Google debuted 'Grounding with Google Search' on AI Studio and the Gemini API
Google has launched 'Grounded with Google Search', a paid feature that allows Gemini to access current search results for more accurate responses with citations and search suggestions. Grounding features a dynamic retrieval configuration to optimize when the feature is used.
In addition to announcing a major expansion that makes AI Overviews available in over 100 locations, Google recently launched 'Grounding with Google Search', a feature that enables developers to get more accurate, up-to-date responses from Gemini by giving it access to Google Search. With 'Grounding with Google Search', not only does Gemini increase its accuracy when answering queries that go beyond the models' knowledge cut-off date, but it can also provide in-line citations to the search results from which it extracted the information.
The feature also allows Gemini to display 'Search Suggestions'—links that take users to the relevant search results pages. According to Google, these measures reduce hallucinations by enabling the models to access more factual information, making the AI-powered assistant useful for more real-world scenarios. They also increase transparency and drive traffic to websites because users can see where the information displayed by Gemini originated, and quickly access these sources by clicking the links in the in-line citations.
While this new feature is free to try on the Google AI Studio, API access requires a paid tier and costs $35 for 1000 queries. Since it will generally be the case that not all queries will require grounding, Grounding with Google Search includes a dynamic retrieval configuration. With dynamic retrieval, prompts are given a score between 0 and 1, depending on the likelihood that they'll benefit from grounding. The default configuration sets 0.3 as a threshold, but developers can modify this value to adjust to their application's demands and preferences.