Are AI Photo Calorie Tracking Apps Accurate? What a New NIH Study Found

Taking a photo of your plate and getting calories in seconds is the biggest change in food tracking in years. But how close are those numbers to what you actually ate? A new test from researchers at the U.S. National Institutes of Health gives some of the first hard data.

What the NIH-led study found

Researchers at the National Institute of Diabetes and Digestive and Kidney Diseases tested four photo-based apps (MyFitnessPal, Lose It!, Cal AI and Appediet) on 102 meals that had been precisely weighed for an NIH diet study, then tested more than 200 more (ScienceDaily / American Society for Nutrition). Key results:

• Calories were underestimated by roughly 250 to 345 per meal on average across the apps.
• Fat was underestimated by about 30 g per meal, and overall estimates were about one-third too low for calories and fat.
• Estimates were more consistent for carbohydrates than for other nutrients.
• High-fat (keto-style) meals were especially hard for the apps to measure.

Why photo estimates miss calories

• Hidden fats: oil, butter and dressings soak in and are nearly invisible in a photo, yet a single tablespoon of oil is about 120 calories.
• Portion depth: a photo is 2D. It's hard to tell a thin layer of rice from a deep bowl of it.
• Mixed dishes: stews, curries, casseroles and sandwiches hide what's inside.
• Look-alike foods: full-fat and low-fat versions of yogurt, cheese or milk look the same.

So is tracking still worth it?

Yes. A systematic review of 59 behavioural weight-loss studies found that programs using dietary self-monitoring generally produced more weight loss than comparison groups, and 12 of the 18 studies that linked tracking adherence to results found a significant positive relationship (Raber et al., 2021).

In other words, the habit of tracking matters, and faster tools make it easier to keep that habit. The trick is knowing where to double-check.

7 ways to make photo tracking more accurate

1. Always review the estimate before saving it. Adjust portion sizes rather than accepting the first guess.
2. Add cooking oil, butter, sauces and dressings separately. They're the most common misses.
3. Use a barcode for packaged food. Label data is far more precise than a photo.
4. Use official restaurant nutrition when a chain publishes it, instead of a photo guess.
5. Weigh your staples once or twice, like your usual bowl of rice or scoop of peanut butter, so you can spot bad estimates. Our guide to counting macros explains why this pays off.
6. Photograph before you eat, from above and with the whole plate in frame.
7. Watch your weekly weight trend. If you're tracking a deficit but not losing weight over a few weeks, you're probably undercounting. Here's how to check and adjust your calorie deficit.

How Forma approaches photo logging

In Forma, photo results are shown as estimates you confirm, not facts: you can change any food or portion before logging. For packaged food, Forma uses barcode label data, and for eating out it includes official nutrition published by chains like Tim Hortons, A&W, Starbucks and Mucho Burrito, so you only need a photo when there's no better source. See how that works at Tim Hortons.