AI dietician: Unveiling the accuracy of ChatGPT's nutritional estimations

• Published in: Nutrition (Burbank, Los Angeles County, Calif.) Los Angeles County, Calif.), 2024-03, Vol.119, p.112325-112325, Article 112325
• Main Authors: Haman, Michael, Školník, Milan, Lošťák, Michal
• Format: Article
• Language: English
• Subjects: Accuracy; Agriculture; Artificial intelligence; Calories; Carbohydrates; Chatbots; ChatGPT; Chronic illnesses; Coefficient of variation; Diabetes; Diet; Dietary planning; Energy; Energy value; False information; Food; Lipids; Meals; Nutrients; Nutrition research; Nutritional data; Planning; Proteins; Public health; Weight control; Weight management
• ISSN: 0899-9007; 1873-1244
• DOI: 10.1016/j.nut.2023.112325

•ChatGPT provides reasonably accurate nutritional data for diet management.•ChatGPT has high accuracy with caloric values close to the United States Department of Agriculture data.•Consistent nutrient data from ChatGPT are supported by low variation. We investigate the accuracy and reliability of ChatGPT, an artificial intelligence model developed by OpenAI, in providing nutritional information for dietary planning and weight management. The results have a reasonable level of accuracy, with energy values having the highest level of conformity: 97% of the artificial intelligence values fall within a 40% difference from United States Department of Agriculture data. Additionally, ChatGPT displayed consistency in its provision of nutritional data, as indicated by relatively low coefficient of variation values for each nutrient. The artificial intelligence model also proved efficient in generating a daily meal plan within a specified caloric limit, with all the meals falling within a 30% bound of the United States Department of Agriculture's caloric values. These findings suggest that ChatGPT can provide reasonably accurate and consistent nutritional information. Further research is recommended to assess the model's performance across a broader range of foods and meals.