Web Grounding

Web grounding is when a language model backs its answer with live search results instead of relying only on training data: the model runs a search, retrieves pages or snippets, and composes a cited answer. Grounded answers are what GEO measures, since they determine which sites get cited today. Engines ground against different indexes — ChatGPT search leans heavily on Bing's index, for example.