AI calorie-counting apps miss a third of meal calories

Clinical takeaway: Patients who log meals by photo are likely eating more than their app reports, especially fat. Treat app-reported intake as a floor when reviewing diet records.

Diet assessment has always leaned on self-reporting from individuals, which has always tended toward underestimation. Photo-based AI logging promised to fix that by taking the human out of the process. However, an NIH analysis of four popular apps suggests it hasn't. Tested against meals weighed to the tenth of a gram, the apps missed about a third of the energy on the plate.

All four apps underestimated fat by about 30 g per meal. Carbohydrate estimates held up better, with two apps showing no significant carb error at all. Beyond those results from an initial 102-meal analysis, a follow-up look at more than 200 additional meals points in the same direction: the apps struggled most with high-fat, low-carbohydrate meals, the pattern patients on ketogenic diets are photographing every day.

All four apps, MyFitnessPal, LoseIt!, CalAI, and Appediet, ran significantly low, with average errors from about 250 calories per meal at best to nearly 345 at worst. MyFitnessPal and LoseIt! were more accurate on higher-calorie meals than lower-calorie ones. Consistency from meal to meal was poor for everything except carbohydrates.

Researchers photographed meals prepared for an NIH controlled feeding study, where a metabolic kitchen weighs ingredients to 0.1 gram. They ran standardized photos through each app and compared outputs against food composition tables. Meals averaged 918 calories, ranging from 592 to 1,211.

The investigators suggest pairing photo-based features with traditional diet assessment methods to improve real-world accuracy, and their conclusion frames the open question as whether photo AI can cut the burden of conventional diet assessment rather than replace it. The findings are from a conference abstract and should be considered preliminary until peer-reviewed publication.

"People using a photo-based tracking app without adjusting the portions or entering the amounts of food should take the results with a grain of salt," said Aaron Hengist, a postdoctoral visiting fellow at the National Institute of Diabetes and Digestive and Kidney Diseases, part of the National Institutes of Health. "These apps tend to underestimate calories, especially from fats, so what they actually ate is likely higher than what the app shows."

Source: Charles O, et al. (2026 Jul 25) NUTRITION 2026, American Society for Nutrition. Photograph-based AI features in calorie-tracking apps underestimate energy content of meals