AI diet plans are risky for teens

Reports warn AI‑generated diet advice is telling some teens to eat dangerously low calories — as much as ~700 fewer calories per day than medical guidance recommends, raising safety flags about unregulated meal plans report local. A peer‑reviewed study [published] (frontiersin.org) in Frontiers in Nutrition on March 12, 2026 lists Ayşe Betül Bilen, an assistant professor at Istanbul Atlas University, as the lead author [author/affiliation] (frontiersin.org). Researchers instructed five AI models—ChatGPT‑4o, Gemini 2.5 Pro, Claude 4.1, Bing Chat‑5GPT and Perplexity—to generate diet plans and produced a total of 60 three‑day menus across two sessions for four standardized 15‑year‑old profiles [methods/sample size] (public-pages-files-2025.frontiersin.org). Compared with dietitian reference plans, the AI outputs showed systematic macronutrient deviations including an average excess of about 19.9 g protein and 15.8 g lipids and a shortfall of roughly 114.6 g carbohydrates per day [macronutrient biases] (public-pages-files-2025.frontiersin.org). AI plans delivered protein energy shares of 21.5–23.7% and lipid shares of 41.5–44.5%, while carbohydrate shares sat at 32.4–36.3%—ratios the paper reports as outside recommended adolescent distributions [AI macronutrient percentages] (public-pages-files-2025.frontiersin.org); by comparison, Acceptable Macronutrient Distribution Ranges for adolescents recommend roughly 45–65% of calories from carbohydrates, 10–30% from protein and about 25–35% from fat [AMDR ranges] (mtsu.pressbooks.pub). The authors recorded substantial variation between models in micronutrient content and concluded that no single AI consistently approximated the dietitian across all macro‑ and micronutrients [model variability finding] (public-pages-files-2025.frontiersin.org). In the paper the lead author warned that following unbalanced or overly restrictive AI meal plans “may negatively affect growth, metabolic health, and eating behaviors,” and the manuscript notes it was received Dec. 11, 2025 and accepted Jan. 20, 2026 [author warning; received/accepted dates] (public-pages-files-2025.frontiersin.org).