Grounding is the practice of giving an AI system relevant evidence to base its answer on. That evidence might be a knowledge base, search result, customer record, product catalog, policy document, or API response.
In practice
Grounding helps reduce guesswork. Instead of relying only on the model memory, the system can retrieve current or private information and ask the model to answer from that material.
What to watch
Grounding improves reliability, but it does not guarantee perfection. The retrieved information can be incomplete, outdated, irrelevant, or misread by the model.