Google uses the name Gemini Apps for the web app of its AI assistant Gemini, the mobile apps, and other access points such as Gemini in Chrome and in Google Maps. When someone asks Gemini about providers or products, or asks for recommendations, Gemini writes the answer itself. Whether and how your business appears in those answers is part of your AI visibility. The most direct way to see it is to ask Gemini the questions that matter to your customers and compare the answers across several queries.

Google launched the assistant as an experiment called Bard in March 2023 and made it available in Germany, including in German, in July 2023. Google introduced the Gemini model family in December 2023, and the assistant took on its name in February 2024. Gemini models also power other Google products, such as AI Overviews and AI Mode in Google Search.

How Gemini answers

The Gemini app can answer a question from what its models learned in training. It can also base an answer on content it retrieves at the moment the question is asked. Google calls this grounding: the model receives content from Google’s web index to make the answer more factual and relevant.

Answers can include links to sources, such as public websites. According to Google, when Gemini quotes at length directly from a web page, it links that page in the list of sources. Such a link is an AI citation; when Gemini names a business without a link, that is a mention. Grounding is meant to reduce false statements but cannot rule them out: Google itself notes that Gemini can hallucinate and present inaccurate information as fact.

Google-Extended: one rule for training and grounding

For Gemini, a website can set a rule in its robots.txt file for Google-Extended, a name with no crawler of its own behind it that, according to Google, governs both whether crawled content may be used to train future Gemini models and whether Gemini Apps may use it for grounding. Blocking Google-Extended to keep content out of training therefore also keeps it out of grounding in Gemini Apps; whether a website appears in AI Overviews or AI Mode, according to Google, is not decided by this rule.

Gemini in other companies’ applications

Gemini models also run in applications built by other companies, such as chatbots or AI agents that carry out tasks on their own. These companies connect the models through the Gemini API, Google’s programming interface for developers, or through Google Cloud’s Gemini Enterprise Agent Platform, which grew out of Vertex AI in April 2026. Content from your website can also flow into answers through such applications. For the Gemini API, Google documents, among others, two tools that let a model draw on web content:

  • Grounding with Google Search: The model checks whether looking something up on the web can improve the answer, writes one or more queries itself, and returns the answer with source references. According to Google, Google-Extended applies to this tool on Gemini Enterprise Agent Platform.
  • URL context: The model reads up to 20 web addresses passed along with a request. It first checks an internal cache and otherwise fetches the page live. Pages must be publicly accessible; paywalled content is not supported.