Researchers created a new computer tool called NutriRAG that uses artificial intelligence to understand what people eat when they type their meals into an app. The tool was tested with 77 adults who were trying to lose weight. Some participants ate only during an 8-hour window each day, while others reduced their calories by 15%. The AI system was very good at recognizing different foods from written descriptions, achieving 82% accuracy. The study shows that AI could help people better understand their eating habits and make healthier food choices by automatically analyzing what they eat without requiring them to count calories manually.

The Quick Take

  • What they studied: Whether a new artificial intelligence system could accurately identify and categorize foods that people typed into a diet-tracking app, and whether this technology could help understand how eating patterns changed during different diet experiments.
  • Who participated: 77 adults with obesity living in the Twin Cities area. They were divided into three groups: one group ate only during an 8-hour window each day, another group reduced their daily calories by 15%, and a third group ate normally without restrictions.
  • Key finding: The AI system correctly identified foods 82% of the time when people described their meals in their own words. People who restricted their eating times ate less food at night, while people who reduced calories ate fewer snacks and sugary foods.
  • What it means for you: This technology could make diet tracking easier and more accurate by automatically understanding what you eat without requiring manual entry of every food item. However, this is still new technology being tested, so it’s not yet widely available in consumer apps.

The Research Details

Researchers created a new computer system called NutriRAG that combines two technologies: a database of food examples and a powerful language model (similar to ChatGPT) that can understand and classify foods. The system works by first looking up similar foods in its database, then using that information to help the AI make accurate decisions about what food a person described. The researchers tested this system using real data from 77 people who participated in a 12-week study where they typed their meals into an app. Three different groups tried different eating approaches: limiting eating to 8 hours per day, reducing calories by 15%, or eating normally. The AI system analyzed all the typed meal descriptions to see how well it could identify the foods and understand eating patterns.

This research approach is important because most diet-tracking apps require people to manually search for and select foods from a list, which is time-consuming and often inaccurate. By using AI that understands natural language (the way people actually talk and write), the system can work with whatever description someone types, making diet tracking faster and easier. This could help more people stick with healthy eating plans because the technology removes a major barrier to tracking food intake.

This study has several strengths: it used real data from an actual diet study with people randomly assigned to different groups, the AI system was tested on real-world food descriptions (not just perfect descriptions), and the results showed strong accuracy. However, the study only included 77 people from one geographic area, so results might differ in other populations. The study was published in a respected medical informatics journal, which means it went through expert review. The main limitation is that this is a proof-of-concept study showing the technology works, not yet a study proving it helps people lose weight or improve health.

What the Results Show

The NutriRAG system achieved 82% accuracy in correctly identifying foods from written descriptions, which is significantly better than previous methods. This means that when someone typed a description like ‘scrambled eggs with toast and coffee,’ the system correctly identified these foods about 4 out of 5 times. The system worked well because it combined a database of food examples with advanced AI, allowing it to understand variations in how people describe foods (like ‘OJ’ versus ‘orange juice’). The accuracy was consistent across different types of foods, suggesting the system is reliable for various dietary analyses.

Beyond just identifying foods, the system revealed important patterns in how people ate during the study. People who practiced time-restricted eating (eating only during an 8-hour window) significantly reduced their nighttime eating, which suggests this approach naturally limits late-night snacking. People who reduced their calories by 15% ate fewer snacks and sugary foods overall, indicating they made more intentional food choices. These patterns emerged automatically from the AI analysis without researchers having to manually review thousands of meal entries, demonstrating the system’s value for understanding dietary behavior.

Previous methods for analyzing diet from text required researchers to manually read and categorize each meal entry, which was slow and prone to human error. Some older computer systems could only recognize foods if they were typed in very specific ways. NutriRAG improves on these approaches by understanding natural language variation and working much faster. The 82% accuracy rate is substantially higher than earlier AI food recognition systems, though still not perfect. This represents a meaningful step forward in making dietary analysis more practical for real-world use.

The study included only 77 people from one area (Twin Cities), so results might be different for people from other regions or with different backgrounds. The study was relatively short (12 weeks), so we don’t know if the system works as well over longer periods. The AI system was trained on certain types of foods, so it might not work as well with very unusual foods or cuisines it hasn’t seen before. The study shows the technology works for identifying foods, but doesn’t prove it actually helps people lose weight or improve their health—that would require a different type of study. Finally, the system still made mistakes 18% of the time, so it’s not perfect and would likely need human review in some situations.

The Bottom Line

This technology shows promise for making diet tracking easier and more accurate, but it’s still in the research phase. If you use diet-tracking apps, this suggests that future versions might be able to understand your food descriptions better without requiring you to search through menus. For now, the recommendation is to stay aware of this emerging technology while continuing to use whatever diet-tracking method works best for you. The evidence is moderate that this specific technology will improve health outcomes, as the study focused on whether the AI works, not whether it helps people achieve health goals.

This research is most relevant to people who struggle with diet tracking because it’s tedious or confusing, people who want to understand their eating patterns better, and developers of health and nutrition apps. Healthcare providers interested in digital health tools should also pay attention to this development. People who are not interested in tracking their diet or who prefer not to use technology don’t need to worry about this research. This is particularly relevant for people with obesity or those managing weight-related health conditions, as the study was conducted with this population.

If this technology becomes available in consumer apps, you could see improvements in ease of use within 1-2 years as companies integrate similar systems. However, seeing actual health benefits (weight loss, improved blood sugar, etc.) would take longer—typically 8-12 weeks to see meaningful changes, similar to any diet change. The technology itself works immediately (it identifies foods as you type), but behavioral changes and health improvements take time.

Want to Apply This Research?

  • Track the percentage of meals you successfully log using voice or text descriptions (rather than searching menus) over a 2-week period. Measure how many meals you complete logging in under 2 minutes each. This shows whether AI-powered food recognition actually saves you time.
  • Start describing your meals in natural language (the way you’d tell a friend) rather than searching for exact menu matches. For example, type ‘chicken sandwich with lettuce and mayo’ instead of searching for ‘sandwich, chicken, on wheat bread.’ This helps the AI learn your personal food vocabulary and improves accuracy over time.
  • Weekly, review the foods the app identified from your descriptions and correct any mistakes. This creates a personalized learning profile for your eating patterns. Track trends in your eating times and food types monthly to see if patterns emerge (like whether you tend to snack more at certain times or eat certain foods regularly).

This research describes a new technology for identifying foods from written descriptions, not a medical treatment or health intervention. The study was conducted with adults with obesity and may not apply to all populations. While the AI system showed good accuracy in identifying foods, it is not perfect and should not be used as the sole method for medical nutrition therapy without professional oversight. Anyone making significant dietary changes, especially those with diabetes, heart disease, or other health conditions, should consult with a healthcare provider or registered dietitian before relying on automated food tracking systems. This technology is still in research phases and not yet widely available in consumer applications. The findings suggest potential benefits but do not prove that using this technology will lead to weight loss or improved health outcomes.