In plain English
Grounding connects a model's outputs to verifiable, real-world data sources so its answers are based on actual information rather than only on patterns in its training. By giving the model relevant documents to work from, and ideally citing them, grounding reduces hallucinations and improves factual accuracy. It is a key idea behind retrieval-augmented generation.
Why it matters
Ungrounded models state things confidently whether or not they are true. Grounding is one of the most practical ways to make AI answers trustworthy enough to act on.
A worked example
A support assistant that answers from your company's current help articles, and links to them, is grounded; one that answers from memory alone is not.
Common confusion
Grounding reduces hallucination but does not eliminate it. A model can still misread or misquote a source, so verification still matters.