Grounding means an AI system bases its answer on information it retrieves at that moment. Google describes Grounding with Google Search as connecting the Gemini model to current web content. According to Google, this can reduce hallucinations and lets the model cite verifiable sources, including on topics from after its knowledge cutoff.
In its guide for websites, Google treats grounding as another name for retrieval-augmented generation.
Why does it matter?
With grounding, the system uses the pages it finds as its basis. That is the moment your website can serve as a source, provided the page can be found and read.
How do you control it?
For some Google products, you control grounding with the robots.txt token Google-Extended. According to Google’s crawler overview, it covers training of Gemini models and grounding in products including the Gemini apps. Google Search is not covered.
User-agent: Google-Extended
Disallow: /Common mistake
Treating grounding as a guarantee of truth. A grounded answer can still summarise a source incorrectly, or rely on a source that is itself wrong. See hallucination.
More in How do AI assistants decide which websites to mention?.