Google AI Studio has introduced a new feature called "Search Grounding," which allows developers to ground AI model responses using real-time Google Search results. This feature is available in both Google AI Studio and the Gemini API, providing more accurate and up-to-date responses by pulling in relevant web search context.
The grounding feature helps reduce hallucinations (incorrect or fabricated information) and includes inline citations and links to sources, making AI-generated content more trustworthy. It also offers dynamic retrieval, where developers can control when grounding is activated based on a configurable threshold. This allows for fine-tuning the balance between response accuracy and system performance.
The feature can be accessed through the "Tools" section in AI Studio or via an API call, and it supports all Gemini 1.5 models. However, it comes with a cost of $35 per 1,000 grounded queries, which may influence how frequently developers choose to use it.
This update positions Google ahead of competitors by integrating native web search grounding directly into its API, simplifying the development process for web-aware AI applications. Despite some concerns about latency and costs, this feature is expected to enhance the reliability of AI outputs across various use cases.
