Live grounded retrieval augments language model generation with real-time web access through tools like Google Search or URL context fetching. By parsing the resulting grounding metadata into reference snippets, domain matches, and cited URLs, it anchors model responses to factual online sources rather than internal parametric memory.