Can artificial intelligence (AI) chatbot tools be used effectively for nutritional management in obesity?

Abstract

Background Artificial intelligence (AI), particularly Chat Generative Pre-Trained Transformer (ChatGPT), has been suggested as a tool for dietary planning in different diseases. Aim The study aimed to compare the energy, macro and micronutrients of the sample menu components presented by ChatGPT-4o and ChatGPT-4 for obesity with the Turkish Dietary Guidelines (TDG)-2022, evaluating their accuracy and clarity in medical nutrition management. Due to higher accuracy levels and the most preferred AI, ChatGPT-4o and ChatGPT-4 were selected for comparison. Methods A comparative content analysis was conducted using ChatGPT-4o, and ChatGPT-4 to generate 1800 kcal daily diet plans for a 20-year-old female with obesity. AI models provided recommendations for dietary management, the nutrition care process, and menu planning. Three dietitians evaluated the outputs. Data were analyzed using SPSS 24.0. Results ChatGPT-generated menus were inconsistent with dietary recommendations. Both ChatGPT-4o and ChatGPT-4 offered 5-day menu samples with low calorie content of “1800 kcal prompt” compared to the TDG-2022 (P < 0.001 for ChatGPT-4o). Additionally, key nutrients, particularly fats (P = 0.003), carbohydrates (%), potassium, and calcium (P < 0.05 for all) were inadequately compared to the TDG-2022. Nutrient analysis revealed that both models underperformed in meeting recommended intakes for critical micronutrients such as calcium, and had an unbalanced distribution of macronutrients. Conclusion ChatGPT-4o and ChatGPT-4 have limitations when used to provide accurate dietary management. While AI chatbots offer useful insights, they cannot replace expertise of dietitians in clinical planning; as a result, caution is advised when using these tools in this context.