Coaley Peak GEO Research Centre Engine guide
Gemini How Google's assistant grounds its answers in Search, and how to be its source.
Gemini is Google's AI assistant, launched as Bard in 2023 and now used by more than a billion people a month. It answers from Google's Gemini models and grounds many answers in Google Search. This guide covers its timeline, the four layers behind its answers, the Google-Extended token and Google's fetchers, and how to opt out.
Independent guide: Coaley Peak is not associated with Google LLC. Trademarks
- Owner
- Google LLC
- Launched
- March 2023 as Bard; renamed Gemini February 2024
- Home market
- More than 230 countries and territories
- Languages
- More than 70
- Open weights
- No: Gemini is closed; Gemma is Google's open family
- Crawlers
- Googlebot, the Google-Extended token and user-triggered fetchers
Overview
What Gemini is, and who uses it
Google announced Bard in February 2023, opened it in the US and UK that March and in more than 180 countries in May. It introduced the Gemini family of models in December 2023, and on 8 February 2024 renamed Bard to Gemini, with a new mobile app. Gemini 2.0, built for agentic uses, followed in December 2024, Gemini 2.5 with built-in thinking in March 2025, and Gemini 3 in November 2025.
Google said in August 2026 that the Gemini app had passed a billion monthly users, up from 650 million the previous November. The web app is available in more than 230 countries and territories and over 70 languages. Gemini 4 Argon was announced at the end of September 2026, initially only for trusted cyber defenders.
Gemini differs from AI Overviews in a way that matters to site owners. Search Console's generative AI control covers AI Overviews, AI Mode and Discover only; for the Gemini app, the documented lever is Google-Extended, a robots.txt token that governs both training of future Gemini models and grounding of Gemini's answers.
That makes Gemini the engine where the choice between visibility and control is starkest. Blocking Google-Extended keeps your future content out of Gemini training and reduces its use in Gemini answers, while leaving Google Search untouched.
Timeline
Gemini, from launch to today
Google announces Bard, its experimental conversational AI service.
Bard opens to users in the US and UK.
Google removes Bard's waitlist and opens it in more than 180 countries and territories.
Google introduces the Google-Extended robots.txt control for Bard and Vertex AI.
Google introduces the Gemini 1.0 family of models.
Bard is renamed Gemini and launches as a mobile app, alongside Gemini Advanced.
Google introduces Gemini 2.0, aimed at agentic uses.
Google releases Gemini 2.5, with built-in thinking.
The UK CMA designates Google with strategic market status in search, but keeps the Gemini assistant out of scope, under review.
Gemini 3 launches in the Gemini app and, the same day, in Search's AI Mode.
Gemini 3.5 Flash rolls out to Gemini app users worldwide at Google I/O 2026.
Google says the Gemini app has passed a billion monthly users.
Google announces Gemini 4 Argon, initially released only to trusted cyber defenders.
Things to know
Things to know about Gemini
Blocking Google-Extended affects Gemini training and the grounding of Gemini Apps only. It does not remove a site from Google Search or change its ranking.
Search Console's generative AI control is documented for AI Overviews, AI Mode and Discover. For the Gemini app, the documented lever is Google-Extended.
Some Gemini answers are grounded in Google Search results, so classic indexability still matters for Gemini visibility.
Gemini shows a Sources button only when it provides links. Long quotations from a page are cited to that page.
Google-Extended has no user agent of its own: Googlebot does the crawling, and the token is read only as a usage control.
User-triggered Google fetchers, such as Gemini Notebook and Google-Agent, generally ignore robots.txt.
The NotebookLM-era fetcher token, Google-NotebookLM, has been replaced by Google-GeminiNotebook, and the old token was supported only until August 2026, so log filters need updating.
Recent Gemini Flash models have a published knowledge cut-off of March 2026, with some domains limited to January 2025, so fresher facts depend on Search grounding.
People can object to processing, or ask for inaccurate personal data in Gemini answers to be corrected, through Google's legal help centre or the in-app 'Report legal issue'.
The UK CMA kept the Gemini assistant outside Google's strategic market status designation in search in October 2025, but said it would keep this under review.
The four layers
How Gemini builds an answer
Every AI answer is made in four layers: what the model learned in pre-training, how it was tuned, what it retrieves live, and how it puts the answer together. Here is each one for Gemini, and what it means for a brand that wants to be cited.
