Google AI Overviews and AI Mode: what actually decides whether you're in

Inclusion in Google AI Overviews and AI Mode is controlled by the same things that control inclusion in Google Search: Googlebot access, indexability, and the snippet controls. Blocking Google-Extended does not remove you from either surface. Google’s documentation says so directly: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.”

That sentence is the most misread line in AI search. This article covers what it does and does not mean, and what has changed about Google’s surfaces in 2026.

How big are these surfaces now?

At Google I/O on May 19, 2026, Sundar Pichai announced that AI Mode had crossed one billion monthly active users roughly a year after launch, and that AI Overviews reach 2.5 billion monthly users. Independent trackers disagree on how many queries trigger an Overview: BrightEdge measured Overviews on about 48% of tracked queries in February 2026, while Semrush’s methodology showed a peak near 25% in July 2025 followed by a decline. The difference is mostly query mix, but either figure is a large multiple of the 6.5% Semrush measured in January 2025.

The traffic consequence is the same under both counts. Semrush data from late 2025 found 92–94% of AI Mode sessions end without a click to an external site. An Overview citation is a visibility event far more often than it is a visit.

What controls inclusion?

Google’s AI features documentation names exactly one access control and three display controls.

Access: “robots.txt directives for Googlebot is the control for site owners.” If Googlebot can crawl and index the page, it is eligible. There is no separate AI Overviews user agent to allow or block.

Display: the same preview controls that have governed snippets for years apply to AI features:

Control	Effect on AI Overviews and AI Mode
`<meta name="robots" content="nosnippet">`	Page text is not used in the response at all
`<meta name="robots" content="max-snippet:N">`	Caps how much text from the page can be used
`data-nosnippet` attribute	Excludes a specific element’s text
`noindex`	Removes the page from Search, and therefore from AI features

The audit’s own layout template sets max-snippet:-1 for this reason. A max-snippet:160 left over from an old SEO plugin caps the amount of your page an AI Overview can quote.

What does Google-Extended actually do?

Google-Extended is a robots.txt token, not a crawler. Google’s crawler documentation describes it as the control publishers use to decide whether crawled content “may be used for training future generations of Gemini models” and “for grounding (providing content from the Google Search index to the model at prompt time),” and then states: “Google-Extended does not impact a site’s inclusion in Google Search nor is it used as a ranking signal in Google Search.”

So a Disallow: / under User-agent: Google-Extended has three effects:

Your pages are not used to train Gemini.
Your pages are not fed to the Gemini app as grounding context.
Nothing changes for Search, AI Overviews, or AI Mode.

The audit’s robots check still counts Google-Extended as a critical agent because of effect 2. The Gemini app is a large answer engine in its own right, and blocking the token removes you from its grounded answers. It is labeled that way on the scorecard so the trade-off is explicit.

How does AI Mode pick what to retrieve?

Google describes AI Mode as using a “query fan-out” technique: the system “issues multiple related searches across subtopics and data sources” and assembles the answer from the union of those result sets. Google says this lets it show “a wider and more diverse set of helpful links” than a single ranking would.

Two consequences follow for a site owner.

Ranking still matters on Google’s surfaces, unlike ChatGPT’s. The 5W synthesis of citation datasets found that 93.67% of AI Overview responses cite at least one top-10 organic result, while only about 12% of URLs cited by ChatGPT and Perplexity overlap with Google’s top ten. Google’s answers are drawn from Google’s index, ranked by Google’s systems. If a page cannot rank for the sub-queries, it cannot be fanned in.

Sub-topic coverage is a retrieval feature, not a writing tip. A fan-out on “best time to replace a roof in Austin” might issue searches for roofing material lifespans, Texas hail seasons, and permit requirements. A page that answers one sub-question cleanly can be pulled in for that sub-question even if it never targeted the head term. That is the strongest argument for the question-shaped headings and front-loaded answers the Content quality checks look for.

Why does the URL match rate look so low?

The same 5W synthesis reports that only 4.5% of AI Overview URLs directly match a page-one organic URL, even though most responses cite a top-10 domain. Overviews frequently cite a deeper page from a ranking site rather than the ranking page itself. Practically: the page that ranks and the page that gets cited are often different pages on the same domain, and the cited one is the one with the tightest answer.

What should you change?

Very little that is Google-specific. In order:

Confirm Googlebot is not blocked by robots.txt or a WAF rule. Search Console’s URL inspection is the ground truth.
Remove snippet caps you did not choose deliberately. Grep templates for max-snippet and nosnippet.
Make the extractable answer obvious. One <h1>, question-shaped <h2>s, and a direct answer in the first paragraph under each. This is the structure that survives fan-out.
Keep dateModified honest and current in Article JSON-LD. Freshness is used in retrieval and shown in the Overview card.
Decide on Google-Extended separately, knowing it is a Gemini decision and not a Search decision.

What has not changed

Google has said since May 2025 that its AI surfaces are built on the core ranking systems and that ordinary SEO remains the way in. Everything in the audit’s Core SEO category applies unchanged. The AEO-specific work is on the other engines, where ranking on Google is nearly irrelevant to being cited.