Google AI Overviews: Meaning, Controls and Measurement
Google AI Overviews are AI-generated summaries that Google Search can show within the results for some searches. Each Overview combines a generated answer with links to web pages, so the reader can check what was said or keep reading at the source. Google decides whether an Overview is useful for a given search. No query, page or website is guaranteed one.
That conditional, generated format is what makes the feature hard to work with. Site owners want to know whether their pages can appear as supporting links, which settings change that, and what Search Console counts when they do. This entry answers those questions from Google's own documentation and one independent study, and it says plainly where the public evidence runs out.
What Google AI Overviews Are and When They Appear
AI Overviews are a feature of Google Search, not a separate chatbot or app. Google's AI features documentation and its Search Help page on AI Overviews describe the same basic shape: a summary generated by Google's AI systems, shown inside Search results, with links that lead to pages on the web.
Display is conditional. Google shows an Overview when its systems judge that one would be useful for the search, and it does not publish the thresholds behind that decision. So the absence of an Overview is not an error, and its presence for one query tells you little about another. Anyone who claims a reliable way to trigger the feature is claiming more than Google has disclosed.
Outside tracking shows how much display can shift. Semrush, which follows a large sample of keywords in its own database, reported that the share of those keywords showing an AI Overview changed considerably between January and November 2025 (Semrush AI Overviews study, December 15, 2025). That finding describes Semrush's keyword sample, not the share of all Google searches, and it is not a rule for when Google shows the feature.
The type of query can matter too. In a June 2026 chapter, the Reuters Institute for the Study of Journalism at the University of Oxford compared screenshots of Google searches in the United Kingdom: some breaking-news searches showed Top Stories, while a more contextual news-related search showed an AI Overview (Reuters Institute, June 16, 2026). These are illustrated examples from one country, not evidence of a general rule for news searches.
Availability Is Not Frequency
Google's Help page lists the countries and languages where AI Overviews are offered, and the list includes Indonesia and the Indonesian language. Being available in a market is not the same as appearing often there. None of the sources reviewed for this entry measure how frequently Overviews appear for searches in Indonesia.
The same Help page explains that searchers can switch to the Web filter, a results view focused on text links. That is a choice made by the person searching. It is not a publisher setting, and it does not remove a site's content from AI Overviews for anyone else.
From Search Labs Experiment to Search Feature
The feature has a short, dated history. In May 2023 Google announced a generative AI experience in Search, known as the Search Generative Experience (SGE), as an experiment in Search Labs (Google, May 10, 2023). A year later, at I/O 2024, Google announced AI Overviews for Search users in the United States (Google, May 14, 2024). These are dated announcements. They do not show that AI Overviews are a simple rename of the experiment, and they say nothing about the systems Google uses today.
How Supporting Links and Query Fan-Out Work
Google says AI Overviews and AI Mode may use a technique it calls query fan-out, which means running several related searches on subtopics to help build a response (Google Search Central). The word "may" matters. Google does not say every Overview uses fan-out, and it does not publish the related searches it runs for any query. A list of the exact sub-queries behind a specific Overview is a guess, not documentation.
An Overview can show more than one supporting link. What the public can see, though, is only the set of visible links. Google has not disclosed how many documents its systems consider before writing a summary, how those documents are scored, or which sentence came from which page. A visible link tells you that Google displayed your page alongside the summary. It does not prove that a particular sentence was derived from that page, and it does not reveal how many other pages were weighed and left out.
What is visible can still be studied. A 2026 paper in the Proceedings of Machine Learning Research proposed citation measures that take a source's search rank and provenance into account, and applied them to Google AI Overviews for high-stakes Your Money or Your Life (YMYL) queries drawn from the MS MARCO Web Search dataset (Kakimov and colleagues, 2026). The authors reported differences in citation patterns after controlling for retrieval rank. Their audit describes observable citations in one query sample. It does not reveal how Google selects sources.
Links, Mentions and Clicks Are Different Events
Measurement vocabulary helps here. The IAB framework Measuring Visibility in the AI Era, published in August 2026, separates being mentioned in an AI answer from being cited with a link, and both from a person actually clicking through. Those distinctions apply directly to AI Overviews. A brand can be named without a link, linked without a click, and clicked without a sale. The IAB document is an industry framework. It does not describe Google's internal attribution, and it does not define Search Console metrics.
