews?
AI Overviews use pages fetched by Googlebot, the same crawler that feeds regular search. Blocking Googlebot in robots.txt removes a page from AI Overviews, but it also removes it from normal search results. Google-Extended, the robots.txt token Google offers for controlling use of content in Gemini models, does not control AI Overviews. The distinction between these bots is covered in what are AI crawlers.
For finer control, the snippet rules that already exist for regular search apply. A nosnippet robots meta tag stops Google from showing any text from the page, which also keeps it out of AI Overviews. The max-snippet rule limits how much text can be used, and the data-nosnippet attribute excludes specific parts of a page. Each of these also changes how the page looks in ordinary results, so they reduce visibility in both places at once.
What do AI Overviews mean for traffic and measurement?
When an Overview answers a question fully, some searchers do not click any result. When they do click, the visit usually lands on a page linked in the Overview or on a regular result below it. In Google Search Console, impressions and clicks from AI Overviews are counted inside the normal web search performance report, with no separate filter. Google Analytics 4 records these visits as ordinary Google organic traffic, so they are hard to separate from classic search clicks. Referrals from ChatGPT or Perplexity are easier to isolate, as covered in how to see AI traffic in Google Analytics 4.
Since analytics cannot isolate Overview clicks, the usual way to know whether a site appears is to check the queries directly. The text and the cited links can change between searches, so a single check is a weak signal, as explained in measuring AI visibility. Running the same set of questions on a schedule and recording which sites are linked gives a clearer picture of AI visibility in Google, and the Searcherries tracker runs this kind of repeated check across AI platforms. On the website side, the inputs are the ones that already drive Google rankings: the page must be indexed and allowed to show a snippet, and its passages should answer a question in a few clear sentences.
