Researchers tested three popular artificial intelligence systems (ChatGPT, Grok, and DeepSeek) to see if they could create healthy meal plans for people with type 2 diabetes. Using pretend patient profiles instead of real people, they asked each AI to create 3-day meal plans and checked if the meals matched official health guidelines. While the AI systems showed some promise in creating basic meal ideas, none of them worked as well as a real dietitian would. The study found that AI sometimes made mistakes and didn’t fully consider each person’s unique health situation. Before doctors could use these AI systems to help patients, they would need significant improvements and expert supervision.

The Quick Take

  • What they studied: Whether three different AI computer programs could create appropriate meal plans for people with type 2 diabetes that follow official medical nutrition guidelines.
  • Who participated: The study didn’t use real people. Instead, researchers created 24 pretend patient profiles with different combinations of gender and body weight categories, and asked the AI systems to create meal plans for these fictional patients.
  • Key finding: ChatGPT-4.1 got the calorie amounts right about 71% of the time, while Grok-3 was more accurate at 83%. However, all three AI systems made mistakes with specific nutrients and didn’t fully follow the complete medical nutrition care process that real dietitians use.
  • What it means for you: AI meal planning tools may help provide basic nutrition ideas, but they’re not ready to replace talking with a real dietitian, especially if you have diabetes or other health conditions. These tools need more development and expert review before they should be used in actual medical care.

The Research Details

This study used a simulation-based approach, meaning researchers created fictional patient scenarios rather than working with real people. They developed 24 different pretend patient profiles that varied by gender and body weight categories. For each pretend patient, they asked three different AI systems (ChatGPT-4.1, Grok-3, and DeepSeek) to create 3-day meal plans, all in the Turkish language. The researchers then carefully checked each meal plan to see if it matched official diabetes nutrition guidelines and included the right amounts of calories, proteins, fats, carbohydrates, and vitamins.

The study compared how well each AI system followed established medical nutrition guidelines for diabetes care. The researchers looked at whether the meal plans had the correct number of calories, the right balance of nutrients, and whether they followed the complete nutrition care process that professional dietitians use when helping patients.

This research approach is important because it allows scientists to test AI systems in a controlled way without risking harm to real patients. By using pretend patient profiles, researchers could test many different scenarios quickly and fairly. This helps identify problems with AI meal planning before it’s ever used with actual patients. The study also shows what improvements are needed before AI could safely support real medical care.

This study has some important limitations to understand. It only tested AI systems with pretend patients, not real people, so the results may not reflect how well these systems would work in actual medical practice. The study was conducted in Turkish, which may affect how well the findings apply to other languages. The researchers were careful to note that they were only checking if the AI matched written guidelines, not whether the meal plans would actually help real patients get healthier. The study clearly states that these results should not be seen as proof that AI can replace dietitian care.

What the Results Show

When researchers checked the meal plans for calorie accuracy, ChatGPT-4.1 got it right about 71% of the time, while Grok-3 was more accurate at 83%. However, each AI system had different problems. ChatGPT-4.1 often included too much fat in the meal plans. Grok-3 sometimes didn’t include enough of certain vitamins and minerals. DeepSeek adjusted the amount of protein based on body weight, which was good, but it didn’t include enough carbohydrates.

None of the three AI systems fully followed the complete nutrition care process that professional dietitians use. This process includes diagnosing nutrition problems, planning treatment, providing nutrition education, and monitoring progress over time. The AI systems were mainly just creating meal plans without considering the full picture of a patient’s health needs.

The researchers also found that all three AI systems sometimes made up information or gave answers that didn’t make sense in a medical context. This is called “hallucination” in AI research. The AI systems also didn’t ask questions or gather enough information about each pretend patient to truly personalize the meal plans.

The study found that different AI systems made different types of mistakes. Some were better at calculating calories but worse at including the right vitamins. Others adjusted nutrients based on body weight but missed other important details. The researchers noted that the AI systems didn’t follow a structured approach to nutrition care, which is important for medical treatment. The systems also didn’t explain their reasoning or provide the kind of personalized guidance that a real dietitian would give.

This is one of the first studies to systematically test whether popular AI systems can create diabetes meal plans that match official medical guidelines. Previous research has raised concerns about AI in healthcare, and this study confirms that while AI shows promise for basic tasks, it still has significant limitations in medical nutrition care. The findings align with other research showing that AI systems can make mistakes and need expert oversight in healthcare settings.

The biggest limitation is that this study only tested AI with pretend patients, not real people. The results may not reflect how well these systems would actually work in real medical practice. The study only checked if meal plans matched written guidelines, not whether they would actually help patients. The research was done in Turkish, so results might be different in other languages. The study didn’t test how patients would respond to these meal plans or whether they would actually follow them. Additionally, AI systems change and improve over time, so these results may become outdated quickly.

The Bottom Line

Based on this research, AI meal planning tools should NOT be used as a replacement for professional dietitian care, especially for people with diabetes or other health conditions. If you have diabetes, continue working with a registered dietitian who can personalize nutrition advice to your specific needs. AI tools might eventually help provide basic nutrition information or support between dietitian visits, but only after significant improvements and with professional oversight. Confidence level: High - this recommendation is based on clear study findings showing AI limitations.

People with type 2 diabetes should be especially aware of these findings and should not rely on AI meal planning tools as their primary nutrition guidance. Healthcare providers and hospital systems considering AI tools for patient care should understand these limitations. Developers of AI systems should use these findings to improve their tools. People interested in nutrition technology and AI in healthcare will find this research relevant. However, people without diabetes or serious health conditions may find AI meal planning tools more helpful for basic nutrition ideas, though professional guidance is still preferable.

If AI meal planning tools are improved and eventually approved for medical use, it would likely take several years of development and testing. In the near term (next 1-2 years), these tools are not ready for clinical use. If you’re currently using an AI tool for meal planning, you should expect to see basic nutrition ideas, but you should verify the information with a real healthcare provider. Long-term benefits would only come after significant improvements to how AI systems work with medical guidelines and patient information.

Want to Apply This Research?

  • If using a nutrition app, track the actual nutrients in your meals (calories, carbohydrates, protein, fat, and key vitamins) for one week and compare them to your dietitian’s recommendations. This helps you see if AI-generated suggestions match professional advice.
  • Use an app to log your meals and compare AI suggestions to guidance from your real dietitian. When an app suggests a meal plan, ask your dietitian to review it before you follow it. This combines AI convenience with professional expertise.
  • Over 4 weeks, track how you feel on any meal plan (energy levels, blood sugar if you monitor it, hunger levels) and share this information with your dietitian. This helps determine if a meal plan is actually working for you, regardless of whether it came from AI or a professional.

This research tested AI systems with pretend patient profiles and should not be interpreted as evidence that AI meal planning tools are safe or effective for real patients. If you have type 2 diabetes or any other health condition, you should work with a registered dietitian or healthcare provider for personalized nutrition advice rather than relying on AI tools alone. AI-generated meal plans have not been proven to improve health outcomes and may contain errors. Always consult with qualified healthcare professionals before making significant changes to your diet, especially if you take medications or have chronic health conditions. This study reflects how AI systems performed in a controlled test environment and may not represent real-world performance.