Your AI Calorie Counter Knows It’s Pasta. It Doesn’t Know How Much

admin
By admin
7 Min Read

AI food-tracking apps make nutrition logging easier than ever. It can tell a bowl of pasta from a bowl of soup in an instant. What it can’t reliably tell you is whether you’re eating 400 calories or 800.

That gap matters because AI food-tracking apps promise to make nutrition logging effortless. Instead of weighing food or typing in every ingredient, you take a picture of your meal and get instant estimates of calories, protein, and carbohydrates.

However, research suggests those estimates can miss broadly, particularly when it comes to portion sizes. Experts say the apps can still help people recognize eating patterns and make healthier food choices, just don’t treat the numbers as exact measurements.

How Far Off Is AI at Counting Calories?

A 2025 study tested three AI models on 52 photos of researcher-prepared meals, individual foods, and a few prepackaged dishes to see how accurately they could estimate calories and portion sizes. ChatGPT-4o and Claude 3.5 Sonnet were off by about 36 percent on average, and Gemini 1.5 Pro missed by even more.

Errors ran in both directions, but one pattern was consistent: All three models underestimated larger portions. In other words, the bigger the meal, the more calories were likely to go uncounted.

The models were tested in 2024, and newer versions are now available. However, more recent research suggests that AI still struggles to accurately estimate portions and calories from food photos.
Independent testing is also thin. A 2025 review found just 13 studies of AI-based dietary tracking, with mixed results. Eight had a moderate risk of bias, often because researchers hadn’t adequately accounted for other factors that could influence the results. The studies also used different methods to measure accuracy, so the results are difficult to compare.

Traditional methods aren’t flawless either. Food diaries and dietary recalls rely on memory and portion estimates, which can also produce errors.

Why Portion Size and Hidden Ingredients Trip Up AI

The core problem, said Erin Hennessy, an associate professor at Tufts University who holds a doctorate in food policy and applied nutrition, is that recognizing a food is only one step in calculating its nutritional content. “‘The AI recognized my meal’ doesn’t mean ‘the AI knows exactly how many calories were in my meal,’” she told The Epoch Times.

Identifying a bowl of pasta is one thing. Knowing whether it contains one cup or three is another. A photograph flattens a three-dimensional meal into two dimensions, which makes volume and weight hard to judge. Camera distance and angle can make a portion look larger or smaller, and an overhead shot may hide how deep the plate of food is.

Portion size isn’t the only challenge. Consider lasagna. An app may recognize it immediately, but a photo cannot show how much cheese, meat, or oil went into the recipe. “Mixed dishes create an additional problem because a photograph shows appearance, not a recipe,” Hennessy said.

Restaurant meals, casseroles, soups, and dishes with sauces can be especially difficult to assess because ingredients that significantly affect calories may be hidden.

AI can also struggle with unfamiliar cuisines or foods that weren’t well represented in the images and information used to train it, Hennessy said. So, if an AI model has seen only a few examples of a traditional dish, it may have trouble recognizing the ingredients or estimating its calories.

Where AI Food Tracking Can Still Help

The main advantage is convenience. Traditional food logging can be time-consuming, requiring people to measure portions, search food databases, and record each ingredient. Snapping a photo takes seconds.

Tracking itself works. Dietary self-monitoring of food intake has long been part of behavioral weight-management programs, and research consistently links it to weight loss, Hennessy said. Recording meals may also help people recognize habits that otherwise go unnoticed, such as regularly skipping vegetables, eating little protein at breakfast, or relying heavily on takeout.

“If AI makes dietary self-monitoring easier and therefore makes someone more likely to continue doing it, the tool can have behavioral value even if its nutrient estimates aren’t perfectly accurate,” Hennessy said.

Tracking isn’t appropriate for everyone, however. Calorie-focused monitoring can increase anxiety or reinforce compulsive behaviors in people with eating disorders or disordered eating, Leslie Bonci, a registered dietitian, told The Epoch Times. She encourages people to look beyond calorie counts and focus on dietary goals, such as eating more protein or fiber. “Calories alone are not an indicator of nutritional adequacy.”

How to Get More Out of an AI Food Tracker

You can improve the odds that the AI numbers are useful:

  • Check and Correct: Review the foods the app identifies and adjust portion sizes when needed. Research suggests that correcting AI-generated food entries can improve accuracy.

  • Photograph Ingredients Separately: Bonci recommends photographing ingredients before combining them, especially for mixed dishes such as stir-fries.

  • Watch Trends, Not Exact Numbers: Hennessy recommends focusing on eating patterns over time rather than treating individual calorie estimates as precise.

  • Be Wary of Black Boxes: Hennessy advises caution with apps that don’t explain how their accuracy was tested or which nutrition databases they use.

  • Measure When It Counts: When accurate calorie or nutrient counts are important, measure portions rather than relying on photographs alone.

AI food trackers should be used as a guide, not a scale. Their convenience may encourage people to track their meals more consistently, and that habit can reveal patterns worth changing. But when exact numbers matter, rely on validated methods.

We had a problem loading this article. Please enable javascript or use a different browser. If the issue persists, please visit our help center.

Source link

Share This Article
Leave a Comment

Leave a Reply

Your email address will not be published. Required fields are marked *