Layer 1
Pre-training: the base (foundation) model
Gemini runs on Google's closed Gemini 3 family: the app offers Flash-Lite, Flash and Pro tiers, with Gemini 3.5 Flash and 3.6 Flash rolled out to all users in 2026. Published knowledge cut-offs are March 2026 for Gemini 3.6, 3.7 and 3.8 Flash (January 2025 in some domains) and January 2025 for Gemini 3 Pro. Training data, per Google's model card, includes public web documents, crawled data gathered in line with robots.txt, licensed data, user data under Google's controls and synthetic data. Gemini app chats improve Google's AI when Keep Activity is on.
What it means for you: Google-Extended decides whether your future content trains Gemini. Allowing it lets Gemini learn your brand directly; refusing it keeps future pages out of training, though earlier crawls may already be in the weights.
Layer 2
Post-training: tuning and alignment
Gemini 3 was post-trained with instruction tuning, reinforcement learning and human preference data, with reinforcement learning for multi-step reasoning. Human reviewers, including trained staff at service providers, review a subset of chats, which are kept for up to three years. Behaviour is governed by Google's Generative AI Prohibited Use Policy, the Gemini app's policy guidelines and Google DeepMind's Frontier Safety Framework.
What it means for you: Gemini's policies make it careful on health, finance and other sensitive topics. In those categories, being a source Google already trusts in Search counts for more than any wording on your page.
Layer 3
Retrieval: live search
Google says some Gemini responses are grounded in Google Search results, and Google-Extended governs whether crawled content may be used for that grounding in Gemini Apps and in Grounding with Google Search on Vertex AI. Gemini can also use connected apps such as Gmail and Drive, uploaded files and, with permission, Personal Intelligence. Gemini Notebook fetches the URLs users add, and agents on Google's infrastructure browse as Google-Agent.
What it means for you: To be used in Gemini's grounded answers, your pages need to be indexed in Google Search and allowed for Google-Extended. Blocking Google-Extended trades Gemini visibility for control over training.
Layer 4
The answer: what people read
When sources exist, a Sources button appears inline or under the answer and opens a panel of links, which can include public websites, uploaded files or connected Workspace content. If Gemini quotes a large amount of text from a page it lists that page, and web images show their source. No Sources button means no links were provided. Other features include Deep Research, Canvas, Gemini Live and skills.
What it means for you: Gemini does not always show sources, so being named in the answer matters as much as being linked. Pages with distinctive, quotable passages are the ones Gemini is most likely to quote, and therefore cite.
Crawlers
Gemini's crawlers and user agents
The user-agent tokens to look for in your server logs and to name in robots.txt.
| User-agent token | What it does | robots.txt |
|---|---|---|
| Google-Extended Documentation | Lets sites decide whether pages Google crawls may train future Gemini models and be used for grounding in Gemini Apps and Vertex AI. It has no user agent of its own and no effect on Google Search. | A robots.txt control token, read by Google's crawlers. |
| Googlebot Documentation | Builds the Google Search index that Gemini's Search grounding draws on. | Follows robots.txt. |
| Google-GeminiNotebook Documentation | Fetches the URLs Gemini Notebook users add as sources for their projects. It replaced the Google-NotebookLM token, which was supported until August 2026. | Generally ignores robots.txt, as a user-triggered fetcher. |
| Google-Agent Documentation | Used by agents on Google's infrastructure to navigate the web and act at a user's request; Google does not list the products. Some requests are signed with Web Bot Auth as https://agent.bot.goog. | Generally ignores robots.txt, as a user-triggered fetcher. |
| Google-CloudVertexBot Documentation | Crawls sites at their owners' request to build Vertex AI Agents. No effect on Google Search or other products. | Follows robots.txt. |
Because Google-Extended rides on Googlebot's crawl, you will not see it in your logs. You will see Googlebot, and Google applies your Google-Extended rule when it decides how the content may be used.
Crawl sources
Where Gemini crawls from, and how to verify it
Google publishes IP ranges for each class of crawler as JSON under developers.google.com/static/crawling/ipranges/: common-crawlers.json for Googlebot and Google-CloudVertexBot, user-triggered-fetchers-google.json and user-triggered-fetchers.json for fetchers such as Gemini Notebook, and user-triggered-agents.json for Google-Agent.
Verify with a reverse DNS lookup and then a forward lookup. Googlebot resolves to googlebot.com hostnames; Google-controlled user-triggered fetchers to google-proxy hostnames on google.com; and fetches from a user's own Google Cloud project to gae.googleusercontent.com.
