Ground truth is the verified reference information used to evaluate whether a claim, prediction, or extracted result is correct. It may consist of measured observations, reviewed labels, or documented facts with an established source. It is a reference standard for a defined task, not a claim that every reference is infallible.
Grounding is different: it supplies evidence to an AI system while it generates an answer. A grounded answer can still be wrong if its sources are outdated or misinterpreted. Ground truth is what an evaluator checks that answer against.
Practical relevance
To assess whether an assistant accurately describes a brand's service regions, use a reviewed, dated list of those regions as the reference. Compare each answer against that list, not against another unverified AI response.
Reference facts need scope and maintenance. A claim about availability may be correct in one market or on one date and incorrect in another. Keep the evidence, applicable market, and effective date with the fact.