Calorie counting apps that rely on AI allege results far from artificial. New research from the American Society for Nutrition, however, suggests they tend to severely underestimate how many calories are actually being consumed.
The apps in question purport to analyze the nutritional content of a meal by using a single photo.
Most calorie counting apps ask that you input the food yourself — whether it be by searching for an item, scanning a barcode or filling out the food’s macros manually. When compared with snapping a photo, it can feel like it’s much easier to go the AI-assisted app route.
But the discrepancies between the number of calories you think you’re eating versus the ones you are could make it difficult to have an accurate read.
In the research project, team members took more than 100 photos of meals cooked in a controlled metabolic kitchen — where foods were measured with precision, to the nearest tenth of a gram. Then, they uploaded them to four common calorie counting apps that now offer AI-assisted estimation.
The apps missed between 250 to 350 calories per meal. Calories from fat, in particular, was underestimated by about 30 grams. Two apps were better at analyzing meals with high calorie counts than low ones, and all four apps were most accurate at counting carbohydrates over other macros.
The research project was a part of a larger National Institutes of Health study exploring how the body processes nutrients on different kinds of diets.
Now, no one is saying you can’t keep uploading photos of your treats and eats to the apps … but you might want to add in your standard calorie counting methods, too.
After all, a picture could be worth a 1,000 calories.