AI agent glossary
Grounding
Tying a model's answers to specific, trusted sources such as documents, search results, or tool outputs, rather than relying on what it remembers.
A model's built-in knowledge is broad but can be outdated, incomplete, or simply wrong. Grounding means giving the model the relevant source material and asking it to base its answer on that material.
In practice, grounding often comes with a requirement to cite: each claim points back to the passage it came from. This lets readers and automated checks verify the answer, and it makes unsupported claims easier to spot.
Grounding reduces hallucination but does not eliminate it. A model can still misread a source or stitch together sources incorrectly, so grounded systems still benefit from evals and spot checks.
A research agent is instructed to answer only from the pages it retrieved, to quote the sentence supporting each claim, and to say 'not found in sources' when the evidence is missing.
Learn more in the course
Related terms
Go from definitions to doing
Learn to use and build AI agents in 65 practical lessons.