Abstract

Using Conversational AI Effectively: Prompt Engineering for Medical Scientists and Professionals

by Giordano Daniela

Generative AI and conversational AI systems are rapidly becoming part of the everyday toolkit of researchers and healthcare professionals, supporting activities such as literature exploration, scientific writing, data interpretation, patient communication, education, and research workflows. Their effectiveness, however, depends strongly on how tasks are formulated, how context is provided, and how outputs are critically evaluated. This practical, demonstration-based session introduces the principles of prompt engineering for the effective, responsible, and reproducible use of Generative AI in medical and scientific contexts. Through live examples, participants will explore how prompt structure, contextual information, constraints, examples, and iterative refinement can improve the relevance and quality of AI-generated outputs. The session will cover applications such as summarizing and comparing scientific evidence, generating and revising scientific text, extracting structured information, adapting explanations to different audiences, and supporting research and professional documentation. Particular attention will be devoted to multimodal tasks, showing how current AI systems can work across text, images, charts, tables, and documents, and how multimodal prompting can support the interpretation and integration of heterogeneous information. Participants will also explore techniques for asking models to clarify assumptions, identify uncertainty, compare alternative interpretations, and critically review their own outputs. Key limitations will be highlighted, including hallucinations, bias, overconfidence, unverifiable information, and risks associated with sensitive data. A human-in-the-loop approach will therefore be emphasized, with AI-generated content systematically assessed against authoritative sources, domain knowledge, and professional judgement. The aim is not to teach “magic prompts”, but to provide a transferable method for interacting with increasingly multimodal AI systems: defining the task, providing appropriate context and data, specifying the expected output, evaluating the response, and iteratively refining the interaction.