AI-powered calorie tracking apps miss hundreds of calories per meal when estimating food from photographs, according to new research presented at the American Society for Nutrition’s annual meeting, NUTRITION 2026, held July 25-28 in National Harbor, Maryland. In a direct evaluation of four smartphone applications using precisely measured meals from a metabolic kitchen, investigators found that software estimates for calories and fat fell roughly one-third short of actual plate contents.
How AI Calorie Apps Underestimate Meals
Photo-based diet applications rely on artificial intelligence image recognition to scan a picture, identify visible food items, estimate portion sizes, and cross-reference internal nutrition databases.
When the research team submitted those images to MyFitnessPal, LoseIt!, CalAI, and Appediet, the results revealed significant shortfalls. Across all four tested platforms, estimated calorie totals ranged between 250 and 345 calories too low per meal on average. Furthermore, the apps underestimated fat content by approximately 30 grams per meal. Olivia Charles, a postbaccalaureate intramural research training fellow at NIDDK, presented these findings during the President’s Oral Session at the Gaylord National Resort & Convention Center.
Metabolic Kitchen Testing Reveals Carbohydrate and Fat Gaps
The investigation demonstrated that app performance varied depending on the macronutrient composition of the food. All four evaluated platforms produced more consistent carbohydrate estimates than fat or overall calorie metrics. Additionally, MyFitnessPal and LoseIt! exhibited higher accuracy when processing higher-calorie meals compared to lower-calorie plates.
The broader nutrition study at the NIH Clinical Center investigates how the human body processes nutrients on a low-carbohydrate ketogenic diet versus a standard diet. Following the initial analysis, researchers tested more than 200 additional meals to isolate specific variables affecting app precision. Preliminary findings indicate that ketogenic diet meals present a tougher challenge for AI tools, primarily because high-fat content items are consistently underestimated by the software.
Practical Implications for Diet Tracking Users
Because these digital tools frequently miss dense macronutrients like fats, consumers relying solely on automatic photo recognition may inadvertently consume significantly more calories than their logs indicate. Hengist advises that individuals using photo-based tracking features without manually adjusting portion sizes or entering exact food amounts should treat the software outputs with caution.
Researchers suggest that combining AI photo-recognition features with traditional manual food logging and rigorous diet evaluation methods can improve tracking accuracy for everyday users. Because abstracts presented at NUTRITION 2026 are selected by expert committees but have not yet completed full peer review, scientists emphasize that these results remain preliminary until published in a peer-reviewed scientific journal.
Worth a look