When LLMs guide the plate: exploring diet recommendations for mental health support

Large Language Models (LLMs) are increasingly used in mental health support, yet their role in nutritional psychiatry remains underexplored. This study examines how LLMs generate dietary recommendations for anxiety, depression, and stress. We generated 300 personalized meal plans using synthetic user profiles and evaluated outputs using descriptive statistics, co-occurrence analysis, qualitative content analysis, and expert validation by seven registered dietitians. Results show that the LLM consistently produced structured, nutrient-dense meal plans aligned with established nutritional guidelines. Common foods such as spinach, quinoa, banana, and salmon appeared frequently across all conditions, alongside recurring meal patterns including oats-based breakfasts and balanced protein–vegetable pairings. While recommendations were largely consistent, minor variations across conditions were observed. The model also showed limited contextual adaptation, including culturally relevant foods. Overall, LLMs show potential as complementary tools for translating dietary guidance into practical recommendations, though further validation and refinement are required for clinical application.