Skip to content

AI Glossary · Concepts

What is Grounding?

Grounding

Grounding means tying a model's answer to specific, supplied sources, such as retrieved documents or database results, so that each claim can be traced back and checked.

Also known as: grounded generation

· Updated · Chain of Thought

RAG & Retrieval

  • 4 explainers
  • 4 episodes
  • 1 source

An ungrounded answer comes from whatever the model absorbed in training, with no way to see where a claim came from. A grounded answer is built from material put in front of the model at the time of the question, usually by retrieval, and ideally cites which source supports which claim. That gives a reader something to check and gives the model less reason to invent.

Grounding can reduce hallucinations but doesn’t guarantee accuracy. The sources can be wrong or stale, the retrieval can miss the right document, and the model can still go beyond what the sources say. Teams measure the last of these with faithfulness checks, and many enterprise hallucination problems turn out to be data problems.

A faithful answer from the wrong version

An invented case: a company’s knowledge base holds two versions of its travel policy. The 2025 page caps meals at $50 a day; the 2026 page raises the cap to $75. Retrieval returns the 2025 page, and the assistant answers “$50 a day” and cites it.

Every claim in that answer is supported by its source, so a faithfulness check passes. The answer is still wrong for an employee travelling this year. Grounding made the error easy to trace: the citation shows which page was used, and the date on that page shows why it no longer applies. It didn’t prevent the error. Check that the supplied evidence is current and applies to the question, as well as whether the answer stays within it.

Sources

Go deeper

From the conversation