How do you audit an AI-generated customer response?
An auditable AI response retains the customer input, retrieved evidence, source versions, model output, confidence or policy checks, human interventions, final response, and timestamps as one reviewable chain.
In practical terms
A model log alone is not an audit trail. Reviewers need enough context to reconstruct why a response was produced and what changed before it reached the customer. The exact evidence retained should follow the organization's legal, privacy, records-retention, and operational requirements.