The Importance of Primary Care Subject Matter Experts: Output Quality in Large Language Models Prompt Engineering

Large language models (LLMs) offer new opportunities to support primary care by generating documentation, summarizing records, answering messages, and providing education. While LLMs show promise, their effectiveness depends on how well their outputs align with real-world clinical needs. This commentary describes one team's development of Primary Care (PC) Navigator, a multimodal artificial intelligence (AI) tool that combines audio and video from clinical encounters with an LLM to generate behavior change plans. Our pilot study showed that prompt design plays a critical role in shaping output quality. Through collaboration between clinicians and engineers, we created more effective prompts and recommend that others use the CARE framework (Context, Action, Result, Example) to ensure the outputs are accurate, relevant, and actionable. Clinician involvement is essential not just for evaluating the LLM's performance, but also for shaping its development. By participating in prompt engineering, clinicians move from passive users to co-developers. Their input ensures that LLMs in primary care are both technically functional and aligned with patient needs, professional values, and the realities of frontline practice.