Gemini SEO: the technical guide to Gemini citations

How Gemini grounds answers in Google Search#

Google defines grounding as “providing content from the Google Search index to the model at prompt time to improve factuality and relevancy” — in the Gemini app and in Grounding with Google Search on Vertex AI (Google’s crawler documentation). There’s no separate Gemini index to get into. If Googlebot hasn’t indexed a page, Gemini can’t ground on it.

How the consumer app decides when to search isn’t published. The Gemini API describes the pattern: the model “analyzes the prompt and determines if a Google Search can improve the answer,” then generates “one or multiple search queries” — the same fan-out Google uses in AI Mode.

1. 1

   Gemini decides whether Search would help

   Time-sensitive, specific and factual questions trigger grounding; many creative or conversational ones don’t. Google notes that “not all responses include related links or sources.”

2. 2

   It writes one or more search queries

   Like AI Mode, Gemini fans a question out into related searches. Ahrefs found over 80% of assistant citations don’t rank at all for the original prompt — they rank for the sub-queries.

   Your lever: Cover the sub-questions around your topic with depth, not one thin page per variant.

3. 3

   Google's ranking systems retrieve pages

   Candidates come from the Search index via Google’s core ranking systems — the same systems behind AI Overviews. Pages blocked from Google-Extended are excluded from Gemini-app grounding.

   Your lever: Be indexed, and leave Google-Extended allowed if you want Gemini citations.

4. 4

   Gemini answers and attaches sources

   Sources appear behind a “Sources” button or as inline links, opening a side panel. When Gemini quotes a large amount of text from a page, it always links to that page.

   Your lever: Write specific, quotable passages — they’re the ones that carry a guaranteed link.

5. 5

   Deep Research and agents go further

   Deep Research “can automatically browse up to hundreds of websites” and is powered by Gemini 3. Agents that browse from Google’s servers on a user’s behalf identify as Google-Agent; Gemini in Chrome’s auto browse works inside the user’s own browser.

   Your lever: Keep pages agent-friendly: real buttons and links, labeled forms, no overlays.

Google-Extended and the other controls#

This is where Gemini differs most from Google’s Search features, and where most advice is wrong. Google-Extended isn’t a crawler — “crawling is done with existing Google user agent strings” — it’s a robots.txt token. Google says it controls training of future Gemini models and “grounding in Gemini Apps and Grounding with Google Search on Vertex AI,” and that it “does not impact a site’s inclusion in Google Search nor is it used as a ranking signal.”

What each control switches off

So the two AI opt-outs are mirror images: the Search Console control (live worldwide since 31 August 2026) takes you out of AI Overviews and AI Mode but not Gemini; Google-Extended takes you out of Gemini but not AI Overviews. To stay in Search and its AI features while leaving Gemini:

# Search, AI Overviews and AI Mode
User-agent: Googlebot
Allow: /
 
# Opts out of Gemini-app grounding AND Gemini training.
# Leave this out if you want to be cited by Gemini.
User-agent: Google-Extended
Disallow: /

Technical requirements#

Because grounding rides on the Search index, Gemini’s technical bar is Google’s. Google’s AI optimization guide is written for AI Overviews and AI Mode, but nothing suggests the Gemini app differs on these points.

Indexed by GoogleRequired

Confirm in Search Console’s URL Inspection. There’s no other way into Gemini’s grounding pool.

Google-Extended allowedRequired

For Gemini-app grounding specifically. Check robots.txt on every subdomain, and any “block AI” presets your CDN adds.

Key content in the first 2 MBRequired

Googlebot fetches the first 2 MB of HTML; anything after is “not fetched… not rendered… not indexed.” Put head tags and main content early; avoid huge inline scripts and base64 images.

JavaScript renderingHelps

Google’s renderer runs JavaScript “similar to a modern browser,” but it’s stateless and slower than HTML. Server-render critical text, and don’t rely on local storage.

Merchant Center feedHelps

Gemini shops from the Shopping Graph with agentic checkout. UCP checkout needs Merchant Center and the native_commerce attribute (US, CA, AU; early access).

Google Business ProfileHelps

Google says Business Profile and Merchant Center “can help your products and services to be visible in… AI responses.”

Agent-friendly markupHelps

Real <button> and <a> elements, labeled fields, stable layouts, no invisible overlays — agents read the DOM and accessibility tree.

Structured dataNo effect

Google: “not required… no special schema.org markup” for AI features. Keep it valid and matching the page for rich results.

llms.txtNo effect

Google Search “doesn’t use them” and says they “neither harm nor help.” Nothing suggests the Gemini app differs.

Snippet controlsUnconfirmed

nosnippet and max-snippet are documented for AI Overviews and AI Mode only. Their effect on the Gemini app isn’t stated.

What Gemini cites#

Studies of Gemini’s sources disagree more than for any other engine — largely because they use different prompt sets. Read them together and a consistent picture emerges: Google’s quality systems pick the candidates, Gemini cites few of them, and the ones it cites lean either towards institutions or towards forums and video depending on the kind of question.

• Google's core ranking

  Official

  Grounding retrieves from the Search index via core ranking systems. Helpful, reliable, people-first content and E-E-A-T apply unchanged.

• Non-commodity content

  Official

  Google calls first-hand, unique content the biggest long-run factor for its AI features. Generic answers are what the model already knows.

• Forums and video

  Observed

  In Ahrefs’ 3M-query study, Reddit took 28.5% of Gemini’s top-50-domain citations, YouTube 14.6% and Wikipedia 8.8%.

• Institutions

  Observed

  BrightEdge found the opposite mix on its prompts: authority sources 26% of citations and user-generated content 0.2%. Health, finance and B2B queries lean this way.

• Few links, many mentions

  Observed

  Gemini cites about 3 sources per response against ChatGPT’s 15, and names brands in text far more than it links them. Being mentioned is half the game.

• Freshness and entities

  Our read

  Freshness weighting and Knowledge Graph strength plausibly drive Gemini’s unlinked brand mentions, but neither is documented or measured directly.

Tracking Gemini traffic#

The Gemini app doesn’t appear in Search Console at all — the new Generative AI report covers AI Overviews and AI Mode only. Gemini is measured in analytics. Since 13 May 2026, GA4’s default AI Assistant channel groups Gemini referrals automatically (Google names Gemini in the definition). Before that date, and as a backstop, look for gemini.google.com / referral — or bard.google.com in older data.

# Session source — Gemini app
^(gemini\.google\.com|bard\.google\.com)$
 
# All major AI assistants
gemini\.google\.com|chatgpt\.com|perplexity\.ai|copilot\.microsoft\.com|claude\.ai

• Some Gemini traffic loses its referrer — practitioners report it when Gemini runs as the Android assistant or inside Google apps — and lands in Direct or Organic.
• AI Overviews and AI Mode clicks are not in this channel. They arrive from google.com as Organic Search; see the AI Overviews guide for those.
• Mentions need their own tracking. With links on barely a fifth of brand appearances, a prompt panel that records unlinked mentions shows far more of your Gemini visibility than analytics can.
• In logs, watch for Google-Agent and Google-GeminiNotebook and verify them against Google’s new IP-range files.

Myths worth dropping#

The action checklist#

Everything above, in the order we’d do it. Tick items off as you go — your progress is saved in this browser.

Gemini SEO: frequently asked questions#