AI Nutrition Apps Underestimate Calories by About One‑Third, Study Finds

Researchers at the U.S. National Institutes of Health evaluated four popular AI‑driven food‑logging apps—MyFitnessPal, Lose It!, Cal AI and Appediet—by photographing 102 meals whose exact nutrient content was pre‑measured in a metabolic kitchen. Comparing the apps' estimates with the known values, the study found that each app consistently reported lower calories (by roughly 250–345 kcal per meal) and lower fat (by about 30 g), amounting to an underestimation of roughly one‑third of the actual content.

The bias was systematic, not random, and was most pronounced for meals high in fat and low in carbohydrates, such as those typical of ketogenic diets. After the initial analysis, the team examined more than 200 additional meals to explore factors influencing accuracy. The findings were presented at the Nutrition 2026 meeting of the American Society for Nutrition and raise concerns for users who rely solely on photo‑based estimates for diet tracking.

The authors recommend that consumers combine AI photo estimates with traditional portion‑size inputs or manual verification to improve tracking accuracy.