AI-powered calorie tracking apps can estimate the nutritional content of a meal from a single photo. The technology offers a quick and convenient alternative to manually entering every food item and portion, but new research shows the results may be significantly lower than what’s actually on your plate.
In testing four photo-based apps, researchers found that the calorie and fat estimates were on average about a third too low.
How AI estimates calories from food photos
Photo-based calorie tracking relies on AI image recognition to identify foods shown in photos and estimate the size of each portion. The app then compares those estimates to a nutritional database to calculate calories and other nutrients.
“Photo-based calorie tracking apps are very popular, especially for people trying to manage their health or lose weight,” said Aaron Hengist, a postdoctoral fellow in the intramural program at the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK), part of the National Institutes of Health. “However, many of these apps have not been fully evaluated for accuracy. Our study helps address this issue by examining whether these apps can reliably estimate calories.”
Olivia Charles, a post-baccalaureate intramural research fellow at NIDDK, presented her research findings at NUTRITION 2026, the American Academy of Nutrition’s flagship annual meeting, held July 25-28 in National Harbor, Maryland, just outside Washington, DC.
Testing the app with accurately measured meals
This project is part of the NIH Clinical Center’s broader nutrition research investigating how the body processes nutrients on either a low-carbohydrate (ketogenic) diet or a standard diet.
Meals used in clinical trials are prepared in metabolically controlled kitchens, with ingredients measured to the nearest 0.1 gram by researchers. This gave the team very precise criteria to evaluate the app.
The researchers collected standardized photographs of 102 meals prepared for the dietary study. We then sent the images to MyFitnessPal, LoseIt!, CalAI, and Appediet to see how well each app’s estimates matched the known nutritional content.
“By using meals prepared in a tightly controlled metabolic kitchen, we were able to compare the app’s estimates to an accurate standard,” Hengist said. “This kind of direct, high-quality comparison has not been available until now.”
Lose hundreds of calories with apps
For all four apps, total calorie estimates were too low, averaging about 250 to 345 calories per meal. The app also underestimated fat by about 30 grams.
MyFitnessPal and LoseIt! were more accurate in analyzing high-calorie meals than low-calorie meals. All four apps also gave more consistent estimates for carbohydrates than for other macronutrients.
“People who use photo-based tracking apps without adjusting or entering food amounts should take the results with a grain of salt,” Hengist says. “These apps tend to underestimate calories, especially fat calories, so the amount you actually eat may be higher than the calories shown on the app.”
Keto diet can be difficult to measure with AI
Following the initial analysis, the researchers tested over 200 additional meals to investigate factors that may affect the app’s accuracy.
Preliminary findings indicate that apps can be very difficult to evaluate low-carb ketogenic diet meals. These meals were often high in fat, and the app tended to consistently underestimate fat.
Researchers suggest that combining photo-based tools with traditional methods of assessing food intake and diet quality could make calorie tracking more accurate in everyday use.
Dr. Charles presented this research (abstract) during a Presidential Oral Session at the Gaylord National Resort & Convention Center on Saturday, July 25th.
Abstracts presented at NUTRITION 2026 were reviewed and selected by a panel of experts. However, the full peer review process required for publication in scientific journals is generally not completed. Therefore, results should be considered preliminary until published in a peer-reviewed publication.
