Grounding
Also called: Groundedness
Grounding is the practice of tying a model's answers to specific source material that can be checked, rather than to whatever it absorbed during training. A grounded answer arrives with the passage it came from, so a reader can open the document and confirm it. Retrieval-augmented generation is the usual way of implementing this: the relevant documents are fetched at the moment the question is asked and placed in front of the model.
In practice
Grounding is the working antidote to hallucination, and often the difference between an AI answer a regulator will accept and one it will not, because a claim with no source attached stands out. When assessing a tool, ask whether every answer carries a citation to a retrievable document, and what the system does when it finds nothing relevant: saying so is a better failure than a confident guess.