How it discovers and cites pages
The Gemini app doesn't search the web on every prompt. Its grounding with Google Search works per-prompt: the model first decides whether a live search would improve the answer, and only then retrieves results and attaches sources. When a response is grounded, a Sources button or inline links expose the pages it drew from; when it answers from its training data alone, no sources appear at all. Retrieval runs on Google's search index — the same index behind AI Overviews and AI Mode — so being indexed and rankable in Google Search is the prerequisite. One control-token wrinkle: Google-Extended in robots.txt governs whether your content may be used for Gemini training and grounding, but it does not affect AI Overviews or AI Mode, which follow normal Googlebot and snippet controls.
What this means for getting cited
To be cited by Google Gemini, your page first has to be reachable through the index it draws on (Google Search, via per-prompt 'grounding' (the model decides when to search)) and crawlable by the relevant bots, then it has to answer the user's question clearly enough to be quoted. I test exactly these variables at HZ and publish what moves the needle.