Answers are grounded in this paper and cite their sources.
Background Generative artificial intelligence (AI) systems are increasingly being explored for menu planning in institutional foodservice settings; however, evidence regarding the nutritional adequacy of AI-generated menus remains limited. Aim This study aimed to evaluate and compare the nutritional adequacy of menus generated by ChatGPT, Claude, and Gemini for three institutional foodservice settings serving different life stages: kindergarten, university cafeteria, and nursing home. Methods A theoretical menu analysis was conducted using 4-week weekday menu cycles generated by ChatGPT, Claude, and Gemini. Menus were developed for three institutional settings, yielding 180 daily menus. Energy, nutrient, and food group contents were analyzed using the Nutrition Information System (BEBIS 9.0), and adequacy was assessed according to the Turkish Dietary Guidelines. Summary Exploratory comparisons revealed differences in the nutritional composition of the generated menu sets across institutional settings. Mean energy content ranged from 887.9 to 1161.0 kcal in kindergarten menus and from 842.9 to 1189.0 kcal in university lunch menus (p < 0.001). Claude-generated menus generally contained higher amounts of protein and several micronutrients, whereas Gemini-generated menus provided lower amounts of certain nutrients, including folate, vitamin C, potassium, and calcium. Excessive estimated sodium content was a consistent concern across all institutional settings, while calcium contributions were generally low in kindergarten and university menus. These findings highlight the importance of professional nutritional evaluation of AI-generated menus, particularly with respect to sodium content and micronutrient adequacy. Given the single-generation design, the observed differences should be interpreted as exploratory findings specific to the evaluated menu cycles.
No takes yet. Share an insight, caveat, or question.
Bahar Yalçın (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: