Query Deserves Grounding
Google's internal decision about whether a query needs a live web search; below its threshold, the model answers from memory.
When you search using an AI system, it has to make a crucial decision. Should it answer your question from its own trained memory, or does it need to search the live web for fresh information? This process is called Query Deserves Grounding.
In Google's system, this decision is controlled by a dynamic threshold. If a query scores high enough, the AI decides it needs grounding and triggers a live web search. If the score is too low, the AI simply answers from what it already knows.
This distinction is incredibly important for anyone trying to be found online. Your website's content can only be cited if the AI decides a query needs grounding and searches the live web. If the AI relies purely on its memory, it will only use information it absorbed during its initial training.
By analyzing thousands of real AI responses, researchers have reverse-engineered this behavior. It turns out the system is designed to keep things efficient. The AI answers quickly and cheaply from memory whenever it can, reserving the power of live web retrieval for questions that genuinely require up-to-date facts.
Query Deserves Grounding (QDG) is the decision an AI search system makes about whether a given query needs a live web search or can be answered from the model's trained-in knowledge. Google's documented default dynamic retrieval threshold for this is 0.3 — score above it and the query gets grounded, below it and the model answers from memory.
The distinction is decisive for AI visibility. Only grounded queries pull in live web content, so only grounded queries are ones your pages can be cited in. A query answered purely from parametric memory reflects whatever the model already absorbed during training.
We reverse-engineered this behaviour in our Grounding Classifier, trained on 10,000 real Gemini responses, and demonstrate it in our QDG tool. It's the same idea Google applies internally to keep fast, confident answers cheap and reserve web retrieval for questions that genuinely need fresh facts.