Four popular AI calorie-tracking apps underestimated the calories in real meals by about 250 to 345 calories each, roughly one-third of the meal, according to research from the National Institute of Diabetes and Digestive and Kidney Diseases presented on July 25 at NUTRITION 2026 in National Harbor, Maryland. The same apps missed about 30 grams of fat per meal, and early follow-up work suggests they do worst on high-fat ketogenic food.
Key takeaways
- MyFitnessPal, LoseIt!, CalAI and Appediet all undercounted calories by about 250 to 345 per meal and fat by about 30 grams.
- The reference meals were weighed to the nearest 0.1 gram in an NIH metabolic kitchen, so the comparison was against a true value, not another estimate.
- This was a conference abstract, not a peer-reviewed paper, so treat the exact figures as preliminary.
What the study found
The work came out of a larger diet study at the NIH Clinical Center comparing how the body handles a low-carbohydrate ketogenic diet versus a standard diet. Meals for that trial are made in a controlled metabolic kitchen where, as the American Society for Nutrition release describes it, ingredients are weighed to the nearest 0.1 gram.
The researchers took standardized photos of 102 of those meals and ran the images through MyFitnessPal, LoseIt!, CalAI and Appediet. All four came in low, by about 250 to 345 calories and roughly 30 grams of fat per meal.
Not everything failed equally. All four apps estimated carbohydrates more consistently than the other macronutrients. MyFitnessPal and LoseIt! did better on higher-calorie meals than on lower-calorie ones.
The team then looked at more than 200 additional meals to find out what drives the error. Preliminary results point to low-carb ketogenic meals as the hardest case, likely because those meals carry more fat and fat is what the apps consistently miss.
“People using a photo-based tracking app without adjusting the portions or entering the amounts of food should take the results with a grain of salt,” said Aaron Hengist, a postdoctoral visiting fellow at NIDDK, in the release. “These apps tend to underestimate calories, especially from fats, so what they actually ate is likely higher than what the app shows.”
Olivia Charles, a postbaccalaureate research fellow at NIDDK, presented the findings during the President’s Oral Session.
Dr. Kumar’s take
Here is the arithmetic nobody else is doing. If you photograph one meal a day and the app is short by 250 to 345 calories, that is roughly 91,000 to 126,000 calories a year that never appear in your log. Photograph three meals a day and it triples. That explains a complaint I hear constantly: people track carefully, the app says they are in a deficit, and the scale does not move. In many cases the deficit was never real.
The error is not random either, and that is what makes it worse. It runs in one direction, and it lands hardest on fat. The people most exposed are the ones who tend to trust these tools most: the keto and high-protein crowd, who eat the meals the apps read worst. Butter melted into a ribeye is close to invisible to an image model, and oil absorbed into a vegetable looks like a plain vegetable. Carbohydrates tend to be discrete and countable in a photo, which is probably why they held up best here.
Two honest caveats. This was presented as a conference abstract, and abstracts at NUTRITION 2026 are selected by expert reviewers but have not gone through the peer review a journal requires, so the exact spread may shift when a full paper appears. And these meals came from a research kitchen, not a restaurant, so I would not assume the numbers transfer cleanly to a burrito of unknown construction.
What I do not think will change is the direction. Every app tested was low, on the nutrient that is hardest to see.
What it means for you
Use the photo as a starting point, not an answer. The app’s number is closer to a floor than an estimate.
If a meal is high in fat, meaning oils, butter, cheese, nuts, fatty cuts of meat, correct the entry by hand. Adjusting portions or typing in amounts is the specific fix the researchers pointed to.
A kitchen scale for a few weeks will teach you more than a year of photographs. You do not have to weigh forever, just long enough to see how far off your eye is.
Simplify what you have to track. There is evidence that eating the same meals every day helps people lose more weight, partly because a familiar plate is easy to log accurately.
And if you have been tracking diligently without results, consider this before you conclude your metabolism is broken. The likelier explanation is that your log has been quietly missing a third of your food.