Prompt design acts as a primary, modifiable factor determining the performance of artificial intelligence chatbots used for health coaching, according to a scoping review evaluating early literature on large language models. Investigators identified twelve peer-reviewed studies published through October 2025 across multiple health contexts, including medication use, mental health, physical activity, and nutrition. The findings demonstrate that how instructions are crafted directly shapes the advice delivered by conversational agents, though real-world clinical validation remains remarkably sparse across current literature.
To assess how prompt engineering strategies have been tested and reported, researchers conducted a scoping review adhering to PRISMA-ScR guidelines across PubMed, Embase/Ovid MEDLINE ALL, PsycINFO, IEEE Xplore, and Scopus. Following data extraction with a standardized form, the authors synthesized findings narratively. Across the twelve analyzed investigations, the underlying research frameworks relied primarily on simulation-based tests rather than live interactions with patients, leaving substantial gaps in practical implementation data.
The review classified prompt strategies into three broad categories: structural modifications, specialized prompting techniques, and role-based approaches. Structural alterations, which included specifying precise constraints and adding detail, emerged as the most common strategy. These structural adjustments consistently improved the clarity and overall response quality of chatbot outputs. In contrast, specialized prompting techniques improved specific task performance metrics but occasionally introduced trade-offs, such as reduced focus or lower conversational consistency across interactions.
Role-based prompting distinctly altered the persona of the models, modifying tone, response specificity, and adherence to established clinical guidelines. Across the analyzed literature, evaluation endpoints centered predominantly on response quality metrics. Noticeably fewer investigations evaluated system safety or patient user experience, and only a single study tracked clinical outcomes, highlighting an urgent need for deeper verification in applied healthcare delivery.
The authors concluded that while prompt design fundamentally shapes chatbot capabilities, current evidence remains constrained by heterogeneous methodologies and minimal evaluation involving real users. Addressing these structural limitations and expanding real-world assessments will be essential to ensure artificial intelligence chatbots can safely and effectively support human health coaching.
Subject of Research: Prompt design strategies and evaluation in large language model health coaching chatbots
Article Title: Prompt design in large language model health coaching chatbots: A scoping review
Article References: Incze, T., Khademioore, S., Bengizi, A., Mbuagbaw, L., Alvarez, E., & Anderson, L. N. (2026). Prompt design in large language model health coaching chatbots: A scoping review. PLOS Digital Health, 5(11), e0001827. https://doi.org/10.1371/journal.pdig.0001827
Image Credits: AI Generated
DOI: 10.1371/journal.pdig.0001827
Keywords: artificial intelligence, prompt design, large language models, health coaching, chatbots, scoping review, digital health, simulation-based evaluation, mental health, medication use, physical activity, nutrition, structural modifications, role-based prompting, response quality, clinical outcomes
