Source-linked AI summary
Watching What We Eat: Information Quality and Body Image in Diet-Related YouTube Videos
Maddalena Ghiotti, Daniela Paolotti, Yelena Mejova
TL;DR
Diet-related social media content raises concerns about information quality and body-focused messaging, while much prior research sampled only popular or recommended content. This study assesses YouTube videos using adapted quality and body-content measures and an LLM-based pipeline. Quality and body emphasis were largely uncorrelated, while distinct body-related mentions were associated with views or engagement.
Problem
Prior research often examined only popular or algorithm-recommended diet-related content, despite concerns about its information quality and associations with body dissatisfaction and disordered eating.
Source: §1 · p. 1Method
The study combines daily and popular YouTube video samples with adapted PRHISM, HONcode, and SMEC quality measures, body-related measures, and LLM transcript analysis.
Source: §1 · p. 1, §3.3 · p. 3, §5 · p. 8Results
Quality averaged about 49 out of 100 and was largely uncorrelated with body emphasis; weight measurement and negative body image were associated with views, while positive body image was associated with likes and comments per view.
Source: §5 · p. 8Takeaways & Limitations
Personal storytelling is the dominant theme in the dataset and may complicate official health-authority messages on the platform.
Source: §5 · p. 8Because data collection was restricted primarily to U.S.-seen content, the findings likely do not generalize to other countries.
Source: §5 · p. 9
Abstract
from arXiv · showhide
The widespread use of social media, particularly image- and video-based platforms, has turned them into key sources of both normative and informational content related to health and diet. This may contribute to the development of disordered eating behaviors or, potentially, eating disorders. This study uses mixed-methods analysis applied to 3129 YouTube videos about diet and weight loss in order to quantify the level of risk of low-quality information and heightened focus on the body image. We adapt three quality measurement frameworks from the literature -- PRHISM, HONcode and SMEC -- to the online video context, perform manual annotation of a sample of the data, and design an LLM content characterization pipeline to score the videos on the quality of their content and the focus on the human body. Surprisingly, we find that videos around personal storytelling and mindset & motivation are associated with higher-quality content, whereas supplement reviews (arguably more medically sensitive ones) are not. Further, the body-related mentions of weight measurement and negative body image are associated with an increased viewership, whereas the mentions of positive body image are associated with an increased engagement rate in terms of likes and comments, but not viewership. Worryingly, we find a cluster of videos categorized as "music" which promote the dietary supplements Mitolyn and the injectable weight loss drug Mounjaro. As video-based platforms grow in popularity, particularly among younger audiences, studies such as the one presented here are essential for developing empirically grounded tools to enhance the detection of harmful content and inform more effective moderation practices.
1 Introduction
Diet-related social media content can combine weak information quality with body-focused messaging and is associated with risks including body dissatisfaction and disordered eating. This study assesses YouTube diet content at scale, examining information quality, body focus, content attributes, and engagement.
1 Introduction
- Diet-related social media is widespread, but popular videos may promote rapid weight loss, restrictive diets, or anecdotal advice without expert validation.Content may also prioritize aesthetic appeal over informational accuracy.Source: §1 · p. 1
- Prior studies link diet-related social media exposure to body dissatisfaction and disordered eating, while research often samples only popular or algorithm-recommended content.The study frames this as a gap in understanding the broader diet-related content ecosystem.Source: §1 · p. 1
- The study uses scalable transcript analysis and structured evaluation frameworks to assess YouTube diet content and its potential risks.Its questions address information quality and body focus, content attributes associated with them, and their relationship to engagement.Source: §1 · pp. 1–2
- By combining computational methods with validated assessment standards, the study aims to characterize the informational landscape around diet and health on social media.Source: §1 · p. 2
2 Related Work
Prior research has documented quality concerns in online health and diet content, including video-based media, while also connecting exposure to body dissatisfaction and disordered eating. This study responds to limits in content-analysis scale and coverage by examining a broader video sample with LLM-assisted quality assessment.
2 Related Work
- Health-information frameworks and coding schemes assess credibility, transparency, accuracy, completeness, and source authority, but detailed content analysis is labor-intensive and difficult to scale.Surveys also find reliance on digital sources and difficulty evaluating credibility.Source: §2 · p. 2
- Studies across YouTube, TikTok, and pro-anorexia communities report quality and reliability concerns, while exposure has been linked to body dissatisfaction and disordered eating.Source: §2 · p. 2
- Earlier quality studies often focused on popular videos or shallow large samples, leaving the long tail of posted content less examined.This study combines longitudinal sampling of diet-related posting activity with popular actors and uses LLMs to examine quality.Source: §2 · p. 2
3 Data & Methods
The study combines daily and retrospective popular-video collection to build a dataset of diet-related YouTube videos. It characterizes topics and textual features, then applies adapted quality and body-content measures through manual annotation and an LLM-based scoring pipeline.
