Key takeaways
Google-Extended switches off Gemini training and Gemini app citations together, and it changes nothing in Google Search.
| What you want | What to do |
|---|---|
| Stay in Google Search, AI Overviews and AI Mode | Keep Googlebot open; Google-Extended plays no part |
| Get cited in Gemini app answers | Allow Google-Extended |
| Keep paid content out of Gemini training | Disallow Google-Extended on those paths only |
| Leave AI Overviews and AI Mode | Use Search Console's Search generative AI setting |
| Opt out of AI training at every lab from 2027 | Disallow __usage_ai-training, and accept losing Gemini citations |
Most B2B sites should allow it by name, then open their live robots.txt today, since a CDN setting can add a block without marketing knowing.
Google-Extended is a robots.txt token that decides how Google's Gemini models may use your website. Block it and Google stops training future Gemini models on your pages, and the Gemini app stops citing them in answers. Your Google rankings stay put. So do AI Overviews and AI Mode. Google spells this out in its crawler documentation, last updated on 30 October 2026.
Here at Searchmaxxed, we're an SEO and AI SEO agency for law firms, clinics and premium builders. We make your company the answer in Google and in the AI engines your buyers ask: ChatGPT, Claude, Perplexity, Gemini and AI Overviews. Our advice for nearly all of them fits in one line: leave Google-Extended open, and name it in robots.txt before the 2027 training token arrives.
What is Google-Extended?
Google-Extended is a standalone product token, Google's term for a name in robots.txt that tells it how it may use the pages it crawls. It has no crawler of its own. Google says it "doesn't have a separate HTTP request user agent string". Its ordinary crawlers fetch your pages. The token then sets what Gemini may do with them.
Google announced it on 28 September 2023, when its chatbot was still called Bard, to cover Bard and the Vertex AI generative APIs. Then Bard became Gemini. In early 2024 Google rewrote the description and added the line every SEO team wanted: the token doesn't affect Google Search.
Today it covers two jobs. The first is training, meaning whether Google may use your pages for "future generations of Gemini models". The second is grounding.
Grounding is when a model pulls pages from Google's Search index while it writes an answer, then cites the pages it relied on. Google-Extended governs grounding in the Gemini app and in Grounding with Google Search on Vertex AI. Other companies build that second one into their own apps.
What does Google-Extended control, and what doesn't it?
It controls training for Google's Gemini models and grounding in the Gemini app and Vertex AI. Your Google rankings, your place in AI Overviews and AI Mode, and the pages Google crawls all sit behind other switches.
| Google product or use | Does Google-Extended affect it? | The control that does |
|---|---|---|
| Training future Gemini models for the Gemini app and Vertex AI | Yes | Google-Extended |
| Training the models behind AI Overviews and AI Mode | Yes | Google-Extended |
| Citations and links in Gemini app answers | Yes | Google-Extended |
| Grounding with Google Search on Vertex AI | Yes | Google-Extended |
| Google Search indexing and rankings | No | Googlebot rules in robots.txt, or noindex |
| Showing up in AI Overviews, AI Mode and Discover's AI features | No | Search Console's Search generative AI setting, or nosnippet |
| Which pages Google fetches at all | No | Googlebot rules in robots.txt |
The second row surprises people. Search Console's help page says its generative AI control "doesn't affect AI training", and points to Google-Extended for the models behind AI Overviews. So the token reaches into Search's AI models. It still never decides whether Search shows your pages.
Picture three separate switches. Googlebot decides whether Google can crawl and rank you, and the Search Console setting decides whether AI Overviews and AI Mode can show you. Google-Extended decides what Gemini may learn from you and whether the Gemini app can cite you.
For the browser experience rather than crawler controls, see how Gemini works in Chrome.
Does blocking Google-Extended hurt your SEO or remove you from AI Overviews?
No. Google says Google-Extended "does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search". AI Overviews and AI Mode are part of Search, so the token can't take you out of them.
That cuts both ways. Some teams block Google-Extended to leave AI Overviews and watch their pages keep appearing, because the working control lives in Search Console instead. Look under Settings, then Search generative AI. Google rolled it out to every site worldwide on 31 August 2026. Choose "exclude" and Google says you'll get no traffic or impressions from AI Overviews, AI Mode or Discover's AI features, usually within a few days.
The older options still exist. Google's guide to AI features lists nosnippet, data-nosnippet, max-snippet and noindex, though those also trim or remove your normal search snippets. For most B2B companies, leaving AI Overviews costs more than it saves. Our guide to ranking in AI Overviews covers the opposite goal.
What do you lose when you block Google-Extended?
You lose your place in Gemini's answers. Google's 30 October 2026 post says sites that allow Google-Extended "are eligible to appear in responses and receive clicks via grounding in Gemini Apps". Block it, and the Gemini app stops citing your pages, along with apps built on Grounding with Google Search on Vertex AI.
