Dynamic Retrieval
Google's mechanism for deciding, per query, whether Gemini runs a live search — governed by a default grounding threshold of 0.3.
When you ask Google’s Gemini a question, it has to make a quick decision. Should it answer directly from its own trained memory, or should it run a live web search to get the most up-to-date information?
This decision-making process is called dynamic retrieval. Rather than searching the web for every single query, Gemini scores how much a question would benefit from fresh, live content. Google uses a default threshold score of point-three to make the call. If a query’s score falls below this threshold, Gemini answers from its static memory. But if the score clears that point-three hurdle, the system decides the query deserves grounding. Only then does it fetch and cite live web pages.
We confirmed this behavior in a live production environment and built a replica of this decision engine called the Grounding Classifier.
For anyone trying to optimize their content for AI search engines, understanding this mechanism is crucial. Dynamic retrieval is the ultimate gatekeeper. It is the single factor that decides whether a search query is even open to your online content in the first first place.
Dynamic retrieval is Google's mechanism for deciding, query by query, whether Gemini should run a live web search before answering or reply from its own trained knowledge. Rather than searching every time, the system scores how much a query would benefit from fresh web content and only grounds when that score clears a threshold.
Google's documented default threshold is 0.3: the point at which a query is judged to deserve grounding. Below it, the model answers from memory; above it, it fetches and cites live pages.
We confirmed this behaviour in a live production environment and built a replica of the decision in our Grounding Classifier. For anyone optimising for AI search, dynamic retrieval is the gate that decides whether a query is even open to your content.