Google-Agent is experimenting with Web Bot Auth, signing some requests as https://agent.bot.goog. Google advises still checking IPs, reverse DNS and user agents, because not every request is signed.
Opting out
How to opt out of Gemini, or limit what it uses
Refuse Gemini training and grounding
User-agent: Google-Extended with Disallow: / stops use of your content for training future Gemini models and for grounding in Gemini Apps and Vertex AI. It does not remove you from Google Search.
Official pageRemove pages from Search's index
noindex, or blocking Googlebot, removes pages from the Search index that grounding uses, but also from Google Search itself.
Official pageObject, or correct personal data
Under laws such as the GDPR, people can object, or ask for inaccurate personal data in Gemini answers to be corrected, through Google's legal help troubleshooter or, in the app, through More and then Report legal issue under an answer.
Official pageLegal removal requests
Ask for content to be removed under Google's policies or the law, for example for defamation or copyright.
Official pageKeep your own chats out of training
Turn off Keep Activity or use temporary chats, which are kept for 72 hours and not used to improve Google's AI unless you send feedback.
Official pageUser-triggered fetchers
No robots.txt opt-out is documented for Gemini Notebook or Google-Agent, which generally ignore robots.txt. Google publishes their user agents and IP ranges so you can identify them.
Official page
Google-Extended works going forward: Gemini models already trained keep what they learned. Because it also governs grounding, blocking it is likely to reduce how often Gemini uses your pages in its answers, so weigh that before you switch it on.
How we help
How we get brands cited in Gemini
The Coaley Peak GEO Research Centre researches and implements GEO for SEO agencies and in-house teams, with foreign markets as our speciality.
Gemini visibility audit
We ask Gemini the questions your customers ask, in each market and language, and record whether you are named, whether you are linked and what it gets wrong.
Google-Extended policy
We help you decide whether to allow Google-Extended, weighing training against grounding, and set it alongside your rules for every other AI crawler.
Search visibility that feeds grounding
Gemini's grounded answers draw on Google Search. We strengthen your indexing and rankings for the queries behind those answers, in every market you sell in.
Quotable content
We write passages Gemini can quote and cite: distinctive, specific and self-contained, so a quotation from your page carries your name with it.
Measurement across Google
We track your presence in Gemini alongside AI Overviews and AI Mode, so you see all of Google's AI surfaces together.
More on our research, services and team at the Coaley Peak GEO Research Centre.
FAQ
Gemini GEO: questions
Does Gemini use Google Search?
Yes, for some answers. Google says some Gemini responses are grounded in Google Search results, so being indexed and ranking well in Google Search matters for Gemini visibility.
What does Google-Extended do?
It is a robots.txt token that lets you refuse use of your content for training future Gemini models and for grounding in Gemini Apps and Vertex AI. It has no effect on Google Search, including AI Overviews and AI Mode.
Does the Search Console AI opt-out cover Gemini?
Its documented scope is AI Overviews, AI Mode and Discover's generative AI features. For the Gemini app, Google's documented control is Google-Extended.
Does Gemini cite its sources?
When it provides links, a Sources button appears with the answer and opens a panel of links. If Gemini quotes a large amount of text from a page, it lists that page. No Sources button means no links were provided.
What is Gemini's knowledge cut-off?
It depends on the model. Google publishes March 2026 for Gemini 3.6, 3.7 and 3.8 Flash, with some domains limited to January 2025, and January 2025 for Gemini 3 Pro.
Can I stop Gemini fetching my pages when a user asks?
Not with robots.txt: user-triggered fetchers such as Gemini Notebook and Google-Agent generally ignore it. Google publishes their user agents and IP ranges, so you can identify them, and block them at the network level if you must.
Want a free chat about Gemini?
Tell us which market and brand you are working on. If we can help, we will set up a free, no-obligation chat with Stephen or Ethan.
More engine guides
Related engines
Independence and trademarks. Gemini, Gemma and Google are trademarks of Google LLC. Coaley Peak is independent: we are not associated with, endorsed by or a partner of Google LLC. This guide is our own research from public documentation and reputable reporting, checked in October 2026.
Reviewed by Stephen Grindley, research centre lead. Last checked .
Document reference: ISO_webpage_geo_engine_gemini_v1
Last modified: 7 October 2026
GEO Research Centre·Engine guides·Gemini