Grounding
The practice of tying a model's output to specific, verifiable source material so that each claim can be traced back and checked.
In more detail
Grounding means an answer is anchored in real evidence rather than the model's unsupported recall. A grounded system retrieves relevant sources, generates from them, and ideally shows which source backs which claim. That traceability is what separates a defensible answer from a plausible guess.
Grounding is the practical response to hallucination. It does not make a model incapable of error, but it changes the failure mode from confident fabrication to a claim you can verify against a named source. For business use, where a wrong answer has consequences, that traceability is often the deciding factor.
Where this shows up at Ceven
Grounding is central to how Ceven's research works: wide and deep research return briefs with citations, so every claim points back to a source the reader can check rather than asking them to trust an unsourced summary. That is the difference between a research brief you can act on and one you have to re-verify by hand.