Gemini

Gemini is the family of generative models made by Google and the assistant built on them. The one name covers the models developers call through Google’s APIs and the consumer app and web assistant a person talks to.

How a website reaches Gemini

Two routes, and Google gives site owners one switch for both. Google’s crawler documentation describes the Google-Extended product token, which controls whether a site’s content is used to train future Gemini models and whether it is supplied to the model as grounding, meaning content pulled from the Search index at the moment a question is asked. Disallowing the token in robots.txt withdraws the site from both. The same page says the token has no user agent of its own, because the crawling is done by Google’s existing crawlers, and that it is not a ranking signal and does not affect whether the site appears in Google Search.

So a business does not get into Gemini’s answers by doing anything Gemini-specific. It gets there by being in Google’s index with a page that says, in one place, what the assistant is being asked. Grounding takes a passage, and the passage has to exist.

What has been measured here

Nothing on Gemini yet. This site’s one large probe put one buyer question per business to Claude with retrieval; the figure it produced is a Claude figure and is not repeated here as if it applied to Gemini. Its visibility panel asks one assistant, Claude, so it produces no Gemini number either; the methodology page describes how each answer is scored. Until Gemini has been measured, this entry states the mechanism and stops.

A common question

Does a business have to do anything Gemini-specific to appear in its answers?

No. Google gives site owners a single token, Google-Extended, and disallowing it in robots.txt withdraws the site from training and from grounding at once; there is nothing to add on the other side of that switch. A business reaches those answers by sitting in Google's index with a page that states, in one place, whatever the assistant is being asked. Grounding pulls a passage out of indexed content at the moment of the question, so the passage has to exist before it can be taken.

Sources

Google-Extended, training and grounding: Google Search Central, common crawlers. The limit stated above is this site’s own: its published probe used one assistant, and it was not Gemini.