Grounding
Tying a model’s answers to real, checkable sources instead of letting it free-associate.
Think of it like
Like a journalist who cites documents rather than writing “trust me.”
Example
A support bot grounded in the actual help docs quotes the real refund window instead of inventing a plausible-sounding one.
How it actually works
Grounding connects generation to external evidence — retrieved documents, tool outputs, databases — so claims can be traced and verified. It’s the main practical defense against hallucination: not because the model gets smarter, but because it’s answering from provided facts rather than parametric memory alone.
For product teams
The difference between a demo and something you’d let talk to customers.
For engineers
Conditioning outputs on retrieved/verified external context and enabling attribution; the core mitigation for hallucination.
Related
- RAG — The pattern that supplies the evidence.
- Hallucination — What it fights.
- Parametric knowledge — Knowledge baked into weights, ungrounded.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome