There is an app for everything, including healthy dieting, but a UVA expert finds quirks in some AI-powered programs and offers advice on how to use them.
Accurately tracking what you eat is notoriously difficult. Still, the rapid rise of AI-powered apps that could simplify the process intrigued Sibylle Kranz, a researcher at the University of Virginia School of Education and Human Development.
A registered dietitian, Kranz signed up for free trials of several popular apps, creating different user profiles and health goals to test features and compare advice. The results proved to be a mixed bag of convenience and possible inaccuracies.
Kranz, an associate professor, studies how technology such as AI, mobile apps and wearable devices supports healthier eating habits. She recently was named an Excellence in Nutrition Fellow of the American Society for Nutrition for her contributions to nutrition science and public understanding.
While diet-tracking apps have existed for decades, Kranz said generative AI has transformed them. Instead of manually logging meals, users can scan a barcode, say what they ate or snap a photo of their meal for instant analysis of portion size and nutrients. Some also offer custom meal plans and nutrition coaching.
Compared with a handwritten food diary or scheduling regular appointments with a dietitian, AI tools make meal tracking easier and more accessible, Kranz said. That can help people stay motivated, consistent and successful at reaching goals.
As the apps learn users’ habits, they can recognise subtle patterns and tailor recommendations in ways that generic meal plans, or even a professional dietitian, can’t. “Over time, it’s almost as if the AI knows what you have in your pantry and in your fridge,” she said. “It knows things about you that you don’t know about yourself.”
The technology may make it easier for people to be honest about what they eat. “Eating and drinking is so personal,” she said, noting that someone uncomfortable sharing such intimate details with a clinician may be willing to share with an app.
That personalisation, however, comes with trade-offs. To generate custom recommendations, apps collect extensive personal information, including eating habits and location data, body measurements and even emotional well-being. “Unless the user carefully reads the data protection policies, they don’t really know what the company is doing with all this data. That’s a big privacy risk,” Kranz said.
Accuracy is another concern, Kranz said. Recent studies have found that AI sometimes underestimates calorie counts or provides incorrect dietary advice. Her own testing of apps backed up those findings.
“I could give you dozens of examples where I took a photo of something I ate, and then I had to correct it because it misinterpreted what’s there. Although AI has come a long way, it’s still a guessing game,” she said.
Some apps rely on crowdsourced databases built from user-submitted entries, which can create inconsistent or inaccurate information. As more users input more data, the recommendations continue to shift. “I experienced that myself, where I took a photo of something it recognised, confirmed it, and put it in the database. A week later, I had the exact same thing, and it identified it differently or changed some of the ingredients. The accuracy is just not there,” she said.
Despite those limitations, Kranz believes AI can be a valuable supplement to professional nutritional guidance, but not a replacement. “You have the diagnosis, you have the expert evaluation, and then you use this tool to support that,” she said.
For people interested in trying an AI nutrition app, Kranz recommends carefully reviewing privacy settings and choosing an app that lets you limit the personal information you share. If your goal is to identify eating patterns, you don’t need to provide every detail it asks for.
She also recommends verifying the app’s accuracy by comparing its recommendations with trusted nutrition sources. She recommends paying attention to whether it asks follow-up questions before making assumptions and giving the app enough time to prove itself by committing to a weeklong test. “Be 100 percent committed and track everything—every piece of gum, every glass of water. And then look at how happy you are with the outcome,” she said.
Exploring AI as a helpful assistant is a good idea, Kranz said, but keep a healthy dose of skepticism. “At the end of the day, they want you to continue using the app. Better yet, they want you to buy the subscription,” she added.
