Prompt design in large language model health coaching chatbots: A scoping review
Prompt-Design in Health-Coaching-Chatbots mit Large Language Models: Scoping Review
PLOS Digital Health · 12.11.2026 · 5 (11) · e0001827
DOI: 10.1371/journal.pdig.0001827 ↗Auf einen Blick
Die Scoping Review fasste 12 Studien zu Prompt-Design in KI-gestützten Health-Coaching-Chatbots zusammen. Am häufigsten wurden strukturelle Anpassungen berichtet, die die Antwortqualität und Klarheit verbesserten.
Fragestellung
Wie wurde Prompt-Design in Large-Language-Model-basierten Health-Coaching-Chatbots angewendet und bewertet?
Methode
Scoping Review nach PRISMA-ScR; Suche in PubMed, Embase/Ovid MEDLINE ALL, PsycINFO, IEEE Xplore und Scopus bis Oktober 2025.
Ergebnisse
12 Studien zu Mental Health, Medikamenteneinnahme, körperlicher Aktivität und Ernährung. Meist simulationsbasiert; wenige mit realen Nutzer:innen. Struktur-, Technik- und Rollen-Prompts wurden berichtet; Endpunkte meist Antwortqualität.
Einschränkungen
Heterogene Studiendesigns und nur geringe Evaluation im realen Einsatz; Sicherheit und Nutzererfahrung seltener untersucht.
Schlussfolgerung der Autor:innen
Prompt-Design ist ein zentraler, veränderbarer Faktor der Chatbot-Leistung. Die Autor:innen sehen Bedarf an mehr Forschung für Qualität, Sicherheit und realen Nutzen.
Kernaussagen
- Strukturelle Prompt-Anpassungen waren am häufigsten
- Prompting-Techniken verbesserten oft die Task-Performance
- Nur eine Studie untersuchte klinische Outcomes
Deutsche Zusammenfassung: KI-gestützt aus dem Original-Abstract erstellt; alle Zahlen wurden automatisch mit dem Original abgeglichen. Maßgeblich ist der Original-Abstract.
Kurz gefasst
Scoping Review in der Zeitschrift PLOS Digital Health, online erschienen am 12.11.2026. Verfasst von Taylor Incze und 5 weiteren Autor:innen. Beteiligte Einrichtung: Department of Health Research Methods, Evidence, and Impact, McMaster University, Hamilton, Ontario, Canada. Die Arbeit nennt 41 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Artificial intelligence (AI) chatbots are being explored to deliver health coaching across multiple contexts. However, the quality and safety of chatbot-generated advice is influenced by how prompts are constructed, and prompt design remains inconsistently described in the literature. This scoping review examined how prompt design has been applied and evaluated in AI health coaching chatbots and summarized the study designs, prompt strategies, and outcomes reported across studies. We conducted a scoping review following PRISMA-ScR guidance and searched PubMed, Embase/Ovid MEDLINE ALL, PsycINFO, IEEE Xplore, and Scopus for peer-reviewed studies published up to October 2025. Eligible studies examined prompt design in large language model–based health coaching chatbots. Data were extracted using a standardized form and synthesized narratively. Twelve studies were included, spanning mental health, medication use, physical activity, and nutrition contexts. Study designs were primarily simulation-based, with few involving real users. Prompt strategies were grouped into structural modifications, prompting techniques, and role-based approaches. Structural modifications, such as increasing specificity and adding constraints, were most common and were reported to improve response quality and clarity. Prompting techniques were reported to improve task performance but sometimes reduced focus or consistency. Role-based prompting influenced tone, specificity, and guideline alignment. Outcomes focused mainly on response quality, with fewer studies assessing safety or user experience and only one evaluating clinical outcomes. Prompt design is a key, modifiable determinant of chatbot performance. However, current evidence is limited by heterogeneous study designs and minimal real-world evaluation. Greater focus on prompt design is needed to improve the quality and safety of AI-delivered health coaching and its real-world impact on health outcomes.
Autor:innen und Institutionen
- Department of Health Research Methods, Evidence, and Impact, McMaster University, Hamilton, Ontario, Canada.
- School of Medicine, University College Dublin, Dublin, Ireland.
- Department of Pediatrics, McMaster University, Hamilton, Ontario, Canada.
- Biostatistics Unit, Father Sean O'Sullivan Research Centre, St. Joseph's Healthcare, Hamilton, Ontario, Canada.
- Centre for Development of Best Practices in Health (CDBPH), Yaoundé Central Hospital, Yaoundé, Cameroon.
- Division of Epidemiology and Biostatistics, Department of Global Health, Stellenbosch University, Cape Town, South Africa.
ORCID-Profile: Taylor Incze, Sahar Khademioore, Alla Bengizi, Elizabeth Alvarez
Zitieren
APA
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
Vancouver
Incze T, Khademioore S, Bengizi A, Mbuagbaw L, Alvarez E, Anderson LN. Prompt design in large language model health coaching chatbots: A scoping review. PLOS Digital Health. 2026;5(11):e0001827. doi:10.1371/journal.pdig.0001827
BibTeX
@article{Incze2026,
author = {Incze, Taylor and Khademioore, Sahar and Bengizi, Alla and Mbuagbaw, Lawrence and Alvarez, Elizabeth and Anderson, Laura N},
title = {Prompt design in large language model health coaching chatbots: A scoping review},
journal = {PLOS Digital Health},
year = {2026},
volume = {5},
number = {11},
pages = {e0001827},
doi = {10.1371/journal.pdig.0001827}
}
EUVIMED