Grounding (AI) in plain English.

Grounding is the practice of connecting an AI model's answers to verifiable, external sources — search results, a company's documents, a database, a live API — rather than relying solely on what the model memorized during training. A grounded answer can point to exactly where its claim came from; an ungrounded answer is only as reliable as the model's internal, sometimes outdated or wrong, recollection. RAG is the most common grounding technique, and citation-style AI search products like Perplexity are built around grounding by design — every claim links back to a source you can check yourself. Grounding doesn't eliminate hallucination entirely (a model can still misread or misquote a real source), but it dramatically reduces the "confidently made something up" failure mode and gives users a way to verify claims themselves, which is why grounding matters most for research, legal, medical, and financial use cases.