Google AI Mode & AI Overviews: the complete 2026 guide

AI Overviews and AI Mode are two different things

This is the most common confusion, including in professional content: the two names get used interchangeably when they describe two distinct surfaces.

AI Overviews are a generated summary shown at the top of a standard search results page. They don't appear on every query, Google only triggers one when it judges that a summary adds something beyond the standard results. The blue links stay below. Google describes them as a way to get to the gist of a complicated topic faster, and as a jumping-off point to explore links.

AI Mode is a separate, conversational search mode: the generated answer takes over the page, you can ask follow-up questions, and the experience reads more like a dialogue than a list of results. Google positions it as particularly helpful for queries that need further exploration, reasoning or complex comparisons.

Both surfaces are enormous: Google states publicly that AI Overviews reach more than 2.5 billion monthly users and AI Mode more than 1 billion. Hold on to the distinction, user behavior differs, the queries that trigger each differ, and yet, as we'll see below, Search Console lumps both into a single report.

Where these features are live: the France rollout as a case study

Availability is uneven by market, and one recent rollout is worth studying because it shows what happens when a large market switches on all at once.

Google launched AI Overviews and AI Mode in France on July 22, 2026, across desktop, mobile and the Google app on Android and iOS. France came late relative to other markets, held back by the regulatory environment Google operates in there. That timing is specific to France, it says nothing about your own market's rollout date, which you should verify separately.

Three lessons generalize to any market that has recently switched on, or is about to:

• Your history is short. Year-over-year comparisons are meaningless on a surface that has only existed locally for weeks. The only valid readings are movements since launch.
• Local content on the topic goes stale fast. A large share of French-language articles still stated "not available here" well after the feature shipped. Check the publication date of anything you read about availability, including this page.
• Benchmarks don't transfer across markets. Trigger rates, which sources get cited, and how crowded a given question is have no reason to match what's been observed in the US over the past two years.

If your site has never been assessed from this angle, a GEO audit is the logical starting point: it checks technical eligibility before anyone starts arguing about content.

Query fan-out: why AI answers cite pages you weren't targeting

Google documents explicitly that both AI Overviews and AI Mode can use a technique called query fan-out: instead of running a single search, the system issues multiple related searches to build its answer, which lets it surface a wider and more diverse set of links than classic search.

The consequence is direct and counterintuitive: the page that gets cited is not necessarily the one ranking first for the query the user typed. It's often a deep page that answers precisely one sub-question the user never asked out loud, but that the system generated for itself.

That changes how you think about a page. The unit is no longer "the primary keyword" but the sub-questions covered. A page that touches ten aspects of a topic superficially is less useful to the system than a page that fully answers one of them. It also explains why pages you never considered strategic show up in the Search Console report, and why you have to read that report page by page rather than as a single site total.

The generative AI performance report in Search Console

On June 3, 2026, Google launched a dedicated generative AI performance report in Search Console. It matters more than it sounds: until then, publishers had no first-party data at all about their presence in Google's AI answers, only manual spot checks and third-party estimates.

The report measures the impressions your site earns inside Google Search's generative AI features, separately from classic search results. It breaks down by pages, countries, devices and over time. A separate equivalent report exists for Discover.

Two caveats before you go looking for it in your own account:

• It's still rolling out. Google states that not all properties have access yet. Not seeing the report does not mean your site is never cited.
• An impression threshold applies. A site needs to have received enough impressions in these features for data to appear at all, the same principle as the Discover report.

What the report measures, and what it doesn't

Read this before drawing any conclusion from a chart.

What the report gives you:

• One metric: impressions. How many times links to your site were shown to a user in a generative AI feature on Google Search.
• Four dimensions: Pages (the final URL linked, after redirects), Countries (where the search originated), Dates (days, weeks or months depending on the granularity you pick) and Devices (desktop, tablet, mobile).
• Two surfaces covered: AI Overviews and AI Mode. Google notes it expects to update that list over time.
• One aggregation rule worth knowing: if two results from the same site appear in the same generative answer, they count as a single impression in the property total.

What the report does not give you:

• No clicks in the generative AI report. Clicks from AI features are counted by Search Console, but inside the classic Performance report under the "Web" search type, blended with every other click and impossible to isolate.
• No queries. You learn which pages were shown, never for which questions.
• No position. Rank has no equivalent inside a generated answer.
• No unlinked brand mentions. The report measures a link to your site being displayed, not your brand being named in the answer text.

