GlossaryRAG
Web grounding
Forcing an LLM's answer to cite fresh, retrievable web sources rather than hallucinating from training data.
Web grounding is the practice of constraining an LLM to base its output on documents retrieved from the live web at the moment of the query. The goal is twofold: factual correctness (the model can't make up information when it has cited sources) and freshness (the model isn't limited to its training cutoff). A grounded answer typically includes inline citations linking back to the source URL.