Multimodal and generative AI systems are increasingly proposed for clinical decision support, but the literature ranges from pragmatic trials to simulated consultations and benchmark-like evaluations.

To identify and synthesize empirical clinical and near-clinical evidence while keeping strict multimodal AI evidence separate from adjacent text-based generative AI clinical decision-support studies.

A structured, PRISMA 2020-informed evidence map and narrative synthesis covered clinician-facing studies published from 2018 to 9 July 2026.

Seventeen empirical studies were included: 12 in the strict multimodal core and five in an expanded contextual set. Patient-important outcome evidence remained limited.

The evidence supports assistive, auditable, human-in-the-loop use for triage, documentation, decision preparation, second reading, and specialist workflow support. Autonomous deployment is not supported.