Position: Clinical Generative AI Must Connect Information, Evidence, and Action

Clinical AI seeks to transform patient information into evidence that can support action. Generative AI makes this challenge increasingly important by producing rich, open-ended clinical information, reasoning, and recommendations at scale. We argue for an explicit progression from information to evidence to action. Our position is that AI-derived information should be evaluated relative to domain-grounded information and qualified for a specified claim; evidence should be explicitly linked to the action it is intended to justify; and the supported claim should guide AI-derived evidence toward meaningful patient-level Translation or population-level Discovery.