AI diet plans undernourish teens
A fresh investigation found AI‑generated diet plans for adolescents systematically underestimate required nutrition and miss key nutrients—at a time when adolescent overweight and obesity rates are rising globally (news-medical.net). That gap underscores the need for human oversight and clinical validation before deploying AI meal plans for vulnerable groups (news-medical.net).
A cross-sectional study led by Ayşe Betül Bilen and Gülen Ecem Kalkan from Istanbul Atlas University was published March 11, 2026 in Frontiers in Nutrition as “Artificial intelligence diet plans underestimate nutrient intake compared to dietitians in adolescents.” (frontiersin.org 1) (frontiersin.org 2) Researchers prompted five AI models—ChatGPT‑4o, Gemini 2.5 Pro, Claude 4.1, Bing Chat‑5GPT and Perplexity—to generate diet plans and compared them to dietitian reference plans created for four standardized adolescent profiles. (frontiersin.org) (frontiersin.org) Across 60 three‑day plans, AI outputs underestimated total energy by a mean bias of +695 kcal and misestimated macronutrients by protein +19.9 g, lipid +15.8 g and carbohydrate +114.6 g compared with the dietitian plans. (frontiersin.org) (frontiersin.org) Macronutrient ratios in AI plans skewed above adolescent guidelines for protein (21.5–23.7%) and lipids (41.5–44.5%) while falling below for carbohydrates (32.4–36.3%), and no single model matched the dietitian across micronutrient targets. (frontiersin.org) (frontiersin.org) The team analyzed energy and micro/macro nutrients with BeBiS software and statistical methods including single‑sample t‑tests, Cohen’s d and Bland–Altman analyses to quantify bias and agreement between AI and dietitian plans. (frontiersin.org) (frontiersin.org) The authors and an Istanbul Atlas University release warned that the deviations pose risks during adolescence—a period when WHO estimates showed roughly 390 million overweight and 160 million obese 5–19‑year‑olds in 2022—and called for professional supervision and clinical validation of AI meal‑planning tools. (frontiersin.org) (atlas.edu.tr) (frontiersin.org)
