Answer verification checks each important claim in an AI output against permitted evidence. You find the supporting record for every date, number, name, and commitment, then flag anything the source doesn't contain. This check is narrower than a broad editorial review of tone or format. It is also distinct from model validation in machine learning, which evaluates a model on data that wasn't used for training. Answer verification traces the claims in the answer you plan to use back to their sources. If an important claim can't be confirmed, you hold or correct the answer.
