Grounding supplies relevant source material to the model and instructs it to answer only from that material, so statements correspond to actual documents rather than to patterns learned during training.
Well-grounded systems also decline to answer when the sources do not contain the information, rather than filling the gap with plausible content.
Alternative Names:
Source Grounding, Grounded Output
Why it Matters?
Grounding is the architectural decision that makes AI usable in litigation. It changes the question from whether the model knows something to whether the documents say it, which is the only question that matters when the answer must be verifiable. The behavior worth testing during evaluation is what the system does when the answer is not in the record, since willingness to say so is a better quality signal than any benchmark.
Frequently Confused with
Related terms
Frequently asked questions
How do I test whether a system is well grounded?
Is grounding the same as retrieval-augmented generation?
