Gemini Visibility Factors, Sorted by Evidence Strength

When people say "Gemini", they usually mean one of four Google paths. Each has its own controls, and fixing one does not fix the others.

• Gemini app. Governed by the Google-Extended token in robots.txt for training and grounding.
• Gemini API, Grounding with Google Search. A developer-invoked, separately billed path. The model writes and runs its own search queries.
• AI Overviews and AI Mode. Part of Google Search. Google says they "may use different models and techniques" and can return different links.
• Retrieval-off answers. What the model says from its training data alone, frozen at a knowledge cutoff. Nothing you publish this quarter reaches these.

Here is how the evidence grades the main factors, as of October 2026.

For the setup steps on each surface, see Gemini visibility: Google-Extended and grounding.

Google's common crawlers page (updated 2026-07-14) says Google-Extended controls training and grounding for Gemini Apps and Vertex AI. The same page says it "does not impact a site's inclusion in Google Search nor is it used as a ranking signal."

That split is the most common mix-up. Blocking Google-Extended keeps you out of Gemini app grounding. It does not remove you from AI Overviews or AI Mode. There is no robots.txt token that blocks AI Overviews or AI Mode. For those, Googlebot rules are the control.

A robots.txt that keeps Search but opts out of Gemini Apps looks like this:

User-agent: Googlebot
Allow: /

User-agent: Google-Extended
Disallow: /

If you want Gemini app grounding, make sure no blanket AI block catches Google-Extended by accident. Allowing it buys eligibility, not a citation.

The second gate is your network. A firewall or CDN can block a crawler that robots.txt allows. Since 2026-09-15, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains and existing free-tier customers. A site can lose AI access without anyone editing it.

One thing matters less here than elsewhere. Googlebot renders JavaScript, so client-side rendering is not the barrier for Google surfaces that it is for ChatGPT and Claude crawlers. It still matters if you care about those engines too.

Google's AI features page (updated 2025-12-10) states the requirement plainly. A page "must be indexed and eligible to be shown in Google Search with a snippet." It adds: "There are no additional technical requirements." It also says indexing and serving isn't guaranteed.

So the checks are mechanical:

1. Confirm the page is indexed in Search Console.
2. Look for noindex, nosnippet, max-snippet:0 and data-nosnippet that a template applied by mistake.
3. Check the Search Console generative AI inclusion control. Inclusion is the default, but a property can be opted out, and child properties inherit from the parent.

Otterly's 2026 experiment shows what losing the index does. Google deindexed both of its test sites, and Gemini, AI Overviews and AI Mode almost never cited them. Microsoft Copilot, which grounds on Bing, kept citing them. One pair of sites proves little on its own, but it matches Google's written rule: no index, no AI citation.

The 2026 critical survey of 45 GEO studies (arXiv:2607.14035) found that "topical relevance and context position are the most reproducible levers." A 252,000-trial factorial study reached a similar finding. Both measured pages already in the model's context, and both come from interested authors. Treat this as corroboration, not proof.

Relevance is not keyword coverage. Google's AI optimization guide says you don't need to worry about capturing every long-tail variation. The harder problem is that you can't see the input. The Gemini API "automatically generates one or multiple search queries", and AI Mode may fan out across subtopics.

The practical move is to cover the questions your buyers actually ask, with specific facts on pages that can be fetched. A stated price or a one-line definition gives the engine a sentence it can quote. The guide to correlation vs causation in AI search studies explains why most relevance findings stop short of a guaranteed lift.

These are sold widely for Gemini. The evidence does not back them.

• Writing style and chunking. Google's AI optimization guide (2026-07-10) says you don't need to write in a specific way for generative AI search, and that there is no ideal page length. C-SEO Bench found most rewriting methods ineffective and often harmful. See whether rewriting content for AI can hurt you.
• Schema markup. Google says structured data isn't required for generative AI search. Ahrefs tracked 1,885 pages adding schema and found AI Overview citations fell 4.6%. Those pages were already heavily cited, so the effect on pages trying to enter the set is untested. Existing markup does no harm. It just isn't a lever.
• FAQ schema. Google deprecated FAQ rich results in May 2026.
• llms.txt. Google says Search ignores it. No engine documents reading it.
• E-E-A-T scores. Google says E-E-A-T "isn't a specific ranking factor." There is no score to raise. The research on Google's helpful content guidance covers what the concept does mean.

A 2025 study did find generative engines prefer prose a model finds predictable, including on Google surfaces. Its causal test rewrote whole pages, so no single style choice was isolated. Nobody can act on it reliably.

Some pages hide instructions aimed at AI systems. This is prompt injection, and it is illegitimate. It also works sometimes.

A June 2026 study (arXiv:2606.16821) measured attack success rates from 0.0% on Claude-Sonnet-4.6 to 31.4% on Gemini-3-Flash. It used a mock search API and never published content to the open web. The figures are point estimates on one attack corpus. They shift with every safety update, so don't read them as fixed model traits.

What it means for you is defensive. Once a manipulated third-party page about your brand reaches the result set, it can distort what an agent says about you. Using injection yourself risks penalties and your reputation. The useful move is to watch the pages that mention your brand and flag hidden instructions.

Google documents eligibility and controls. It does not document how Gemini chooses among eligible pages, how often the app grounds, or which model version serves free users. As of the September 2026 snapshot, Gemini 3.6 Flash was selectable in the app, with no stated default.

Measurement is thin too. Search Console's generative AI report shows impressions only, with no clicks, CTR or position. Any "Gemini rank" from a third-party tool has no ground truth behind it. Score each engine separately rather than blending them, as one dashboard for all AI engines explains.

Start with the checkable gates: Google-Extended state, index status, snippet directives, the Search Console opt-out, and CDN behavior. Then test real buyer questions. The free checker on this site gives a snapshot of whether AI engines name you for three of them. Compare Gemini with how ChatGPT ranking factors and Perplexity visibility factors work.

Does blocking Google-Extended remove me from AI Overviews? No. Google says Google-Extended does not affect inclusion in Google Search and is not a ranking signal. It controls training and grounding for Gemini Apps and Vertex AI. AI Overviews and AI Mode follow Googlebot rules and the Search Console generative AI inclusion control. If you want out of AI Overviews specifically, robots.txt has no dedicated token. The Search Console control or snippet directives are the documented route.

Does Gemini need server-side rendering like ChatGPT does? Less so. Googlebot renders JavaScript, so content that only appears after scripts run can still reach Google surfaces. Crawlers from OpenAI, Anthropic and Perplexity don't execute JavaScript, according to Vercel's crawler study and later re-testing. If you want visibility across engines, putting key facts like pricing in the initial HTML still makes sense. For Gemini alone, rendering is rarely the blocker.

Will adding schema markup help Gemini cite my site? The evidence says no. Google states structured data isn't required for generative AI search. Ahrefs' matched test found AI Overview citations fell 4.6% after pages added schema. Those pages were already heavily cited, so the effect on uncited pages is untested rather than refuted. Keep existing markup, since Google says it does no harm. Don't expect it to earn citations.

Can I see my position in Gemini or AI Mode answers? No. As of October 2026, Google's Search Console report for generative AI features shows impressions only, with no position, clicks or CTR. Google also says third-party tools don't have access to its internal ranking data. Any tool showing a Gemini rank is producing its own model, not a measurement Google can confirm.

Why would Gemini get facts about my company wrong? Two common reasons. In retrieval-off answers, the model relies on training data frozen at a cutoff, so newer facts are missing. In grounded answers, it may pull from pages that state your facts inconsistently or out of date. Check your own site first: make sure pricing, product names and company details match on every page Google can index.