AI Visibility Academy · Chapter 08
Gemini: how it finds and cites web pages
Gemini, Google's AI assistant, answers from what its model learned in training and, when it helps, from content in the Google Search index fetched at the moment you ask. Google calls the second part grounding, and describes it as providing content from the Google Search index to the model at prompt time. For a site owner, that means the pages Gemini can draw on live are pages Google has indexed. Google doesn't document any separate way to submit pages to Gemini.
How sources appear
When Gemini uses outside material, a Sources button appears under the response or inline, and opens a panel of links that can include public websites. Google says not every response includes sources, and that when Gemini quotes at length from a web page, that page is listed. It also warns that Gemini can present inaccurate information as fact, so check what it says about your company rather than assume it read your site. For longer tasks, Deep Research can browse up to hundreds of websites to write a report.
The setting that is Gemini's own
As of October 2026, Google runs two separate switches, and they don't overlap the way many site owners assume:
| Control | Search, AI Overviews, AI Mode | Gemini app |
|---|---|---|
| Google-Extended in robots.txt | No effect on inclusion | Covers training and grounding |
| Search Console generative AI setting | Can exclude your site | Not listed |
Blocking Google-Extended keeps your pages out of grounding in the Gemini app and out of training for future Gemini models, while your Search visibility stays as it is. The Search Console control lists AI Overviews, AI Mode and Discover, not Gemini. Decide each one on purpose. The lesson on search crawlers vs training crawlers walks through the trade-off.
What isn't documented
Google doesn't say how Gemini decides when to search, which results it reads, or why it cites one page over another. Apps built on the Gemini API can use Grounding with Google Search, where the model writes its own queries and returns an answer with citations, so similar mechanics may sit behind tools other than the Gemini app. And when someone adds your page as a source in Gemini Notebook, Google fetches it with a user-triggered fetcher that, like the others in its class, generally ignores robots.txt.
This week
Open your robots.txt and check whether Google-Extended is blocked, and whether that was a decision or a leftover from an old template. Then ask Gemini three buying questions, open Sources on each answer, and note which pages it links.
Sources
- Google, "Google's common crawlers", updated July 14, 2026 (opens in a new tab)
- Gemini Apps Help, "View related sources from Gemini Apps", accessed October 8, 2026 (opens in a new tab)
- Gemini Apps Help, "Learn about responses from Gemini Apps", accessed October 8, 2026 (opens in a new tab)
- Google, "Gemini Deep Research", accessed October 8, 2026 (opens in a new tab)
- Search Console Help, "Search generative AI control", accessed October 8, 2026 (opens in a new tab)
- Google AI for Developers, "Grounding with Google Search", updated September 23, 2026 (opens in a new tab)
- Google, "Google's user-triggered fetchers", updated August 19, 2026 (opens in a new tab)