Tools & retrieval
What is Grounding?
Grounding ties a claim to an observation: a quote, API field, or test result.
Ungrounded answers can be fluent and wrong. Grounding is a designed step, not a model personality.
Why it matters
Grounding connects model claims to evidence. When an agent says "the quarterly revenue was $4.2M," grounding means that claim is tied to a specific API response, document passage, or database query — not hallucinated.
Without grounding, agent outputs look authoritative but may be fabricated. Grounding is the design pattern that prevents this.
Grounding techniques
The simplest grounding technique is citation: include the source observation alongside the claim. More sophisticated approaches include fact-checking tools that verify claims against a knowledge base, and structured output that requires a source field for every assertion.
Grounding is not a model personality ("be accurate"). It is a workflow step that requires observations to be present and cited.
Key takeaways
- 1Grounding ties claims to evidence — quotes, API fields, test results.
- 2Ungrounded answers can be fluent and completely wrong.
- 3Grounding is a designed workflow step, not a prompt instruction.
Common mistakes
- ✕Telling the model to "be accurate" instead of requiring citations.
- ✕Accepting outputs that sound confident without checking against observations.
Related terms
Concept neighborhood
Terms linked from Grounding in the glossary graph.
- Grounding
- Observation
- Verification
- RAG