•  
  •  
 

Document Type

Conference Proceedings

Abstract

Objective: To evaluate the efficacy of ChatGPT in generating culturally appropriate and personalized lifestyle interventions to support clinicians in shared decision-making and lifestyle prescribing. Methods: This descriptive pilot study was conducted in 2025 using ChatGPT-3.5 and Gemini-2.5 Flash AI. Ten culturally diverse patient vignettes representing varied ethnic, dietary, and lifestyle backgrounds were created. For each vignette, ChatGPT generated three culturally tailored meal plans, two snack alternatives, and one home-based physical activity plan. A rubric aligned with American Dietetic Association (ADA) guidelines was developed to assess cultural alignment, ingredient substitutions, and exercise recommendations. Four human researchers and the Gemini model independently evaluated the outputs using this rubric. Readability was assessed using the Flesch-Kincaid Grade Level and Reading Ease Score. Descriptive comparisons assessed agreement between AI and human evaluations. Results: ChatGPT produced lifestyle plans consistent with ADA guidelines and demonstrated strong cultural adaptability. Gemini’s grading closely mirrored researchers’ assessments, except in carbohydrate evaluation for low-carbohydrate diets, where it outperformed human raters. Flesch-Kincaid readability scores had an average of 8.28 and (range 7.4-9.7). The Flesch-Kincaid reading ease score averaged that of 10th to 12th grade and in some vignettes, had values that scored college level. Conclusion: With well-designed prompts, ChatGPT effectively generated guideline-consistent, personalized, and culturally relevant lifestyle plans. However, readability exceeded the ideal level for general patient comprehension, suggesting the need for physician guidance. Further research should explore ethical and practical frameworks for clinical integration.

Download

Share

COinS