AI Diet Advice Falls Short for Adolescents, Study Finds
New research indicates that popular artificial intelligence (AI) tools may provide adolescents with incomplete and imbalanced dietary advice. The study, published in Frontiers in Nutrition, suggests that AI-backed recommendations consistently underestimate the nutritional intake required for growing adolescents.
AI-backed recommendations consistently underestimate the nutritional intake required for growing adolescents.
The global rates of adolescent overweight and obesity are increasing, affecting approximately 390 million adolescents in 2022. This trend drives demand for accessible diet advice. While dietitians offer individualized nutrition plans, their services are not always accessible. This often leads to the use of AI-based tools like chatbots to address these limitations. However, limited studies have evaluated AI's role in adolescent nutrition, and existing research primarily focuses on adults or clinical cases. This indicates that AI tools may not meet safety standards or international nutritional recommendations for this vulnerable population.
Study Methodology
Researchers compared five popular AI models—ChatGPT-4o, Gemini 2.5 Pro, Claude 4.1, Bing Chat-5GPT, and Perplexity—with dietitian-prepared plans. They generated 60 three-day diet plans using these AI models for four standardized adolescent profiles (overweight/obese boy and girl). These AI-generated plans were then compared to one-day dietitian plans that strictly adhered to nutritional recommendations (45–50% carbohydrates, 30–35% lipids, 15–20% protein). The energy and macronutrient content of each plan was thoroughly analyzed.
Key Findings
The study revealed a consistent pattern: AI models recommended less energy and macronutrients than dietitian plans across the board.
- Energy shortfall: 695 kcal
- Protein shortfall: 20 g
- Fat decrease: 16 g
- Carbohydrate decrease: 115 g
AI models often showed a higher protein content (up to 23.7%) and fat content (up to 44.5%) than recommended levels for adolescents, while carbohydrate content (at most 36.3%) was notably lower. This macronutrient imbalance, characterized by lower carbohydrates, higher protein, and higher lipids, suggests AI models may rely more on popular diets like ketogenic approaches rather than established scientific guidelines. Such deviations could significantly affect adolescent growth, metabolism, and cognitive development.
Furthermore, micronutrient composition varied significantly among AI-generated diets and compared to dietitian plans, potentially leading to inadequacies. No AI model consistently aligned with the dietitian's reference across all nutrients. This may indicate a lack of technical expertise in AI models for accurate estimation of energy and macronutrient composition in personalized dietary plans.
Strengths and Limitations
Strengths
The study's strengths include:
- Evaluating five different, widely used AI models.
- Generating three-day plans for consistent pattern assessment.
- Using dietitian-designed plans as a credible reference.
- Offering a comprehensive assessment of both macro- and micronutrients.
Limitations
Limitations noted by the authors include:
- Findings may apply only to the specific AI models tested, which are continuously evolving.
- Standardized adolescent profiles might have lacked some relevant personalization information.
- The statistical approach, using averaged multi-day outputs, could affect result independence.
- Reliance on simulated scenarios rather than real-world adolescent behaviors limits ecological validity.
- Standardized prompts in a single language could restrict generalizability.
Risks of Unsupervised AI Nutrition Advice
The authors caution that AI models exhibited clinically significant deviations in diet plans for adolescents at both macro and micro levels.
They consistently recommended diets with lower energy and carbohydrate content compared to professionally designed plans. These discrepancies highlight a significant concern for adolescents seeking dietary advice from AI tools without professional oversight.
Until these discrepancies are addressed, AI-generated diet plans should not replace professional dietary guidance for adolescents.