EXAMINATION OF THE ACCURACY AND RELIABILITY OF NUTRITIONAL RECOMMENDATIONS PRODUCED BY ARTIFICIAL INTELLIGENCE-BASED LANGUAGE MODELS FROM A PUBLIC HEALTH PERSPECTIVE

Özet

Artificial intelligence–based large language models (LLMs) are increasingly utilized as accessible sources of nutrition-related information by the general population. This study aims to evaluate the accuracy, reliability, and scientific alignment of nutrition advice generated by widely used LLMs from a public health perspective. Nutrition recommendations produced by ChatGPT and Google Gemini were examined using standardized dietary scenarios representing different health profiles. The AI-generated outputs were subjected to qualitative content analysis and systematically compared with evidence-based recommendations outlined in guidelines issued by the World Health Organization (WHO), the Food and Agriculture Organization (FAO), and national nutrition authorities.

The findings demonstrate that both LLMs consistently emphasize general healthy eating principles and convey information in a clear and user-friendly manner. However, the recommendations predominantly remain qualitative in nature and lack the quantitative specificity required for clinical or individualized nutrition counseling. Key deficiencies were identified in the calculation of individual energy requirements, specification of macronutrient distributions, and inclusion of measurable targets such as daily fiber intake, sodium limits, and sugar thresholds. While alignment with public health messages—such as increased fruit and vegetable consumption, moderation of processed foods, and balanced dietary patterns—was observed, this concordance remained superficial and did not reach the level of clinical applicability defined in established nutritional guidelines.

In conclusion, LLM-based nutrition advice may function as a supplementary tool for disseminating general public health information and promoting dietary awareness. Nevertheless, due to limitations in personalization, quantitative accuracy, and clinical specificity, these systems cannot be considered substitutes for evidence-based, individualized nutrition counseling delivered by qualified health professionals. Clear ethical and methodological boundaries are required to guide the responsible integration of artificial intelligence into nutrition and public health practice.

Keywords: Artificial intelligence, large language models, nutrition counseling, public health, digital health