A new study reveals that popular AI tools may be providing incomplete and unbalanced dietary advice to adolescents, raising important questions about whether these technologies are ready to guide growing bodies without expert supervision.
Study: Artificial intelligence meal planning underestimates nutrient intake in adolescents compared to nutritionists. Image credit: ilona.shorokhova/Shutterstock.com
Artificial intelligence (AI) is increasingly being used to plan meals for adolescents, but a new study suggests it may fall short of expectations. The survey results are Frontiers of nutritionfound that AI-backed recommendations may consistently underestimate the nutritional intake needed by adolescents.
Rising obesity among adolescents is driving demand for accessible dietary advice
Overweight and obesity rates among adolescents are rapidly increasing globally, and will affect approximately 390 million adolescents in 2022. In fact, it is now being reported as a major form of malnutrition in several regions. Excess weight is associated with multiple negative health effects, including type 2 diabetes, abnormal blood cholesterol, high blood pressure, and sleep apnea. These young people are more likely to be obese as adults, which reduces their quality of life.
Adolescents are also more likely to have body image concerns and a desire to lose weight, including potentially dangerous methods such as vomiting after meals and excessive use of laxatives.
Dietary modifications are key to improving your child’s health in this area. A dietitian is a medical professional who designs and oversees individualized nutritional plans according to established guidelines. However, their services are not always available and their heavy workloads may mean that young people do not receive the dietary advice and follow-up they need.
AI-based tools such as chatbots have been used to overcome these limitations, but only a handful of studies have evaluated their role in adolescent nutrition. Similarly, large-scale language models (LLMs) such as ChatGPT can provide useful information to support nutrition planning, but they have important limitations.
Existing research shows that they may not meet safety standards or international nutritional recommendations, especially under real-world conditions. Additionally, AI tools are unlikely to provide the same level of patient-tailored service as dietitians. However, most of this evidence is based on adult studies or clinical cases.
The current study sought to directly compare AI-generated meals to meals individually prepared by nutritionists for overweight or obese adolescents. Areas of comparison were energy and nutrient content, safety, and feasibility. This comparison could indicate whether AI chatbots can be used in place of nutritionists in nutritional planning for this category of patients, or as an adjunct under the nutritionist’s supervision.
Researchers compared five AI tools to a nutritionist’s plan
The researchers used five AI models (ChatGPT-4o, Gemini 2.5 Pro, Claude 4.1, Bing Chat-5GPT, and Perplexity) to generate 60 diet plans over two sessions. Three-day meal plans were created by each model in response to prompts using four standardized adolescent profiles: overweight or obese boys and overweight or obese girls.
These were compared to a reference daily meal plan created by a nutritionist for each profile. This was in accordance with nutritional recommendations such as: Energy allocation was 45-50% from carbohydrates, 30-35% from fats, and 15-20% from proteins.
The researchers then analyzed each plan’s energy and macronutrient content.
AI diet underestimates energy and macronutrients
The results revealed a consistent and potentially concerning pattern. The AI model included less energy and macronutrients than the nutritionist included in the plan. Energy deficit was 695kcal, protein was 20g, fat was 16g, and carbohydrates were 115g. The potential energy gap may have important clinical implications, especially considering the high energy demands of adolescents.
The authors suggest that given this typical fat oversupply and reduced carbohydrate content, LLM may rely more on common diets such as the ketogenic diet than on scientific guidelines that describe a low-carbohydrate, high-fat approach. This can disrupt growth, metabolism, and cognitive development during this critical developmental stage. Therefore, the long-term safety of such recommendations is unproven.
Five models recommended protein content up to 23.7% and fat content up to 44.5%. Both were above levels recommended for adolescents. In contrast, carbohydrates made up up to 36.3% of the diet, which was lower than recommended levels.
Dietitian plans contained 44% to 46% carbohydrates depending on profile. The protein percentage varied between 18% and 20%, and the fat between 36% and 37%. Overall, these plans are consistent with national recommendations.
The authors note that “this pattern shows a systematic shift toward a low CHO, high protein, and high fat diet structure across all AI models, indicating that not only the gram-based nutrient amounts but also the macronutrient balance is significantly disrupted in the AI-generated plans.”
The composition of micronutrients varies greatly between AI-generated meals; There was significant variation between models and compared to nutritionist standards. This may contribute to micronutrient deficiencies in adolescents and indicates that these plans may not yet be suitable for clinical use without expert supervision. No model closely followed a dietitian’s standard diet across all nutrients.
The authors note that this is the first time that different LLMs have been compared regarding the nutritional needs of adolescents, with a detailed assessment of macronutrients as well as multiple micronutrients. As previous research has suggested, this may indicate a lack of AI technical expertise in this area. This can hinder accurate estimation of energy and macronutrient composition in AI-generated personalized meal plans.
strengths and limitations
This study has several strengths. We evaluated five different AI models to enhance the robustness and comparative power of our analysis. By creating a three-day meal plan, the researchers were able to assess consistent patterns rather than isolated anomalies, making the results more reliable. Using a dietitian-designed plan based on international dietary guidelines provided a reliable and clinically relevant reference standard. Furthermore, comprehensive assessment of macro- and micronutrients has enabled detailed and multidimensional assessment of diet quality.
Despite these advantages, this study also has limitations. The results may only apply to the specific AI model tested, which is continually evolving, and some potentially relevant information may be missing from standardized youth profiles, limiting personalization. Statistical approaches, including the use of multi-day average outputs, can affect independence and variability estimates of results. Additionally, this study relied on simulated scenarios rather than real-world adolescent behavior, which may limit ecological validity. Finally, the use of standardized prompts in a single language may limit the generalizability of the findings to other populations and settings.
Risks of unsupervised AI nutrition advice
“The AI models showed clinically significant deviations in the adolescents’ dietary plans at both the macro and micro levels.” They consistently recommended diets with lower energy and carbohydrate content than dietitian-designed diets.
Until these gaps are addressed, the authors caution that AI-generated meal plans should not replace professional dietary guidance for adolescents.
Click here to download your PDF copy.
