Grounding
Tying outputs to evidence the user can check.
Providing sources, citations or tool results so the answer rests on verifiable material instead of the model’s memory. Grounding makes hallucination visible and mistakes correctable.
A grounded answer is auditable: click through and see where it came from. If a feature cannot show its sources, users have no way to calibrate trust.
Grounding is a user-interface decision as much as a modelling one. The value is not that the model becomes correct, it is that a reader can check. Systems that return citations with each claim convert an unfalsifiable assertion into a two-second verification.
Citations that point at something real
Answers include references, but they are to the document rather than the passage.
A document-level citation still leaves the reader hunting. Return the supporting span so the check is instant — and, incidentally, so that a broken citation becomes an obvious signal rather than a plausible-looking one.
- docsGrounding overview ↗
Grounding responses against search results or your own corpora, returning support citations that tie individual claims back to sources — the API shape of provenance.
- standardLLM09: Misinformation ↗
Treats overreliance as the risk: the failure is users accepting confident output without verification, so the mitigation is validation and visible limitations, not just better models.
Faithfulness · Grounding
Sticking to the material it was given.
RAG · Retrieval-Augmented Generation
Look things up, then put what you found in the prompt.
Hallucination
Fluent output that is not true.
Contextual Knowledge
What the model knows from the prompt, right now.