That last distinction is the one people get wrong most often: an AI impression is not the same as being quoted. An answer that discusses your brand without linking to you appears nowhere in this report.

How to read and analyze the report

A simple method, in this order:

1. Start with the Pages dimension. It's the most actionable view. You find out which URLs are genuinely being cited, rarely the homepage, often the deep pages that answer one question thoroughly, frequently not the ones you'd have bet on.
2. Filter by country. Essential if you operate across markets that switched on at different dates: a global curve mixes a two-year-old market with one that launched last month, and the blend tells you nothing.
3. Check devices. A sharp desktop/mobile gap is a signal about the kind of queries exposing you.
4. Then look at dates. Hunt for breaks, not absolute values. A clean drop on one page usually maps to a content change, an indexing problem, or a shift in what Google chooses to trigger.
5. Cross-reference with the classic Performance report. This is where the report earns its keep. For your key pages, compare Web impressions against generative AI impressions. Two profiles deserve attention: pages strong in Web but absent from AI (an eligibility or answer-format problem), and pages present in AI but weak in Web (content the system already validates, with ranking upside left on the table).

One habit to adopt immediately: export and archive your data. The report's history does not go back beyond its own launch, and you cannot reconstruct a baseline you never captured. Our GEO tracking tools exist precisely to keep that continuity over time.

How to tell whether Google AI cites your site

Three methods, complementary rather than interchangeable.

1. The Search Console generative AI report. The only first-party source, straight from Google. It tells you which pages were shown and how often. It does not tell you for which questions, or whether anyone clicked.

2. Manual checking. Asking AI Mode, or running the query in Search, the questions your customers actually ask, then recording which sources get cited. This is irreplaceable for context: who gets cited instead of you, how your category is framed, which phrasings trigger a generative answer at all. But answers vary with location, session and timing. A single test proves nothing.

3. Instrumented tracking. Replaying a stable sample of questions at regular intervals and following how the cited sources shift. It's the only way to turn scattered observations into a readable trend, and the only way to compare your presence against competitors.

For a first read with nothing to install, our free audit tool checks your site's technical eligibility and its visibility across answer engines. Used together the three methods give a reliable picture; used alone, each is wrong in its own way.

Why a page gets zero AI impressions

Before concluding you have a content problem, work through this diagnosis in order, technical causes are more common and far faster to fix.

1. The page isn't indexed. No generative surface can cite what isn't in the index. Always check this first.
2. The page is indexed but not snippet-eligible. Google states a page must be indexed and eligible to be shown with a snippet in Search. A nosnippet directive, a max-snippet:0, a misplaced data-nosnippet or a leftover noindex is enough to rule you out.
3. The property is excluded via the Search generative AI control. Covered in the next two sections, a setting sometimes flipped by a previous vendor that nobody remembers.
4. The topic doesn't trigger a generative answer. Google doesn't show an AI Overview on every query. On some intents there is simply nothing to capture.
5. Another source answers the sub-question better. Fan-out selects pages against precise questions: a generalist page loses to a specialist one, even when it outranks it on the head keyword.
6. Volume is below the report's threshold. Few impressions can mean no data displayed, which is not the same as no impressions.
7. The report hasn't reached your property yet. Google is rolling it out gradually. Rule this out first if no data appears anywhere.

The Search generative AI control: excluding your site from AI answers

On July 20, 2026, Google published Search Console Help documentation for a domain-level generative AI exclusion control.

You'll find it under Settings > Search generative AI. The choice is made at property level, and it is binary:

• Include your site's links and content in Search generative AI features. This is the default: the site can appear as a link, help ground the generated answer, and receive impressions and traffic from those surfaces.
• Exclude your site's links and content from those same features. Your content is then prevented from being visible to users there, neither as a link nor as grounding for the answer.

Child properties, subdomains, URL-prefix properties, can additionally inherit the setting from their parent property, or define their own explicitly.

Two things to hold on to. First, the scope goes beyond Search: the control covers AI Overviews, AI Mode and generative AI features in Discover. Second, there is no per-URL granularity. It's a switch for the entire property, not a checkbox you apply page by page.

When not to use the exclusion control

Since that documentation appeared, exclusion has been widely framed as the obvious answer to falling clicks. In most cases that reasoning doesn't hold, for four reasons.

Excluding recovers no clicks. The argument "Google uses my content without sending traffic, so I'll cut it off" assumes the traffic returns to the blue links. Nothing guarantees that. What exclusion definitely removes is the clicks those surfaces already send you.

The setting is global and binary. You cannot exclude only your blog, or only one sensitive section. It's the whole property or nothing.

