Grounding
Grounding is the practice of tying a model's responses to verifiable external information — retrieved documents, tool outputs, or a knowledge base — so answers reflect real, current data rather than only the model's internal parameters. Grounding reduces hallucination and supports source attribution and citation.
Definition
GroundingA model's parametric knowledge is fixed at training time and can be incomplete or outdated. Grounding supplies relevant, authoritative context at inference time and instructs the model to answer from it, ideally citing the specific evidence it used. This makes outputs checkable and keeps them aligned with an organization's own data.
Grounding is most commonly implemented through retrieval-augmented generation, but also includes tool and search results, structured lookups, and connected data via protocols like MCP. Effective grounding pairs good retrieval with prompting that discourages the model from filling gaps with unsupported claims.
Frequently asked questions
What is Grounding?
Grounding is the practice of tying a model's responses to verifiable external information — retrieved documents, tool outputs, or a knowledge base — so answers reflect real, current data rather than only the model's internal parameters. Grounding reduces hallucination and supports source attribution and citation.
Sources
Related terms
← Full AI & prompt engineering glossary
Put Grounding to work with Prompeteer, the Agentic Contextual AI Platform →