3.1 Data Collection
- 3,129 unique videos combine a daily sample of recent uploads with a retrospective sample of popular videos from the same five-month window.The daily sample yielded 2,870 videos with complete metadata; the popular sample added 384 videos.Source: §3.1 · pp. 2–3
3.2 Content Characterization
- Content characterization combines YouTube topic labels, NMF topics from titles, descriptions, and transcripts, and channel-level information.The selected NMF model used 13 topics; surface features included text length, stylistic markers, hashtags, mentions, and links.Source: §3.2 · p. 3
3.3 Annotation for Information Quality and Body-related Content
- Information quality is scored using adapted PRHISM, HONcode, and SMEC principles, averaging applicable ratings and rescaling them to 0–100.Body-related variables adapted from Munro et al. are similarly aggregated into a 0–100 score, with higher scores indicating more body focus.Source: §3.3 · p. 3, Table 1 · p. 4, Table 2 · p. 4
- A 50-video manually annotated validation set informed an automated GPT-4.1 transcript-scoring pipeline for quality and body-related content.The selected zero-shot model achieved Spearman’s ρ=0.669 for information quality and ρ=0.752 for body-related content against manual scores.Source: §3.3 · p. 3
4 Results
The results characterize diet-related videos by their sources and topics, assess their quality and body focus, and relate these measures to engagement. They also identify a cluster of pharmaceutical-promotion videos on channels categorized as music.
4.1 Categories of Diet-related Content
- Individual accounts made up 58% of sources and identifiable commercial accounts 8%; YouTube categories were generic, while NMF topics provided finer-grained content labels.Personal storytelling covered the most videos, followed by metabolism and mindset & motivation; Food channels averaged 1.93 videos per channel, versus about 1.74 for Health and Lifestyle.Source: §4.1 · pp. 4–5, Figure 1 · p. 5
4.2 Information Quality and Body-related Content
- Quality averaged about 49/100 and body score 44/100; their correlation was slight (Pearson r=0.054, p=0.0024), with low-quality videos at both ends of body focus.Manual inspection found low-quality videos promoting diet tricks, while videos with low body scores focused on food and those with high scores explicitly mentioned the body.Source: §4.2 · p. 5
- Medicine & Drugs and Mindset & Motivation had the strongest positive associations with quality, while Trick, Supplement review, and Mounjaro recipe were negatively associated.Duration and description mentions had the largest positive associations among descriptive variables; title length, average word length, and subscriber count were negatively associated with quality.Source: §4.2 · pp. 5–7
- Personal storytelling showed the strongest positive association with body score, followed by Transformation, Workout, and Mindset & Motivation; Metabolism, Recipe, and Supplement review were negatively associated.Shorter words, longer videos, uppercase text, and exclamation marks were also associated with body-related mentions.Source: §4.2 · p. 7
4.3 User Engagement
- Body-related principles such as weight measurement, comparison, and negative body image were mainly positively correlated with views, while positive body image was associated with engagement rate.Quality-score correlations with views and engagement rate were low; personal storytelling was associated with both higher views and engagement rate, whereas several quality principles and topics had differing associations across the two metrics.Source: §4.3 · pp. 7–8, Table 5 · p. 7
4.4 Case Study: Mitolyn and Mounjaro
- Videos promoting Mitolyn and Mounjaro appeared unusually often on music-category channels; inspection found channels repurposed from music posting to pharmaceutical promotion.The same three women appeared across multiple channels, suggesting a single producer; medical review videos on an example channel regularly received over 1K views.Source: §4.4 · p. 8, Figure 4 · p. 8
5 Discussion & Conclusions
The study combines a sample of daily poster activity and popular videos with adapted quality rubrics and LLM assessment at scale. It finds low average information quality, largely independent quality and body-emphasis dimensions, and associations between content features and engagement, while noting important limitations.
5 Discussion & Conclusions
- The study combines daily poster activity and popular videos, adapts existing quality rubrics, and evaluates an LLM approach across thousands of videos.The sample captures both posters’ efforts and the most successful videos over the same period.Source: §5 · p. 8
- 49 out of 100 was the average quality score, and information quality and body emphasis were largely uncorrelated dimensions requiring distinct attention.The dataset displayed a wide range of quality scores.Source: §5 · p. 8
- Weight-measurement and negative-body-image mentions were associated with more views, whereas positive-body-image mentions were associated with higher likes and comments per view.The authors note that passive social media use has been linked to body dissatisfaction and eating concerns, especially in young women, while body-positive content may not have protective effects.Source: §5 · p. 8
- Personal storytelling dominated the dataset and was positively linked with views and engagement, while channel subscriber counts were negatively related to content quality.The authors say further work is needed to evaluate the reach and impact of influencer training and support initiatives.Source: §5 · p. 8
- Music-channel accounts promoted Mitolyn and Mounjaro, while annotators detected AI use in 22% of videos, negatively associated with quality.The authors call for broader review of country patterns and continued tracking of AI’s role in health communication.Source: §5 · p. 8
- The sample primarily reflects content seen by U.S. users and may not generalize to other countries; the LLM scorer underperformed human evaluators, and many correlations were small.The authors suggest adding visual information alongside transcripts to improve detection of body-related focus.Source: §5 · p. 9