# Grounding

> Tying a model's answers to real sources — retrieved documents, databases, live data — so it states what is true instead of what sounds right.

An ungrounded model answers from the internet-shaped blur it was trained on; a grounded one answers from your price list. The mechanism is usually retrieval — find the relevant fragment, put it in the context, instruct the model to use only that — plus the discipline to say 'not in the sources' instead of guessing.

Grounding is the difference between a chatbot that demos well and one you can point at customers: it is also the difference between a defensible answer and a fluent hallucination. Citations back to the source turn a claim into a checkable claim.

## Related terms

- https://dfieldsolutions.com/en/glossary/rag.md
- https://dfieldsolutions.com/en/glossary/hallucination.md
- https://dfieldsolutions.com/en/glossary/knowledge-base.md

---

Source: https://dfieldsolutions.com/en/glossary/grounding
DField Solutions — Dunakeszi, Hungary — dezso@dfieldsolutions.com
Booking: see https://dfieldsolutions.com/en/contact