AI Calorie Tracking Apps May Be Off by 345 Calories
A new study finds that four popular photo-based calorie counting apps underestimated calorie and fat content by about one-third on average, compared to precisely measured meals from a metabolic kitchen.
How Accurate Are AI Food Apps?
AI-powered calorie tracking apps can estimate a meal's nutritional content from a single photograph, offering a fast alternative to manual logging. However, new research suggests the results may be considerably lower than what is actually on the plate.
Researchers tested four apps—MyFitnessPal, LoseIt!, CalAI, and Appediet—using standardized photographs of 102 meals prepared in a controlled metabolic kitchen for a diet study at the NIH Clinical Center. Ingredients were measured to the nearest 0.1 gram, providing a highly accurate reference.
Key Findings
Across all four apps, estimated calorie totals were about 250 to 345 calories too low per meal on average. Fat was underestimated by approximately 30 grams. MyFitnessPal and LoseIt! were more accurate for higher-calorie meals, while all apps struggled with low-carb ketogenic meals that contain more fat.
Aaron Hengist, a postdoctoral visiting fellow at NIDDK, noted that users who rely on photo-based tracking without adjusting portions should take results with a grain of salt—actual intake is likely higher than the app shows. The findings are preliminary and have not yet undergone full peer review for publication. Olivia Charles presented the research on July 25 at NUTRITION 2026 in National Harbor, Maryland.
