AI Calorie Apps Undercount by Up to 345 Calories Per Meal
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- Four photo-based calorie apps (MyFitnessPal, LoseIt!, CalAI, and Appediet) underestimated calories by 250–345 per meal and fat by approximately 30 grams on average when researchers compared their estimates against 102 meals prepared with ingredients measured to 0.1 gram precision at the NIH Clinical Center.
- Olivia Charles, a postbaccalaureate intramural research training fellow at NIDDK, presented the findings on July 25 at NUTRITION 2026, the flagship annual meeting of the American Society for Nutrition held July 25–28 in National Harbor, Maryland.
- Aaron Hengist, a postdoctoral visiting fellow with NIDDK's Intramural Program, warned that the apps "tend to underestimate calories, especially from fats," advising users to "take the results with a grain of salt" when relying on photo-only logging.
- High-fat ketogenic meals posed the greatest difficulty — a follow-up test of more than 200 additional meals indicated the apps had the most trouble with dishes from the low-carb diet, which are richer in fat.
- MyFitnessPal and LoseIt! performed more accurately on higher-calorie meals than on lower-calorie ones, and all four apps produced more consistent estimates for carbohydrates than for any other macronutrient.
- The results are preliminary, having been presented at a conference before completing the full peer-review process required for journal publication.
Why it matters: For the millions using photo-based calorie tracking to manage weight or diabetes, a systematic 250–345-calorie undercount — concentrated in fat, which the apps miss most — means users may log substantially less than they actually consumed. The NIH's controlled-kitchen benchmark, with ingredients measured to 0.1 gram, sets a more rigorous bar than earlier app-accuracy studies.
Ask SkimNews
