Tianpei Yang, Daniel Stahlhoven, Hyun Soo Ko
The progressive advancement of generative artificial intelligence and in particular vision-language models (VLM) has led to its adoption and integration in radiology. Prompting, the process of providing instructional and contextual information (input) to the model, serves as a critical interaction influencing the quality and clinical usefulness of the output. This article provides an overview of the core prompting principles, provides practical examples of radiological application, and discusses limitations of VLMs relevant to clinical practice and education.