AI Calorie Apps Underestimate Meals by Up to 345 Calories

Photo-based food trackers promise to reduce the burden of manually logging ingredients and portions, but systematic errors could undermine weight-management and nutrition decisions. The research grew out of a controlled-feeding study at the U.S. National Institutes of Health Clinical Center, where meals were prepared in a metabolic kitchen and ingredients were weighed to the nearest 0.1 gram, giving investigators a precise benchmark for testing AI estimates.

Researchers presented preliminary findings at the American Society for Nutrition’s NUTRITION 2026 meeting on July 25. They ran standardized images of 102 meals through MyFitnessPal, Lose It!, CalAI and Appediet. All four apps underestimated energy by an average of about 250 to 345 calories per meal and fat by roughly 30 grams, with high-fat ketogenic meals posing greater difficulty. The team said photo tools should be paired with traditional dietary records to improve accuracy.