Grounding

Grounding is the practice of tying an AI response to supplied or retrieved sources so its claims can be checked against real evidence.

Grounding is the practice of tying an AI response to supplied or retrieved sources so its claims can be checked against real evidence. In a grounded response, the model is not asked to invent an answer from memory alone. It is asked to use information that a person can inspect.

Grounding is useful for two reasons: it can improve factual reliability, and it makes it easier to see where a response came from. A visible AI citation is one practical sign that an answer was grounded in a source.

Is grounding the same as RAG?

They are related but not identical. RAG is a common technical pattern for retrieving information before generation. Grounding is the broader outcome: an answer is anchored to evidence, whether that evidence was retrieved automatically, supplied by a user, or selected by a system.