Google AI Overviews and AI Mode: How Websites Become Sources

Google AI Overviews and AI Mode can bring information from a business website into a generated answer, with links that let the reader explore further. Appearing as a source depends on Google's Search systems, the usefulness of the information for that answer, and the site's eligibility and inclusion settings. There is no separate application for a citation and no guaranteed placement.

For a business owner, the important distinction is between making information available and having Google select it. A page can meet the requirements without being used in a particular response. A selected page can support an explanation without its business being recommended. Understanding that sequence makes both an SEO proposal and a visibility report easier to interpret.

Two experiences with different ways of exploring a question

An AI Overview is an AI-generated summary within Google Search results. It can bring together key information from several sources and provide links for further reading. Google says it shows these summaries when its systems consider generative AI especially helpful, such as when a person wants to understand information from a range of sources. An Overview does not appear for every search. Google's AI Overviews explanation describes the feature and its limitations.

AI Mode is a more extensive search experience built around AI responses, follow-up questions and links to the web. A person can continue a comparison or refine a question without starting an entirely separate search. Google describes it as useful for exploration, reasoning and questions that might previously have required several searches. It can also return web links when confidence in an AI answer is insufficient. Google's AI Mode explanation also warns that responses can misinterpret content or miss context.

These are related products, but their source lists are not interchangeable. Google says they can use different models and techniques, producing different responses and links. A business appearing in an Overview for a question therefore does not establish that it appears in AI Mode for the same wording. The Search Central feature guide makes that distinction explicit.

The information comes through Search discovery and retrieval

Google's ordinary search process has several stages. It discovers and requests pages, processes their content for its index, and selects results when people search. Discovery, inclusion in the index and display are separate events. A newly published page is not automatically in Google's index, and an indexed page is not automatically shown for a particular question. Google's explanation of how Search works describes those stages.

Generative search adds a way of using information found through Search. Google's current documentation describes retrieving relevant, up-to-date pages from its index and using information in those pages to help generate a grounded response. Grounding means connecting the answer to information retrieved for the question. The technical term, retrieval-augmented generation, is often shortened to RAG.

This helps explain why a live business webpage matters even when an AI model already knows something about the subject. Retrieving information at the time of the search is a different process from relying only on patterns learned during model training. Current service details or product specifications can be relevant evidence in their own right; they do not have to become part of a model's training before a search system can encounter them.

Google also documents query fan-out: the model can issue several related searches across subtopics and data sources. A single customer question can create more than one information need. Retrieved information is then brought together in a response with supporting links. The current Google generative AI guide explains both retrieval and fan-out.

Consider a hypothetical customer comparing a heat pump with a furnace for an older house. The question could involve climate, installation constraints, operating costs and the home's existing equipment. Different sources may supply different parts of the explanation. A manufacturer's specification may answer one point; a utility's program information may answer another. This illustrates the mechanism rather than predicting Google's actual searches or source choices.

The source set can consequently be broader than the pages an owner sees for the customer's original wording. It is unsafe to assume that the first ordinary result must become the first AI source, or that a page outside one observed result list is ineligible. Google has not published a fixed source-selection formula that lets a business convert a position into a promised citation.

Eligibility makes a page available; it does not choose the page

A supporting webpage must be indexed and eligible to appear in Google Search with a snippet, the short preview of a result. Google's current eligibility guidance also requires the site to be included in Search generative AI features through its Search Console control. That inclusion is the default, subject to a property's inherited or explicitly selected setting.

These conditions describe access and permission. They do not establish that the page is the best evidence for a particular response. Google's systems still decide what to retrieve, which information is useful and which links to display. Google expressly says compliance with requirements and best practices does not guarantee crawling, indexing or serving.

For an owner investigating a missing appearance, those distinctions prevent an immediate leap to rewriting the content. A page unavailable to Search has a different problem from an indexed page that simply was not selected in an observed answer. Likewise, a site excluded through the generative AI control has a participation decision to resolve, rather than a shortage of AI-specific formatting.

The evidence for these situations is also different. Search Console can establish facts about the page's indexing and the property's inclusion setting. A captured answer establishes what appeared in that particular observation. Neither is a complete explanation of Google's internal selection decision. An eligibility check should therefore be described as an eligibility check; it cannot certify a future source placement.

This is also why no provider can guarantee that a particular page will be cited for every searcher or on a fixed date. Source selection remains with Google, and the question, feature and available information can change. The meaningful claim is that a specific obstacle was removed or the information was improved, followed by evidence of what happened.

What a business website can contribute to an answer

A supporting link has a purpose within the answer. It may provide a definition, evidence, an explanation of a product or a detail needed for a comparison. Its role is not necessarily to endorse the organization that publishes it.

For example, a contractor's guide might explain a material's maintenance requirements. Linking to that guide does not, by itself, mean Google selected the contractor for a local job. A product page might support a statement about compatibility while another source supplies the comparison. The business has gained a form of visibility, but the reader's next decision is still separate.

Google's guidance emphasizes material that adds useful information beyond widely repeated summaries. Original research, a properly documented test or a firsthand explanation can give the reader something specific to evaluate. The value comes from the information and its support. An assertion that a company is experienced is different from a clear account of what it knows, what was examined and where the conclusion applies.

That does not mean every business needs a research department. A manufacturer's accurate product limitations, a service company's clear description of what is included, or a documented explanation of a real process can answer questions that generic promotional language leaves open. These are illustrations of reader value, not a list of factors with promised citation weights.

