AI glossary
Grounding
Tying a model’s answer to specific sources — search results, documents, a database — so it is based on checkable facts rather than memory alone.
A model answering from its training alone can be out of date or simply wrong (hallucination). Grounding gives it fresh, relevant material in the context and asks it to answer from that, often with citations you can follow.
RAG and web search are the most common forms. Grounding reduces errors but does not remove them: the model can still misread a source, so check the citations that matter.
Example: You ask about a shop’s returns policy. An assistant with grounding finds the shop’s official page, answers from it and links you to the exact section so you can check.
In practice
- Ask it to cite the source for every important claim.
- Tell the model what to do if the source does not answer: “say it is not in the document”.
- Check that the quote really says what the model claims.
More in our guide to hallucinations and privacy.
