DOI: 10.38079/igusabder.1903214 ISSN: 2536-4499

Can Artificial Intelligence Reliably Detect Risky Eating Narratives? A Comparative Analysis of YouTube Extreme Diet Videos

İrem Başören, Hilal Aksoy, Abdullah Furkan Korkmaz, Zehra Margot Çelik
Show PDF Cite
Aim: This descriptive study was conducted to identify characteristics that are related to binge eating in YouTube videos that have been shared with the keywords “extreme diet” and “diet challenge” and to compare classifications made by expert dietitians with those generated by an artificial intelligence model.Method: A total of 49 YouTube videos that met the inclusion criteria were analyzed. The video characteristics were evaluated in relation to the view count, number of likes and comments, video duration, and time elapsed since upload. The quality of information provided in the video was assessed using the Modified DISCERN instrument, JAMA Benchmark Criteria, and Global Quality Score (GQS). A study-specific Binge Eating–Related Scoring System (BERS), consisting of 15 dichotomous items and based on video transcripts, was independently applied by both expert dietitians and the Gemini 3 Pro artificial intelligence model. In addition, 89,240 user comments were analyzed using R-based sentiment and emotion analysis techniques. Google Trends data for 2021–2026 were also evaluated. Statistical analyses were performed using SPSS 22.0.Results: The median number of views was 2,043,167 (IQR: 1,043,285–5,550,269). AI-based BERS scores showed significant positive correlations with the number of views (r=0.471), likes (r=0.453), comments (r=0.423), and video length (r=0.543) (p<0.05). A moderate positive correlation was found between expert-rated and AI-generated BERS scores (r=0.577, p<0.001). Higher binge eating–related scores were negatively correlated with educational quality scores (GQS). Sentiment analysis revealed that 41.3% of comments were positive, and the most frequently identified emotions were trust and joy. Google Trends indicated a peak in “extreme diet” searches in 2026.Conclusion: Highly engaging YouTube content often not only normalizes extreme eating behaviors but is also received positively by viewers. AI proved to be moderately reliable for identifying binge eating-related features, thus suggesting its potential as a scalable screening tool for monitoring harmful health trends. Interdisciplinary collaboration is required to ensure that AI-powered content analysis aligns with public health priorities.

More from our Archive