Grounding
Tying the answer to something checkable, so it can be verified instead of trusted.
Grounding means the answer is anchored to something outside the model's own generation: a retrieved passage, a tool result, a cited document. The reader can follow the chain back to the source.
It is what separates a useful system from a confident narrator. An ungrounded answer can be right, but the reader has no way to know that. A grounded one can be checked, and when it is wrong, the wrong source is visible too.
Grounding is also the practical defence against hallucination: not that the model never invents, but that inventing has nowhere to hide. The weakness is in the retrieval step, so a system can be perfectly grounded in the wrong document.
Related terms
RAG
Fetch the relevant documents first, then let the model answer using them.
Hallucination
Output that is fluent and wrong. Two different problems hide under one word.
Knowledge Cutoff
The date after which the model has no trained knowledge at all.
Web Search
A tool that lets the model look things up instead of guessing.
Citation
A pointer from a claim back to the material that supports it.
Want this working inside your team?
It starts with one conversation about your workflow.