AI Glossary

Grounding

What does “Grounding” mean?

Grounding: Tying a model's answer to specific, verifiable sources, such as supplied documents, search results or a database, rather than letting it rely only on what it memorized during training. Grounded systems often cite where each claim came from. Grounding reduces, but does not eliminate, made-up answers.

Grounding visual guide: Tying a model's answer to specific, verifiable sources, such as supplied documents, search results or a database, rather than letting it rely only on what it memorized during training. Grounded systems often cite where each claim came from. Grounding reduces, but does not eliminate, made-up answers.
Visual guide Open full size (new tab)

What is an example of “Grounding”?

A grounded research tool might answer with footnotes linking each statement to the web page it was drawn from.

Keep learning

See also

Retrieval-augmented generationTermHallucinationTermKnowledge cutoffTerm

Same topic

Frequency and presence penaltiesTermIn-context learningTermFew-shot promptingTermInference costTermEmbeddingsTerm