Google gives you one switch for two jobs.
You can't tell Google "no training, but cite me in Gemini", because both sit behind one token. A training opt-out is a Gemini opt-out. Microsoft has gone the other way. From 1 February 2027, Bing will read separate preferences for training and for grounding.
| AI company | How you control training | How you control appearing in AI answers |
|---|---|---|
| Google-Extended, plus the shared training token from early 2027 | Google-Extended for Gemini; Search Console for AI Overviews and AI Mode | |
| Microsoft Bing | Shared training token in robots.txt, plus page-level settings, from 1 February 2027 | A separate page-level AI Grounding setting |
| OpenAI | GPTBot, plus the shared token in the coming months | OAI-SearchBot for ChatGPT search |
The OpenAI row comes from its crawler documentation and its 30 October announcement. And a block can't reach backwards. Google's wording covers "future generations" of Gemini models, so models trained before your block keep whatever they learned from your pages.
Should your company block Google-Extended?
For most B2B SaaS, cybersecurity and enterprise sites, no. Your public pages exist so buyers can find you, understand you and shortlist you, and Gemini is one of the places they ask. Block it only where the content itself is what you sell.
| Your content | Google-Extended | Why |
|---|---|---|
| Product, pricing, comparison and solution pages | Allow | These are the pages you want Gemini to quote |
| Public docs and help centre | Allow | Gemini answers product questions from them |
| Threat research, free reports and the blog | Allow | Citations put your name next to your category |
| Paid reports, benchmark data and licensed content | Disallow on that path | The content is your product |
| Customer portals and logged-in areas | Put them behind a login | robots.txt is a request, so it can't lock anything |
Take a cybersecurity company that publishes free threat research and sells an annual benchmark report. We'd allow Google-Extended everywhere except /benchmark/, and keep the full report behind a login. Gemini keeps citing the research. The paid data stays out of training at Google.
Our guide on whether to block AI crawlers makes the same call for every AI bot, from GPTBot and ClaudeBot to PerplexityBot.
How many B2B companies block Google-Extended?
Almost none. On 1 November 2026 we read the live robots.txt files of 70 well-known B2B SaaS and cybersecurity companies, and 69 returned a readable file. Just one blocks Google-Extended across its whole site.
| What the robots.txt file does | Sites (of 69) |
|---|---|
| Blocks Google-Extended across the whole site | 1 |
| Names Google-Extended in its own group and allows it | 13 |
| Never mentions Google-Extended | 55 |
| States "ai-train=no" in a Content-Signal line while Google-Extended stays open | 5 |
| Blocks GPTBot across all or most of the site | 2 |
Five sites look as if they opted out of AI training, just not at Google. Each added a Content-Signal line saying "ai-train=no". Google's robots.txt specification supports four fields: user-agent, allow, disallow and sitemap. Content-Signal isn't one of them. Under the rules Google reads, all five leave Google-Extended open.
And 55 sites never name the token at all. That's fine today, but in 2027 one new line from a legal team could quietly take any of them out of Gemini's answers.
How do you block or allow Google-Extended in robots.txt?
To block Gemini training, add a group that starts with User-agent: Google-Extended and give it a Disallow rule. To allow it, use Allow instead. Put it in the robots.txt file at the root of every host that matters, subdomains included.
To keep your whole site out of Gemini training and Gemini answers:
User-agent: Google-Extended
Disallow: /To protect one paid section and leave everything else open, which suits most companies with gated research:
User-agent: Google-Extended
Disallow: /benchmark/To allow it by name, our default for B2B sites:
User-agent: Google-Extended
Allow: /Four details trip up experienced teams:
- Specific groups win. Google follows only the most specific group that names a crawler and ignores the rest,
User-agent: *included. Repeat any shared Disallow rules inside the Google-Extended group. - Unnamed means default. Leave Google-Extended out and your
*group sets its rules. - Every host has its own file. A rule on www.example.com does nothing for docs.example.com.
- Changes take a day. Google generally caches robots.txt for up to 24 hours.
Our own robots.txt shows the pattern. It names Google-Extended in its own group and allows it, next to GPTBot, ClaudeBot and the other training bots.
What changes in 2027 with the shared AI training token?
From early 2027, Google plans to honour one shared robots.txt token for AI training, written as User-agent: __usage_ai-training. Disallow it and you tell every participating AI company not to train on your content. Google will read that as a Google-Extended block too, unless your file gives Google-Extended a separate group with its own rules.
Google announced the plan on 30 October 2026. The same day, it added a note to its crawler page promising support "in early 2027", and OpenAI and Microsoft made matching commitments. The token comes from an IETF draft built on the AI Preferences working group's work. The IETF's page says the draft has "no formal standing" yet. Its wording can still change.
It's easy to get wrong. Your legal team hears that one line now stops AI training everywhere, and someone adds it. Months later, sales notices Gemini has stopped citing your product pages. Nobody connects the two.
