Start with Google's own words.
Per the crawler documentation, "Google-Extended is a standalone product token that web publishers can use to manage whether content Google crawls from their sites may be used for training future generations of Gemini models that power Gemini Apps and Vertex AI API for Gemini and for grounding (providing content from the Google Search index to the model at prompt time to improve factuality and relevancy) in Gemini Apps and Grounding with Google Search on Vertex AI."
Two things follow, and they are the practical answer for publishers.
First, Google-Extended really does control something here. It gates whether your crawled content trains future Gemini models.
It also gates whether that content can be used to ground a live answer in the Gemini app.
That is a genuine, documented lever.
Second, and this is where it differs from everything on our Search hubs. Google-Extended does not gate whether you appear in an AI Overview or an AI Mode answer.
Those run on ordinary Search indexing.
So a publisher can disallow Google-Extended and keep full Search and AI Overviews eligibility. They still stop their content training Gemini or grounding its chat answers.
One independent measurement complicates that second point without contradicting the documentation. A SIGIR 2026 study found sites disallowing Google-Extended were significantly less likely to be cited in AI Overviews.
That is despite Google's stated access. We cover the study and its limits on the AI Overviews hub.
A second measurement speaks to what the grounding pool looks like rather than to who is in it.
An August 2026 detector study estimates that 8.90% of the pages behind these two channels are optimized for generative engines. The share is higher for pages Gemini's grounded answers cite than for pages a conventional results page returns.
It does not change what the token controls. It is context for the decision.
The content competing for a grounded citation is increasingly written for that purpose.
Eligibility and citation are different things, and the paper measures the second. It rests on one December 2025 snapshot and reports an association rather than a demonstrated cause.
So treat it as a reason to check your own citation data after blocking.
A third measurement speaks to the cost of blocking rather than to citation, and it comes from outside Google entirely.
A Wharton and Rutgers working paper on the 2023 blocking wave studied large publishers who disallowed GenAI crawlers. They lost about 7% of weekly traffic within six weeks.
That is an aggregate effect across every major AI token rather than a Google-Extended finding. Its window stops before AI Overviews launched.
So it cannot tell you what this one token costs. It does say the blocking decision has shown up in traffic data before.
That is a reason to measure your own rather than assume the cost is zero.
Google is explicit that the token costs you nothing in Search: it "does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search".
It also runs no crawler of its own.
As the documentation puts it, "Google-Extended doesn't have a separate HTTP request user agent string. Crawling is done with existing Google user agent strings; the robots.txt user-agent token is used in a control capacity."
What it cannot do is remove you from a conversation entirely. It restricts Google's use of its own crawled copy.
So it does not govern what happens when a user pastes your URL into a chat directly.
No documentation describes a separate mechanism for that case.
Whether Gemini links out, and how much traffic it sends
Gemini does cite, but not always. Google's help page says "When sources are available, you can find the Sources button at the bottom of the response or in-line throughout the response", and equally that "Not all responses include related links or sources."
One case is guaranteed: "If Gemini Apps directly quotes a large amount of text from a webpage, you'll see a link to that webpage in the sources list."
One distinction in that documentation is easy to miss. Google also surfaces "related links", which it says are "not necessarily sources used to generate the response".
A link sitting next to an answer is not proof Gemini used that page for the claims it made.
On volume, Statcounter is the verified backbone. Gemini reached 8.65% of AI assistant referrals to websites in March 2026 and 9% in April, second behind ChatGPT.
Note that this is share of a pool rather than absolute traffic. It measures referrals out to sites rather than usage of the assistant itself.
Figures we deliberately do not print. A set of Gemini referral numbers circulates widely in SEO coverage, attributed to a panel of 101,000 sites.
The figures are +115% in two months and 29% more visitors than Perplexity.
No fetchable primary report exists for them, so they stay off this page until one does.