How AI Overviews Differ From Featured Snippets and AI Mode
Three Google surfaces are often confused, and the differences affect both expectations and measurement. The table below sets them side by side using Google's own descriptions.
| Feature | What the reader sees | How the content is produced | Google documentation |
|---|---|---|---|
| AI Overviews | A summary inside Search results with supporting links | Generated by AI and shown only when Google judges it useful | AI features and your website |
| Featured snippet | Text taken from one identifiable webpage, with a link to that page | Displayed from the page's own text | Featured snippets and your website |
| AI Mode | A separate, more conversational Search experience with follow-up questions | Generated by AI and may use query fan-out | AI Overviews in Google Search Help |
Featured Snippets Show Page Text, Overviews Generate It
A featured snippet displays material from a single webpage that the reader can identify, together with a link to that page (Google Search Central). An AI Overview, by contrast, is generated text with supporting links. The practical difference is attribution. With a featured snippet, the displayed text and the linked page match. With an Overview, the links support the summary, but Google does not publish a sentence-by-sentence map. Google keeps separate documentation for each feature, and neither page describes a fixed selection formula.
AI Mode Is a Separate Experience
AI Mode is a distinct, more conversational Search experience in which a person can ask follow-up questions. Google has also described ways to continue from an AI Overview into a follow-up conversation in AI Mode, for example in its January 2026 update on Gemini upgrades. Announcements like that one, and Google's I/O 2026 Search update from May 2026, are dated statements about product changes. They do not tell you which model produced a specific Overview on a specific day, and features announced for AI Mode should not be read as features of AI Overviews.
Eligibility: What Google Requires and What It Does Not
To appear as a supporting link, a page has to meet the same basic conditions as other Google Search results. It must be indexed, it must be eligible to be shown with a snippet, and it must meet Google's technical requirements (Google Search Central). Since Google began rolling out its Search Console generative AI control on August 31, 2026, the setting for the page's property also has to allow participation (Search Console Help).
Meeting these conditions makes a page eligible. Eligibility is a requirement, not a promise. Google chooses whether to show an Overview at all and which links to include when it does.
This is also where ordinary SEO work fits. Google's 2026 guide on generative AI features says that standard SEO practices remain relevant for these features (Google Search Central). Crawlable pages, clear content and sound technical structure are the foundation, and teams that need help with those basics can review Arfadia's SEO services. That work supports eligibility. It does not guarantee inclusion.
No Special File or Markup Is Required
Google's guide to optimizing for generative AI features, updated in July 2026, addresses several popular shortcuts directly. It says Google Search does not require a special AI text file such as llms.txt, forced content chunking or special schema markup for its generative AI features. That statement covers Google Search only. Other AI platforms document their own behavior separately, and nothing in Google's guide speaks for them.
Where GEO and AEO Fit
The same guide acknowledges the labels generative engine optimization (GEO) and answer engine optimization (AEO), and treats the work they describe, for Google's own product, as SEO. That gives this entry a clear boundary. AI Overviews are one named Google Search feature. GEO is a broader program that tracks visibility across several AI platforms, such as the work described on Arfadia's generative engine optimization services page, and AEO is a related label for answer-focused optimization. The Related Terms section below links to both.
Publisher Controls: Four Separate Mechanisms
Site owners have four documented ways to influence how Google uses their content, and each one controls something different. Treating them as interchangeable can produce settings that do not do what the owner intended.
The Search Console Generative AI Control
Google added a generative AI control to Search Console, with a rollout that began on August 31, 2026 (Search Console Help). Each property can be set to Include, Exclude or Inherit. The setting covers AI Overviews, AI Mode and specified generative features in Discover.
Inheritance follows the property structure. A property set to Inherit takes its state from the nearest ancestor property that has been configured manually. If there is none, the top-level domain setting applies. The default is to include, but an inherited Exclude overrides that default.
A hypothetical example shows how this plays out. A site owner sets the property https://example.com/research/ to Exclude and leaves a narrower property, https://example.com/research/reports/, at Inherit. The narrower property follows its nearest configured ancestor, so it is excluded too, even though the general default would include it.
According to the Help page, Exclude limits whether the property's content and links are shown in the covered features and whether that content is used directly to generate those responses. It does not change rankings elsewhere in Search, it does not affect Google Ads or Merchant Center decisions, and it is not an AI training control. Changes are not instant either. Google describes them taking effect over a period of days, with further delay possible while cached data refreshes.
Googlebot Access and Page-Level Directives
Crawling and indexing are separate steps. A robots.txt rule for Googlebot controls whether Google can crawl a URL, while the noindex rule controls whether a page can be indexed and served (robots meta tag specification; Google's common crawlers). One consequence is easy to miss. If robots.txt blocks a page, Googlebot cannot fetch it, so it cannot see a noindex or nosnippet rule added later. Blocking crawling is therefore not a reliable way to remove a page from results.
