Why the NIH Tested Photo-Based Calorie Trackers

People who rely on AI-powered photo apps to log their meals may be consuming hundreds more calories than they think, according to a study from the National Institutes of Health. Researchers tested four widely used apps—MyFitnessPal, LoseIt!, CalAI and Appediet—against 102 meals prepared in a tightly controlled metabolic kitchen, where ingredients were weighed to the nearest 0.1 gram. On average, the apps underestimated the meal's calorie total by between 250 and 345 calories, and fat by roughly 30 grams. The findings were presented at NUTRITION 2026, the annual meeting of the American Society for Nutrition.

The study took advantage of an ongoing NIH clinical trial that compares a standard diet with a low-carbohydrate ketogenic diet. Every meal was known exactly, giving researchers a rare gold-standard reference. They photographed each plate and ran the images through the four apps, which use AI image recognition to identify foods, estimate portion sizes, and calculate nutritional content from databases. The gap between the apps' estimates and the actual values was consistent across all four tools, though they performed somewhat better on higher-calorie meals and were more accurate for carbohydrates than for fats.

“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 at the National Institute of Diabetes and Digestive and Kidney Diseases. “However, the accuracy of many of these apps has not been thoroughly evaluated. Our study helps address this question.” Early data from an additional 200 meals point to the same trend, with ketogenic, high-fat dishes causing the most trouble.

Where the Apps Fall Short – And Why It Matters

Why fat trips up the AI

The core challenge is that fat is harder to see than carbohydrates. A plate of grilled chicken and avocado may look ordinary, but the calorie density of fats—oils, butter, nuts, dressings—is often invisible in a photo. The apps' image recognition can reasonably gauge the volume of a food, but small, high-fat additions or hidden fats in sauces and cooking oil cannot be detected. Since fat delivers nine calories per gram versus four for carbs or protein, even modest underestimates compound quickly. The 30-gram fat gap the study found translates to about 270 missing calories from fat alone.

The consequence for weight management

A 345-calorie daily error—roughly equivalent to a large latte or a small sandwich—can undermine weeks of dietary effort. For someone aiming for a modest 500-calorie deficit, such an unrecorded surplus could halve expected weight loss. The apps are not universally inaccurate; they performed better on higher-calorie meals, simply because larger portions are easier for AI to discern. But the error is systematic enough that relying solely on photo logging without verifying servings may give users a false sense of control.

What it means for the app developers

The study does not directly critique the business models of MyFitnessPal, LoseIt! or the smaller apps, but it places pressure on their core value proposition. Accuracy is the whole point. If the technology systematically undercounts, users may lose trust or abandon the tools. For developers, the finding highlights a need to invest in better fat-detection algorithms, perhaps by incorporating metadata or multi-angle images. It may also open a door for hybrid solutions that prompt users to manually confirm high-fat ingredients, blending photo capture with structured input. The companies have not yet publicly responded to the findings.

What Users of AI Calorie Apps Should Do Now

  • Don't trust the photo total alone. Use the app's manual entry or barcode scanning for high-fat items (oils, butter, nuts, avocado, dressings) whenever possible. A quick manual adjustment can close much of the gap.
  • Be extra cautious with keto and high-fat meals. The study found the largest errors on ketogenic plates. If you're following a low-carb, high-fat plan, supplement photo logs with weight or measuring-cup checks for fats and oils.
  • Check serving sizes after each scan. The AI often guesses portion size incorrectly. If the app suggests a generic “medium avocado,” override it with the actual weight or a closer estimate.
  • Use the app’s trend, not its daily number. A consistent undercount across meals means your weekly deficit may be less than reported. Instead of obsessing over a single day’s total, track how your weight and measurements change over time and adjust your target if progress stalls.
  • Combine photo logging with a traditional food journal or a dietitian’s review if precision is critical, such as in a medically supervised weight loss program or when training for athletic performance.