What Is Grounding?

Grounding — connecting an AI model's outputs to verifiable external sources or data so the responses can be traced and trusted.

Grounding is a primary defence against fabrication: rather than answering from its parameters alone, a grounded system draws on retrieved documents or authoritative data and can cite them. Retrieval-augmented generation is a common grounding technique. Grounding reduces, but does not eliminate, hallucination — the model can still misuse the sources it retrieves.

Source: Machine-learning practice