Gemini SEO: Google-Extended, Grounding and Gemini App Visibility

Google runs several generative surfaces, and most advice treats them as one. They are not one product.

• AI Overviews and AI Mode live inside Google Search. Google's AI features documentation says they "may use different models and techniques" and can show different links for the same question.
• The Gemini app is a standalone assistant. It has its own models, its own release notes and its own controls.
• Grounding with Google Search in the Gemini API is a fourth path. Developers switch it on inside their own apps and pay for it separately.

The practical consequence: a fix for one surface can do nothing for another. If you want the longer comparison, read whether Gemini is the same as Google AI Overviews.

As of October 2026, Google publishes no robots.txt token that targets AI Overviews or AI Mode alone. Google's position is that AI is built into Search, so Googlebot directives are the control.

Google-Extended is a token you address in robots.txt. Google's common crawlers documentation describes it as the control for whether your content is used to train Gemini models and to ground answers in Gemini Apps and Vertex AI. The same page states that it "does not impact a site's inclusion in Google Search nor is it used as a ranking signal."

That gives you a clean split. Googlebot decides whether you are in Search, which includes AI Overviews and AI Mode. Google-Extended decides whether Gemini can learn from and ground on your pages. The definition of Google-Extended covers the token itself, and the Google-Extended vs Googlebot comparison walks through each control side by side.

To opt out of Gemini training and Gemini app grounding while staying in Search, the rule looks like this:

User-agent: Google-Extended
Disallow: /

Leave your Googlebot rules alone. Blocking Googlebot to keep content out of AI features removes the page from Google Search as a whole.

It helps to split a Gemini answer into four stages, because you control only some of them.

Parametric knowledge

When Gemini answers without searching, it uses what the model learned in training, frozen at a cutoff date. Nothing you publish this quarter reaches that answer. Google-Extended governs whether your content can feed future training. Nobody, including the vendors, has measured whether excluding a site from training changes what a model later says about that brand.

Access

Google has to be able to fetch and store your page. Google-Extended, Googlebot rules, your firewall and your CDN all sit at this gate.

Retrieval

When Gemini grounds an answer, it writes its own searches. Google's Gemini API documentation says the model "automatically generates one or multiple search queries" and runs them. You never see those queries, so you cannot optimize for them directly.

Generation

The model then picks which retrieved sources to use and cite. That choice belongs to the model.

The index matters more than any on-page trick. In OtterlyAI's 2026 experiment, Google deindexed two AI-generated test sites, and Gemini, AI Overviews and AI Mode almost never cited them afterward, while Microsoft Copilot (which grounds on Bing) kept citing them. An engine cannot cite what its index no longer holds.

In your control

• The Google-Extended line in robots.txt, which decides Gemini training and Gemini app grounding.
• Googlebot rules, which decide Search inclusion, and with it AI Overviews and AI Mode.
• Index and snippet eligibility. Google's AI features page says a page must be "indexed and eligible to be shown in Google Search with a snippet" to appear as a supporting link, with "no additional technical requirements."
• Suppression controls: noindex, nosnippet, max-snippet and data-nosnippet. These keep content out of AI answers, on purpose or by accident.
• The Search Console generative AI control. It is opt-out: inclusion is the default. Where it exists, child properties inherit from the parent by default, so a property can be excluded without anyone touching it. It is still rolling out to a subset of owners.
• Firewall and CDN rules, which can block a crawler that robots.txt allows.
• The facts on your own pages. A specific price or one-line definition in fetchable HTML gives the engine a sentence it can quote.

Not in your control

• Whether Gemini searches the web for a given question.
• Which queries it writes and which sources it retrieves.
• Which sources it cites, and what it says about you.
• What the current model already learned in training.

One note on rendering: Googlebot executes JavaScript, unlike most non-Google AI crawlers. Client-side rendering is a bigger risk for ChatGPT and Claude than for Google surfaces, though server-rendered facts are still the safest choice.

Every site owner faces the same choice. Allow Google-Extended (the default if you write no rule), or block it.

If you allow it: Google may use your content to train Gemini models and to ground Gemini app answers. No evidence shows that allowing it gets you cited more. It keeps the grounding path open, nothing more.

If you block it: your content stays out of Gemini training and Gemini app grounding. Per Google, your Search inclusion, AI Overviews eligibility and rankings do not change. The Gemini app can still describe you from other sites that mention you and from what the model already learned.

Two hypothetical cases show how the trade-off plays out.

• Say you publish a paid research newsletter. Your content is the product. Blocking Google-Extended is a reasonable call, and you keep your Search presence.
• Say you sell project management software. You want the Gemini app to quote your current pricing from your own pricing page. Blocking grounding removes your page as a source, so Gemini leans on review sites and older knowledge instead.

The question to ask is simple: is the value of your content in people reading it on your site, or in AI answers repeating your facts correctly? The playbook for blocking AI training while staying in AI search covers the per-bot rules across vendors, and whether blocking Google-Extended changes what Gemini says about your brand goes deeper on the brand-answer side.

Work from the gates outward. Each step depends on the one before it.

1. Confirm Google indexes the pages that matter. Use URL Inspection in Search Console for your homepage, pricing and product pages. Only Search Console shows real index membership.
2. Check for accidental suppression. Look for noindex, nosnippet, max-snippet:0 and data-nosnippet in templates. These often get inherited from a component someone wanted to hide.
3. Check the Search Console generative AI control if your property shows it, including any setting inherited from a parent property.
4. Read your robots.txt for Googlebot and Google-Extended lines. Make the Google-Extended choice on purpose, and write it down.
5. Check your firewall and CDN. A bot-management rule can turn away a crawler your robots.txt allows, and nothing in a dashboard will tell you. The Gemini crawler controls guide covers this layer.
6. State your key facts plainly in HTML. Pricing, product names, what you do in one sentence. Make sure each fact says the same thing on every page.
7. Measure Gemini on its own. Ask the Gemini app the questions your buyers ask and record what it says. The free checker on this site gives a one-time snapshot for three buyer questions. For an ongoing view, see tracking what Gemini says about your brand.

• Blocking Google-Extended to leave AI Overviews. It does not do that. AI Overviews follow Search rules. See whether blocking Google-Extended removes your site from AI Overviews.
• Blocking Googlebot to stop AI features. That removes you from Google Search entirely. Use snippet controls or the Search Console control instead.
• Allowing Google-Extended and expecting citations. Access buys eligibility, not selection. The model still decides.
• Adding schema or an llms.txt for Gemini. Google says structured data is not required for generative AI search and that Google Search ignores llms.txt files. No Gemini documentation says the app reads llms.txt either.
• Rewriting pages in an "AI-friendly" style. Google's optimization guide says you don't need to write in a specific way for generative AI search, and there is no ideal page length.
• Reporting one "Google AI" score. The Gemini app and AI Overviews behave differently. A blended number hides which one is broken.
• Trusting robots.txt alone. It is necessary, not sufficient. Test that your CDN actually lets Google's crawlers through.