Grounding
Definition
Grounding is connecting an AI model’s output to retrieved external information—documents, search results, or tools—so answers stay tied to sources.
Full definition
Without grounding, a model relies mainly on patterns learned in training. With grounding (often via retrieval), the system can cite or condition on current pages, databases, or APIs. That reduces some hallucinations and enables fresher answers.
For site owners, grounding means your public pages may be fetched or indexed for use in answers. Crawl access, clear structure, and accurate facts improve the chance of being selected as grounding material. See RAG and Hallucination.