To leave AI training at every lab, Gemini citations included:
User-agent: __usage_ai-training
Disallow: /To leave training at other labs but keep Gemini citing you, name Google-Extended as well. Google may then train on your pages, since training and Gemini citations share one token:
User-agent: __usage_ai-training
Disallow: /
User-agent: Google-Extended
Allow: /Make the decision once. Write it down, and put Google-Extended in the file by name either way.
Can you see Google-Extended in your server logs?
No. Google-Extended sends no requests of its own, so no log line, firewall event or analytics report will ever carry its name. Your robots.txt file is the one place it exists.
Some bot-management guides suggest a firewall rule matching "Google-Extended" in the user agent, but no request carries that string, so it never fires. Blocking Googlebot instead goes too far, because it takes you out of Google Search.
The bigger risk is a block nobody chose. Cloudflare's robots.txt setting adds User-agent: Google-Extended and Disallow: / to your file when Bot Preference Sync runs with Training on Disallow. A security team can flip that in a dashboard. Check the live file at yourdomain.com/robots.txt rather than the copy in your code. Then check every subdomain after any CDN or security change.
Once the token is open, our guide to ranking in Gemini covers what earns the citation.
How does Searchmaxxed handle Google-Extended for clients?
We make your robots.txt, CDN and Search Console settings match the choice your company made, so you stay visible wherever you want to be. That's Google Search, AI Overviews and Gemini. Only protected content stays out. It's part of our technical SEO work, next to crawling, indexing and site-speed fixes.
We check the live robots.txt on every host, the bot rules at your CDN and the Search Console generative AI setting. We set your robots.txt rules on purpose, including how to treat the 2027 shared token. Then we test real buyer prompts in Gemini to confirm the pages you want cited are the ones it cites.
Crawler settings are only the floor. Getting named takes answer-first pages and other sites that vouch for you, and our guide on how to show up in AI search covers both.
Is Searchmaxxed right for you?
We're a strong fit for marketing, growth and SEO leads at law firms, clinics and premium builders. Maybe you're losing ground in Google, or missing from AI answers. You want buyers to hear your name when they ask Gemini, ChatGPT, Perplexity or Google who to choose. And you want to own everything we build.
One client came to us with a website that wasn't working. Three months later he was on page one for one of the most competitive searches in his industry. Now he has more high-intent leads than he can handle, and we're building the systems to handle them.
The bigger picture:
Searchmaxxed's work has generated more than $15 million in revenue through search for its clients.
"I can't believe this worked." That's what we hear. Clients tell us they finally found a growth partner in search that keeps its word and gets things done.
We're the wrong call in a few cases. Need one robots.txt line changed? Your developer can paste it from this page this afternoon. If you sell content and want every AI use blocked, a lawyer and a login wall will serve you better than an SEO team. And if you need leads this month just to survive, search takes longer than that.
What should you do next?
This check takes about ten minutes:
- Open yourdomain.com/robots.txt and search for "Google-Extended".
- In Search Console, open Settings, then Search generative AI, and confirm "include" is selected.
- Ask your security team whether a CDN setting blocks AI training.
Want a team to check all of it with you? Get your free search analysis. We'll benchmark you against your top competitors across Google, ChatGPT and the other AI engines. We'll show you where they're named instead of you, and the few moves that will matter most.
Frequently asked questions
Is Google-Extended a crawler?
No. It's a robots.txt token with no user agent of its own. Google's regular crawlers fetch your pages, and Google-Extended only sets how Gemini may use them, which is why it never appears in server logs.
Does blocking Google-Extended stop AI Overviews?
No. AI Overviews and AI Mode belong to Google Search, which Google-Extended doesn't touch. To leave them, use Search Console's Search generative AI setting, or the nosnippet and noindex controls.
Does Google-Extended affect Google rankings?
No. Google says the token doesn't affect a site's inclusion in Google Search, and Google doesn't use it as a ranking signal. Blocking it changes what Gemini does, and nothing in your organic results.
How do I stop Gemini training on my personal chats?
That's a different setting, because Google-Extended covers website content only. For your own Gemini conversations, Google lets you turn off Keep Activity and delete past activity in your Gemini Apps Activity settings.
Does blocking Google-Extended remove content Gemini already learned?
No. Google's wording covers future generations of Gemini models, so a block stops new training only. Models trained before the block keep what they learned.
Should I name Google-Extended in robots.txt if I allow everything?
Yes. A named group with Allow: / keeps you in Gemini answers even if someone later adds the 2027 shared training opt-out. Without it, Google plans to read that opt-out as a Google-Extended block.
How long does a Google-Extended change take to work?
Expect at least a day. Google generally caches robots.txt for up to 24 hours, and it doesn't publish how long Gemini takes to reflect the change after that.
Sources
About the editorial team
Founder, Searchmaxxed