Grounding is when an AI model bases its answer on retrieved, verifiable sources — live web results or documents — instead of relying only on what it learned in training.
An ungrounded answer comes from the model's parameters: a compressed memory of its training data, frozen at its knowledge cutoff. A grounded answer folds in fresh retrieval — web search, a document, a database — before responding.
Grounding is why AI rankings move. A model's trained opinion of a category changes only when the model is retrained, but its grounded answers shift as the underlying sources change. That is also why recommendation answers differ between the same model with and without web access.