Several preview rules also apply to AI Overviews. The nosnippet rule prevents a text snippet and also prevents the page's content from being used as direct input for AI Overviews and AI Mode. The max-snippet rule limits snippet length. The data-nosnippet attribute marks specific parts of a page that should not appear in snippets. A robots meta tag applies to the page that carries it, and an X-Robots-Tag header applies to the response that sends it. Neither one is inherited automatically by other URLs in the same folder. The specification also lists exceptions, so check the current text before relying on a rule.
Google-Extended Is Not an AI Overviews Switch
Google-Extended is a product token used in robots.txt, and it has no HTTP user-agent string of its own (Google's common crawlers). Google documents it as a way to manage whether content is used for certain Gemini Apps and Vertex AI purposes, including training and grounding. Google does not document it as an opt-out from AI Overviews, and it should not be treated as one. The Search Console Help page for the generative AI control separately refers to the training of Search models. This entry makes no claim about how the two controls interact in practice.
Measuring AI Overviews in Search Console
Search Console reports AI Overviews in two places, and the two use different rules. Reading them together, without mixing their metrics, is the safest way to understand what happened.
The Standard Search Performance Report
In the standard Search performance report, a click on an external link inside an Overview counts as a click. The Overview itself counts as one result position, and all of the links inside it share that position (Search Console Help). That shared position is not automatically position 1. Impressions follow Search Console's documented rules for when a result counts as shown, and those rules depend on the result type, so it is worth reading the Help article on impressions, position and clicks before interpreting any change.
The Generative AI Performance Report
Google also added a dedicated Generative AI performance report for Search, with a worldwide rollout noted on August 31, 2026 (Search Console Help). It records impressions for AI Overviews and AI Mode. It does not provide its own click or click-through rate metrics, so it cannot be read as a click report. Search Labs experiments are not included, and Discover has a separate report.
The report offers familiar dimensions: pages, countries, dates and devices, plus filters for text and multimodal searches. Dates follow Pacific Time, and pages are usually attributed to their canonical URL. Google's Help page also lists limits worth knowing before anyone builds a dashboard on this data. The table shows up to 1,000 rows, recent periods are marked as preliminary with a dotted line, and totals can differ between the chart and the table and between property-level and page-level views.
A hypothetical walk-through shows how the pieces fit. Suppose a page appears as a supporting link in an Overview, and a reader later clicks that link. The Overview being shown can produce an impression in the Generative AI report. The click appears in the standard Search performance report, at the Overview's shared position. Neither report shows which sentence in the summary relied on the page, and neither one measures business outcomes such as leads.
What an Independent Study Measured
Search Console shows what happened to one site. For user behavior across many searches, the most useful independent source in this review is a Pew Research Center study, Do people click on links in Google AI summaries?, published on July 22, 2025.
Pew analyzed the browsing activity of a panel of 900 U.S. adults during March 2025. Between April 7 and April 17, 2025, it reconstructed the search result pages for 68,879 unique queries, of which 12,593 were classified as producing an AI summary. The reconstruction captured up to three of the links shown in each summary. Every figure in this section comes from Pew's published method and results, and its confidence level is VERIFIED because it was checked against the original study. The table below reports the click measures as Pew defined them.
| Measure (source: Pew Research Center) | Result | Population and period | Confidence |
|---|---|---|---|
| Visits to a results page with an AI summary that included a click on a traditional search result link | 8% | 900 U.S. adults, March 2025 browsing | VERIFIED, observational |
| Visits to a results page without an AI summary that included a click on a traditional search result link | 15% | 900 U.S. adults, March 2025 browsing | VERIFIED, observational |
| Visits to a results page with an AI summary that included a click on a link inside the summary | 1% | 900 U.S. adults, March 2025 browsing | VERIFIED, observational |
These figures describe one U.S. panel in one period. They are observational, so they do not show that AI summaries caused the difference. They also cannot be applied to Indonesia, to search behavior in 2026, or to the traffic of a particular website.
Other studies use different designs, and their results should not be merged with Pew's. Ahrefs, for example, compared aggregated desktop click-through rates from Search Console data for December 2023 and December 2025 across a selected group of keywords with AI Overviews and a comparison group of the same size. It reported an association between AI Overviews and lower click-through for top-ranked pages (Ahrefs, February 4, 2026). That is a vendor's modeled comparison of keyword groups, not a count of individual user visits like Pew's, and not proof of lost traffic for any particular site.