Local and retail information has an additional dimension. Google's current guide says generative responses can include products and local-business information, and identifies Merchant Center and Google Business Profile as relevant ways of maintaining those details. A website article is therefore only one possible contribution to a customer's experience. A shop's availability, location or business details may matter more to a particular question than another general guide.

Google's documentation on establishing business details explains how business profiles, website information and supported structured data can help Google understand a business. The relationship between a brand, its locations and its official information deserves a coherent explanation. Our business-information guide covers that foundation in more detail.

Why a special AI writing format is not an entry requirement

Google does not require an AI-specific schema type, a special text file or a separate Markdown version of a website for these Search features. Its guidance also rejects a required page length or the idea that information must be divided into tiny pieces for AI to understand it. Clear organization can help a reader follow a complicated subject without becoming an artificial format imposed on every page.

This matters when a proposal describes a formatting package as the condition for appearing in Google AI answers. A well-chosen table may make specifications easier to compare. A concise answer may suit a simple definition. A longer explanation may be needed for an important limitation. Those choices have a reader purpose; none grants a source slot.

Structured data is a standardized description of information on a page. It remains relevant to supported Google Search features, and its contents should agree with the visible information. That role should be kept separate from a claim that an added markup field makes Google cite the page. Google's own guide says structured data is not required specifically for generative AI search.

The same distinction applies to publishing more pages. Query fan-out describes what Google's system may do, rather than a requirement for a website to create a separate URL for every possible question. Google warns that creating variations primarily to manipulate rankings or generated responses can violate its scaled content abuse policy. A coherent page can address related questions where they belong together.

These are Google-specific statements. A file or access arrangement used by another service has to be evaluated under that service's documentation. Google's treatment of llms.txt, for example, does not prove how every other AI provider treats it. Nor does inclusion in Google's Search index establish inclusion in another provider's results.

Publishers have several controls with different consequences

Participation in a search feature, permission to crawl a website and permission for model training are different choices. Treating them as one switch can produce consequences the owner did not intend.

The Search generative AI control

Google's Search generative AI control governs whether a site's links and content can appear in AI Overviews, AI Mode and supported generative AI features in Discover. Google says the control became available worldwide on August 31, 2026.

The default is inclusion. A property can instead be excluded or inherit its choice from a parent property. An owner examining a particular section of a website therefore needs to know the effective setting, including inheritance.

Exclusion prevents the site's content and links from appearing in the covered features and prevents crawled content from being an input to their responses or previews. Google says this setting does not act as a ranking or inclusion signal for other parts of Search. It also does not control AI training. Changes require processing time, and other websites can still supply similar information about the same subject.

Page and text preview controls

Separate page-level rules affect how content is indexed or displayed. A noindex rule tells Google not to show the page in search results. nosnippet prevents a text snippet and prevents the content from being used as a direct input for AI Overviews and AI Mode. max-snippet limits the amount of text available for previews and direct AI input, subject to Google's documented exceptions. data-nosnippet excludes designated portions from snippets.

These controls can affect ordinary Search previews as well as AI features. A business wanting to limit one use should understand the wider consequence of the chosen control. The Google robots meta specifications also explain that Google must be able to crawl a page to read the rules placed on it.

Crawling and training permissions

Googlebot is used for Google Search crawling. Blocking its access can affect Search generally. Google-Extended is a separate product token for specified AI uses, including training and grounding uses described in Google's crawler documentation. It is not a substitute for the Search generative AI control or a way to improve rankings.

The distinction extends beyond Google. ChatGPT, Bing and other providers have their own products, access rules and reporting. Even Google's standalone Gemini app is a different experience from AI Mode in Search. A statement about one product's permissions should identify that product rather than refer vaguely to blocking or allowing “AI.”

What Google's current reporting can establish

Google now provides a dedicated Generative AI performance report for Search in Search Console. The official June 2026 announcement and report documentation state that these insights were rolled out to websites worldwide on August 31, 2026. Advice that Google offers no separate generative AI visibility report is out of date.

The report covers impressions in AI Overviews and AI Mode. An impression records that a link to the site was shown under Google's counting rules. The documented views include pages, countries, devices and dates, with text-based and multimodal Web search types. The dedicated view is part of the wider Search performance picture rather than an additional pool of traffic to add to the ordinary totals.

Its scope is visibility. The current documented report is not a query-by-query list of generated answers, an AI citation position report or a separate AI click-and-conversion report. It does not include Search Labs experiments. A report may also be unavailable when a site has too few impressions; its absence alone does not prove a technical failure.

Aggregation matters when reading the numbers. Google documents different treatment of property totals and page rows, so several links from the same site do not necessarily produce several impressions in the property's chart. An increase in that chart should not be described as an equal increase in unique people, visits or leads.

The business interpretation follows the event being measured. More impressions indicate more reported visibility within the covered features. A screenshot can show the wording and linked sources in one observed answer. Website analytics can supply evidence about visits and subsequent actions where tracking permits it. These records can complement each other without proving that one particular answer caused a sale.

Our guide to mentions, citations and referrals develops those measurement distinctions. A source link, a named recommendation, a visit and an inquiry answer different business questions.

What this means for SEO

Google's AI features use Search foundations while changing how information can be assembled and presented. A business still benefits from being understandable and accessible to Search. The additional interpretation is that a page can become part of a larger answer, and its contribution may be narrower than the business's whole offer.

An owner can therefore read a claim about “showing up in AI” more precisely: which Google experience, which page or business information, what observed appearance, and what evidence of the next action? Eligibility, selection, visibility and business results each require their own support. Preserving those distinctions makes the opportunity understandable without turning a changing search experience into a promised result.