Three Google surfaces, two sets of rules
Gemini is the name of Google’s models and of its assistant app, and the two are easy to conflate. For a site owner the distinction that matters is which rules govern access.
| Surface | What it is | What controls access |
|---|---|---|
| AI Overviews | Generative summary in Search results | Googlebot robots.txt rules; preview controls |
| AI Mode | Conversational Search, “a custom version of our advanced Gemini models” | Googlebot robots.txt rules; preview controls |
| Gemini app | Google’s assistant, “some responses are grounded on Search results” | Google-Extended for training and grounding |
DOCUMENTED Google’s site-owner documentation says that because AI is integral to Search, robots.txt directives for Googlebot are the control for AI Overviews and AI Mode, and points to Google-Extended “to limit AI training and grounding in some of Google’s other systems”.
What Google-Extended is
DOCUMENTED 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.” It 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.”
DOCUMENTED And the line that settles the most common worry: “Google-Extended does not impact a site’s inclusion in Google Search nor is it used as a ranking signal in Google Search.”
The practical reading: a site that disallows Google-Extended keeps its Search presence, including AI Overviews and AI Mode, and gives up being used to ground Gemini app answers and to train Gemini models. A site that wants to appear in Gemini app answers allows it. Blocking it “to be safe” is a decision to be absent from one surface, and should be made on purpose.
How Gemini grounds answers
DOCUMENTED Google’s Gemini privacy documentation states that Gemini Apps rely on other Google services and that “some Gemini responses are grounded on Search results”. Google’s definition of grounding, from the crawler documentation, is “providing content from the Google Search index to the model at prompt time to improve factuality and relevancy”.
INFERRED Two consequences follow for a Perth business. The Search index is the substrate, so a page that is not indexed by Google is not available for grounding. And a grounded Gemini answer is composed from Search results at prompt time, so the same consistency and retrievability that matter for AI Overviews matter here. Which Search results Gemini grounds on, and how it chooses among them, is not documented.
What a Perth business can check today
- Google-Extended. Read robots.txt for a
User-agent: Google-Extendedgroup. If it disallows, that is a choice to be out of Gemini grounding; confirm it was deliberate. The crawler access guide covers the check. - Indexing. Confirm the pages that should answer Perth questions are indexed in Search Console; unindexed pages cannot be grounded on.
- Consistency. Ask the Gemini app what the business does, where it is and who runs it, in fresh sessions, more than once. Hedging or confusion is an entity problem, not a Gemini one.
How Gemini applies in Perth
HYPOTHESIS The planned Gemini Perth study runs the same Perth recommendation prompts on the Gemini app and on Google Search on the same days and counts the overlap in businesses and domains. The hypothesis is that grounded Gemini answers draw on a similar source set to AI Overviews for local prompts, with differences in how many businesses are named. That is a hypothesis to count, not a finding.
What is documented and what is inferred
- Documented: grounding on Search results; Google-Extended’s scope and its non-effect on Search; AI Mode’s use of Gemini models and its Australian launch.
- Inferred: indexing as a precondition for grounding; consistency carrying over.
- Hypothesis: the Gemini versus AI Overviews source overlap for Perth prompts.
- Undisclosed: how the Gemini app selects and cites grounded sources.