Accuracy and How to Read AI Overviews
Google's Help page tells users that AI Overviews can contain mistakes. In May 2024, shortly after the U.S. launch, Google published a post explaining several erroneous Overviews and the changes it made in response (Google, May 30, 2024). That post is Google's own account. It offers examples and remedies, not an independent accuracy rate, and none of the sources reviewed here provide a current independent accuracy rate.
For readers, the practical rule is simple. Treat an Overview as a starting point and check important claims against the linked pages or other primary sources. For marketers and analysts, it means keeping three questions apart: was the page shown, was it clicked, and did that visit matter to the business. Search Console answers the first two only partly. The third needs the site's own analytics, and no source here establishes how AI Overviews affect a particular website's traffic or revenue.
Frequently Asked Questions
What are Google AI Overviews?
Google AI Overviews are AI-generated summaries that Google Search can show for some searches, with links to web pages that support or expand on the summary. Google decides when an Overview is useful, so none is guaranteed for any query, page or site.
Why don't AI Overviews appear for every search?
Google shows an Overview only when its systems judge that one would be useful for the search, and it does not publish the thresholds it uses. A missing Overview is expected behavior, not an error, and there is no documented way to trigger one reliably.
How is an AI Overview different from a featured snippet?
A featured snippet displays text taken from one identifiable webpage, with a link to that page. An AI Overview is generated text with supporting links, and Google does not publish which sentence relied on which link.
Are AI Overviews the same as AI Mode?
No. AI Mode is a separate, more conversational Search experience in which people can ask follow-up questions. Both are Google Search features and both may use query fan-out, but they are distinct experiences, and AI Mode announcements should not be read as AI Overviews features.
Does a site need llms.txt or special schema to appear in AI Overviews?
No. Google's 2026 guide says Google Search does not require a special AI text file such as llms.txt, forced content chunking or special schema for its generative AI features. A page must be indexed, eligible for a snippet, technically accessible and allowed by its Search Console setting, and even then display is not guaranteed.
How can a site owner keep content out of AI Overviews?
The main option is the Search Console generative AI control, which can set a property to Include, Exclude or Inherit. The page-level nosnippet rule also prevents content from being used as direct input for AI Overviews. Google-Extended is not an AI Overviews opt-out, and blocking Googlebot in robots.txt is not a reliable removal method.
What does the Generative AI performance report in Search Console measure?
It records impressions for AI Overviews and AI Mode in Google Search. It does not provide separate click or click-through rate metrics. Clicks on links inside an Overview appear in the standard Search performance report, where the whole Overview counts as one position shared by its links.
Related Terms
- Featured Snippet: the conventional answer feature that shows text from one webpage
- Generative Engine Optimization (GEO): the broader practice of building visibility across AI platforms
- AI Citation: what a visible source link in an AI answer does and does not show
- Answer Engine Optimization (AEO): a related label for optimizing content for direct answers
Sources
- Google Search Central. AI features and your website. Updated December 10, 2025.
- Google Search Central. Optimizing your website for generative AI features on Google Search. Updated July 10, 2026.
- Google Search Help. Find information in faster and easier ways with AI Overviews in Google Search. Publication date not stated; accessed October 16, 2026.
- Google Search Console Help. Search generative AI control. Publication date not stated; accessed October 16, 2026.
- Google Search Console Help. Generative AI performance report (Search). Publication date not stated; accessed October 16, 2026.
- Google Search Console Help. What are impressions, position, and clicks?. Publication date not stated; accessed October 16, 2026.
- Google Search Central. Robots meta tag, data-nosnippet, and X-Robots-Tag specifications. Updated March 24, 2026.
- Google Crawling Infrastructure. Google's common crawlers. Updated October 15, 2026.
- Google Search Central. Featured snippets and your website. Updated December 10, 2025.
- Pew Research Center, Athena Chapekis and Anna Lieb. Do people click on links in Google AI summaries?. Published July 22, 2025.
- Interactive Advertising Bureau (IAB). Measuring Visibility in the AI Era. Published August 2026.
- Google, Elizabeth Reid. Supercharging Search with generative AI. Published May 10, 2023.
- Google, Elizabeth Reid. Google I/O 2024: New generative AI experiences in Search. Published May 14, 2024.
- Google. AI Mode in Google Search and AI Overviews get Gemini upgrades. Published January 27, 2026.
- Google, Elizabeth Reid. Google Search's I/O 2026 updates: AI agents and more. Published May 19, 2026.
- Google. What happened with AI Overviews and next steps. Published May 30, 2024.
