Grounding
TL;DR
Anchoring AI responses to real, verifiable sources.
What does this mean?
Grounding is the technique of tying AI responses to verifiable facts or sources. Instead of generating freely, the model responds only on the basis of concrete documents or data.
How it works
The model is provided with relevant documents or data sources. It is instructed to use only information from those sources and ideally to cite them.
Example
A customer service bot that only answers based on official FAQs and product documentation — not on the basis of its general training.
Why it matters
Grounding drastically reduces hallucinations and makes AI responses reliable and verifiable.
Related terms