GenAIWiki
Retrieval

Grounding

Grounding means tying model claims to retrieved documents, tools, or other checkable evidence instead of parametric memory alone.

Expanded definition

Grounded systems retrieve or fetch evidence, then require the answer to use it. RAG is the common pattern; tool results and databases are others. Grounding is a product property you measure with faithfulness evals, not a checkbox on a model card. Citations must use real source IDs. Search grounding features from providers still need your own allowlists and logging.

Related terms

Explore adjacent ideas in the knowledge graph.

Grounding FAQ

What is Grounding?

Grounding means tying model claims to retrieved documents, tools, or other checkable evidence instead of parametric memory alone.

How is Grounding used in AI systems?

Grounded systems retrieve or fetch evidence, then require the answer to use it. RAG is the common pattern; tool results and databases are others. Grounding is a product property you measure with faithfulness evals, not a checkbox on a model card. Citations must use real source IDs. Search grounding features from providers still need your own allowlists and logging.

Related

Comparisons, tools, and models that connect to this idea.