This cross-sectional study compared three LLMs (Grok 4.1, Gemini 3, and ChatGPT 5.2) in classifying ultra-processed foods (UPF) using best-selling products from leading supermarket chains covering 53.2% of the national market. Of 3001 products, 2920 with complete ingredient information were included; two trained dietitians assigned NOVA groups as the reference standard. In the reference classification, 74.3% of products were UPF. Under the baseline prompt, all models underestimated UPF prevalence compared with the reference standard (p< 0.001). ChatGPT 5.2 yielded the highest binary UPF detection performance (accuracy: 69.01%; sensitivity: 59.01%; specificity: 98.00%; F1: 73.88%). Prompt sensitivity analyses revealed that a minimal prompt substantially outperformed the detailed baseline for most models (Gemini 3 F1: 94.20%; ChatGPT 5.2 F1: 92.62%). Low inter-run agreement (κ: 0.01–0.27) indicated sensitivity to model updates, supporting prompt calibration and human oversight. These findings suggest that off-the-shelf LLMs require prompt calibration and human oversight before UPF surveillance workflows.
H. Bayram, Sedat Arslan, Arda Ozturkcan· International Journal of Foo...· 0 citations
Background: Mobile health (mHealth) applications are increasingly used for dietary assessment and self-monitoring, yet their validity and comprehensiveness remain uncertain. Aim: This study aimed to evaluate the quality, features, and nutritional accuracy of popular nutrition-related mobile applications available in Türkiye by comparing their outputs with standardized 3-day food records (3D-FRs). Methods: A total of 807 apps were identified in the App Store in Türkiye using the search terms 'diet' and 'food'. Seventeen applications meeting the inclusion criteria were systematically evaluated. App quality and functionality were evaluated using the Mobile Application Rating Scale (MARS). Features were categorized into dietary, tracking, insights, technical, educational, social, and artificial intelligence (AI) domains. Energy and macronutrient estimates from the apps were compared with 3D-FRs to determine validity. Summary: MyNetDiary, Fitatu, and HealthifyMe provided the broadest range of functionalities, while only five apps incorporated AI-based features. The mean MARS score for the 17 apps was 3.61/5, with the highest average in the 'functionality' domain (4.03) and the lowest in 'engagement' (3.27). Compared with 3D-FRs, 10 of the 13 apps overestimated energy intake (+327 kcal), 8 overestimated carbohydrate (+33 g), 12 overestimated protein (+23 g), and 12 overestimated fat (+19 g). The greatest discrepancies were observed in Cronometer, Diyet 7/24, Fitbit, Lifesum, and HealthifyMe, whereas MyNetDiary and FatSecret yielded estimates closest to reference records. Despite strong performance in usability and tracking features, most apps showed systematic overestimation with notable variability (≈+327 kcal; +19-33 g among overestimating apps), although findings are limited by the use of standardized menus.
H. Bayram, Arda Ozturkcan· Nutrition and Health· 0 citations