Eligibility is ordinary indexing
The most useful thing to understand about AI Overviews is how little of it is new. There is no separate crawler, no separate index and no opt-in. Google assembles these summaries from pages already indexed by Googlebot, which means the prerequisites are the familiar ones: the page is crawlable, indexable, canonical, renders its content reliably, and is relevant to the query.
This is genuinely good news for anyone with a functioning SEO programme, because the foundation transfers wholesale. It is also why the AI Overviews advice that recommends exotic new tactics should be treated sceptically. If a page is not indexed, no amount of passage engineering will surface it. If it is indexed and relevant, most of the remaining work is about extractability.
Google-Extended is a different control
The confusion worth clearing up is Google-Extended. It is a robots.txt directive governing whether your content is used for Gemini model training and grounding. It is not Googlebot and it does not affect Search.
You can block Google-Extended and remain fully present in Search results and in AI Overviews sourced through the Search index. Publishers who want to decline training use without losing search visibility can do exactly that, and the two decisions are genuinely independent.
The controls that do affect Overviews are the snippet directives — nosnippet, max-snippet and data-nosnippet — and they are blunt instruments. They restrict how much of your content Google can display anywhere, including in ordinary result snippets, which typically costs more in click-through than it saves in summarisation. There is no AI-Overviews-only switch.
What gets selected
Within the set of indexed, relevant pages, selection favours content that can be extracted cleanly and attributed confidently. The patterns are consistent with the other surfaces.
Direct answers positioned early do better than answers that arrive after several paragraphs of context. Clear question-shaped headings help, because they map onto the way these queries are phrased. Lists and tables are extracted more readily than the same information in prose, because the boundaries between items are unambiguous. Specific, checkable claims outperform general assertions.
Comprehensiveness helps only when it is navigable. A long page organised into clearly headed sections offers many retrieval points; the same length as undifferentiated prose offers few.
Where structured data actually helps
Structured data is worth implementing, and it is worth being precise about why. It does not make a page eligible for an Overview and there is no schema type that acts as a lever. What it does is remove ambiguity about what a page is, who wrote it, when it was updated and how its parts relate — and reduced ambiguity lowers the chance of your content being misread or misattributed.
Article with accurate datePublished and dateModified, a real author, FAQPage where you genuinely answer questions, BreadcrumbList for hierarchy and Organization for publisher identity together cover most of what is useful. Implement them accurately rather than extensively; schema that misdescribes a page is worse than none.
Measuring presence honestly
AI Overviews are harder to measure than the other surfaces, and it is worth saying so plainly. Search Console does not break out Overview impressions separately, so you cannot simply read the number off a dashboard.
What you can do is run your fixed query set and record, for each: whether an Overview appeared at all, whether you were cited within it, and which domains were. Because Google shows Overviews selectively and that determination shifts, the appearance rate itself is a moving target — which means a change in your citation share can reflect a change in how often Overviews are shown rather than anything about your content.
Track both numbers separately for this reason. Reporting citation share without also reporting how often an Overview appeared at all produces a metric that moves for reasons nobody in the room can explain.
What this does to your traffic, honestly
The question every publisher asks about AI Overviews is whether they cost clicks, and the honest answer is that it depends on the query and that anyone quoting a single universal figure is generalising past their evidence.
For a query fully satisfied by three sentences — a definition, a conversion, a simple factual lookup — the summary plausibly absorbs a visit that would previously have landed on a page. That traffic was often low-value anyway: a visitor who needed one number and left. For queries where the summary is a starting point rather than an answer, the cited links remain a route through, and the visitor arriving that way has already been pre-qualified by reading the summary.
The strategic consequence is not to fight the summary but to notice which of your pages exist mainly to answer a question that a summary now answers. Those pages were always vulnerable, and the useful response is to make them the entry point to something the summary cannot replace — a tool, a dataset, a comparison, a calculator, original research.
Measure it on your own site rather than accepting the industry number. Segment your pages by whether their primary query type triggers an Overview, and compare impression and click trends between those segments. That comparison is specific to you, and it is the only version of this question worth acting on.
Which queries produce an Overview at all
Before optimising for the summary it is worth knowing that most queries never trigger one, and the pattern is reasonably legible.
Informational questions with a reasonably settled answer are the strongest candidates — definitions, how-to, explanatory and comparative queries where multiple credible sources broadly agree. Navigational queries rarely trigger one, because the user wants a specific destination rather than a synthesis. Transactional queries are inconsistent. And Google is visibly more conservative on topics touching health, finance, legal and safety, where the cost of a confidently wrong summary is highest.
The practical consequence is that your query set should record whether an Overview appeared at all, separately from whether you were in it. A topic where Overviews are rarely shown is not one to spend effort on, and you cannot tell those apart from topics where you are simply losing unless you track both.
Expect this to move. The set of queries Google considers suitable has widened and narrowed more than once, and a topic that produces no Overview today may produce one next quarter. That is an argument for recording the appearance rate over time rather than sampling it once and filing the conclusion.
The practical priority order
For most sites the order is unglamorous. Confirm the pages you care about are actually indexed. Confirm they render their primary content server-side. Put the direct answer in the opening lines of each section under a question-shaped heading. Convert comparison prose into real tables. Add accurate dates and real authorship. Implement Article and FAQPage schema correctly.
Every item on that list improves classic search results as well, which is the reassuring part: unlike the engine-specific surfaces, work aimed at AI Overviews is almost never wasted if the summaries turn out to matter less than expected.
Primary references
Common questions
Is there a separate crawler for AI Overviews?+
No. AI Overviews draw on Google's existing Search index, crawled by Googlebot. Google-Extended is a distinct control governing Gemini training and grounding, and blocking it does not remove you from Search or from AI Overviews.
Can I opt out of AI Overviews?+
Not selectively while remaining in Search. The available controls are blunt: nosnippet, max-snippet and data-nosnippet limit how much of your content can be displayed anywhere, including in ordinary result snippets. There is no AI-Overviews-only switch.
Does structured data get me into AI Overviews?+
Not directly. Structured data helps Google understand what a page is about and is worth implementing, but there is no schema type that makes a page eligible. Eligibility comes from being indexed and relevant.
Why does the same query show an Overview sometimes and not others?+
Google shows them selectively based on query type, confidence and available sources, and that determination changes over time. Presence is not stable, which is another reason to measure across repeated dated runs rather than treating a single check as the answer.
Do AI Overviews reduce clicks?+
For queries fully answered in the summary, plausibly yes. For queries where the summary is a starting point, the cited links remain a route through. The honest position is that the effect varies by query type and that anyone quoting a single universal figure is generalising past their evidence.
Does blocking Google-Extended affect my Search rankings?+
No. It governs Gemini training and grounding only. Your Search presence, including AI Overviews sourced through the Search index, is unaffected.