You lose the measurement too. No impressions means no data in the generative AI report, you give up the very instrument that would have let you evaluate the decision.

Finer controls already exist. Before reaching for the global switch, max-snippet limits how much text can be shown, and data-nosnippet protects specific blocks within a page. These address a targeted problem without removing an entire surface.

There are, however, legitimate cases: contractual or legal obligations on licensed content, a regulated sector requiring strict control over how material is redistributed, or a site whose audience genuinely has nothing to gain from these surfaces. In those situations the decision gets made deliberately and documented, not in a panic after one bad traffic week. The setting is reversible, but recovery isn't instantaneous.

What actually influences being cited

On this point Google's position is explicit and worth quoting as-is: there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary. No dedicated markup, no special file, no "AI" meta tag is required for these surfaces.

What actually gates eligibility:

• Indexing and snippet eligibility. The page must be in the index and able to be shown with a snippet. That's the non-negotiable entry condition.
• Technical fundamentals. Crawlability, internal linking, page experience. Nothing new, and nothing optional.
• Helpful, reliable, people-first content that complies with Search policies.

Where real craft comes in is the shape of the answer, given how fan-out works: covering sub-questions explicitly rather than circling a keyword, writing direct and extractable answers, grounding claims in verifiable and dated facts, and making each page stand on its own without depending on its neighbors for context. That is exactly what Generative Engine Optimization is about.

One frequent mix-up to clear: Google-Extended is not a lever for AI Overviews or AI Mode eligibility. That control governs other Google AI systems. Blocking it does not remove you from these two surfaces, and unblocking it does not get you into them.

AI Overviews, AI Mode and classic SEO: how they relate

Two opposing narratives circulate, and neither is accurate.

The first says generative AI is a separate channel to be optimized on its own terms. It isn't: same index, same eligibility rules, same fundamentals. A site invisible in classic Search will not be cited by these surfaces.

The second says nothing has changed and good SEO is enough. That falls short. Fan-out changes the unit of selection: it's no longer a page ranking for a query, it's a passage answering a sub-question. And the fact that Google built a dedicated Search Console report at all shows it now treats these impressions as a measurable category in their own right.

The accurate reading fits in one sentence: the SEO foundation is a necessary condition, and answer-shaped structure is the differentiator. Sites gaining ground on these surfaces are rarely the ones that bolted on a technical layer, they're the ones that rewrote pages to answer rather than to rank.

What KPIs to build from AI impressions

Because the report carries neither clicks nor queries, several familiar metrics are simply out of reach. Don't try to calculate an "AI CTR": the data doesn't exist, and any ratio built on reconstructed clicks will be wrong.

Four indicators hold up:

• Coverage, the number of pages with at least one AI impression in the period. Early on, this is the best progress signal you have.
• AI impression trend on your key pages, page by page rather than as a site total.
• Per-market share, isolated by country, so you don't blend markets that opened on different dates.
• The Web/AI gap, pages that perform in classic search but are absent from generative surfaces. That's your work queue, in priority order.

One methodological caution to close on: don't conclude that a click decline was caused by AI Overviews without isolating it. The report cannot establish that causal link, and plenty of other variables move at the same time, seasonality, algorithm updates, changes on your own site. Observe and document, but don't attribute what the available data can't prove.

Practical checklist

• Check whether the generative AI report is available on your Search Console property, if it isn't, an absence of data proves nothing.
• Export and archive your data now: the history doesn't go back beyond the report's own launch.
• Filter by country to isolate markets that switched on at different dates.
• Analyze page by page first, never as a single site total.
• Confirm your key pages are indexed and snippet-eligible: check nosnippet, max-snippet, data-nosnippet and noindex.
• Verify under Settings > Search generative AI that your property isn't excluded without your knowledge, subdomains included.
• Don't flip the exclusion switch in reaction to a traffic dip: measure first, and prefer max-snippet or data-nosnippet for a targeted need.
• List the questions your customers actually ask and check manually, on a repeated sample, who gets cited instead of you.
• Rewrite your priority pages as answers to sub-questions rather than as content built around a keyword.
• Run a free audit for the technical baseline, then structure the work with our GEO audit method.

Summary table: AI Overviews, AI Mode and classic results

To close, here are the three surfaces side by side: what each one is, how a site appears in it, and the instrument available to measure it. The most revealing column is the last one, it's where the real limit of today's measurement sits, since Search Console does not separate AI Overviews impressions from AI Mode impressions.