AI-generated menus versus the Türkiye Dietary Guidelines: nutrient adequacy, ultra-processed food burden, cost, and carbon footprint

Summary

This study evaluated whether menus generated by two large language models (ChatGPT-4o and Gemini Pro) align with Türkiye's national dietary guidelines across eight life stages and physiological conditions. Although both AI tools achieved similar energy adequacy to guidelines (~95%), Gemini produced significantly higher carbohydrate adequacy whilst ChatGPT delivered better dietary fibre; more concerning, Gemini-generated menus contained substantially more ultra-processed food components (41.7% versus 16.7%) and generated higher carbon footprints, despite lower average costs. The findings suggest caution in relying on commercial AI tools for meal planning without explicit guardrails for food processing level and environmental sustainability.

Regional applicability

Whilst this study was conducted in Turkey, the findings are relevant to United Kingdom practice and policy: they highlight the need for critical evaluation of AI-based meal planning tools against national dietary guidelines, food processing standards, and sustainability metrics. Similar validation would be valuable for UK guidance (e.g. Eatwell Guide) to ensure AI tools properly reflect nutrient adequacy targets and the Ultrarian principles prioritised by UK public health policy.

Key measures

Energy and macronutrient adequacy (%), ultra-processed food (NOVA 4) proportion, daily cost (Turkish lira), carbon footprint (kg CO₂-eq/day), temporal consistency

Outcomes reported

The study compared AI-generated menus from ChatGPT and Gemini with Türkiye Dietary Guidelines across nutrient adequacy, ultra-processed food content, cost, and carbon footprint for eight population groups. Energy adequacy was similar across AI tools (95.2–95.1%) and guidelines (93.4%), but carbohydrate and dietary fibre adequacy differed significantly; Gemini menus contained more ultra-processed foods (41.7% vs 16.7% for ChatGPT) and had higher carbon footprints (7.3 vs 6.4 kg CO₂-eq/day).

Topic tags