Can Social Media Algorithms Exacerbate Eating Disorders?

A new study analyzed the algorithms of people with and without an eating disorder.

Reviewed by Abigail Fagan

Key points

• The TikTok algorithms of people with an eating disorder deliver strikingly more problematic content.
• Greater severity of eating disorder symptoms was related to more problematic TikTok algorithms.
• TikTok algorithms could play a role in maintaining eating disorders or impeding recovery.

It’s a situation many of us know all know too well: You find yourself endlessly scrolling through social media, only to realize that much more time has passed than you realized, or intended to spend on social media in the first place! Nowadays, what we see on our social media feeds is determined by algorithms (i.e., sets of rules and conditions), and their primary aim is to increase the amount of time we spend there.

Algorithms use many types of information to learn how to get us hooked for longer. This includes not only the obvious, such as what we are looking at and whether we are “liking” and sharing it, but also details such as how long we linger on a post, how quickly we scroll through our feed, whether we replay a reel, and much more. The algorithms of social media companies may even evolve to such a degree that they become “black box algorithms” that are so complex, that not even the developers themselves understand how they work.

In a new study, researchers at the University of Melbourne wanted to look more closely at these powerful algorithms and how they might be related to our mental health. In particular, they investigated the potential role of social media algorithms in the maintenance of eating disorders.

The Research Design

To do so, the researchers recruited 112 Australian adults, about half of whom currently had an eating disorder. Though algorithms are used by all social media platforms, this study focused on TikTok because it is one of the most popular platforms among young people, who may be most vulnerable to developing an eating disorder.

All participants were asked to obtain their user data from TikTok, which was sent to the research team. Using these data, the researchers could identify what types of videos the participants saw on TikTok over the previous month, as well as whether the participants “liked” the video. Namely, the researchers examined four categories of videos that are relevant to eating disorders:

1. Appearance-related videos (e.g., #makeuptips, #beautychallenge);
2. Dieting videos (e.g., #cleaneating, #lowcarb);
3. Exercise videos (e.g., #workout, #gymmotivation);
4. “Toxic eating disorder” videos, which explicitly encouraged disordered eating (e.g., #thinspo, #proANA”).

In addition to providing their TikTok data, the participants also completed questionnaires to assess their current levels of eating disorder symptoms.

Using all of these different data, the researchers could then conduct statistical analyses to investigate the relationships between the participants’ TikTok algorithms and symptoms of eating disorders.

The Key Findings

First of all, there were no differences between participants with and without an eating disorder in terms of how much they used TikTok (e.g., duration or frequency of use). However, there were striking differences in terms of what content they were exposed to via their TikTok algorithms.

Namely, the algorithms of participants with an eating disorder delivered dramatically more appearance-related videos (+146%), dieting videos (+335%), exercise videos (+142%), and “toxic eating disorder” videos (+4343%).

In addition, participants with more severe eating disorder symptoms had TikTok algorithms that were significantly biased toward delivering more appearance-related videos, dieting videos, exercise videos, and “toxic eating disorder” videos.

Importantly, although participants with an eating disorder were more likely to “like” these types of videos, this likelihood was far eclipsed by much higher rates of delivery of these videos by their algorithms. The researchers even identified a subset of participants who appeared to be in an eating disorder echo chamber. This meant that their algorithms delivered over 1,000 appearance-related videos over the past month, and that a majority of the videos shown to them were appearance-related.

The Take-Home Messages

To conclude, the findings suggest that algorithms could play a role in exacerbating eating disorder symptoms, with people with an eating disorder having algorithms that deliver strikingly more problematic content. The fact that these delivery rates far exceed the potential role of active engagement with content (e.g., “liking” behavior) suggests that algorithms may be preferencing less conscious viewer actions.

Of course, the present study cannot determine the causal role of TikTok algorithms in contributing to an eating disorder, and the relationship is likely to be complex. Yet, as the authors of this study described:

“Our concern is that a vulnerable person may spend slightly more time watching a problematic video, or delay skipping over a problematic video, because it touches upon insecurities and briefly intensifies anxiety or other negative mood states (e.g., guilt) and/or induces an unfavorable appearance comparison. If the algorithm is sensitized toward less volitional, more passive user actions like slightly longer video watch times, and desensitized toward more volitional, more active user actions like “liking” a video, then a user with an eating disorder may consequently be delivered more of the videos that intensify their eating disorder psychopathology despite not having “liked” them.” (p. 9)

One important implication of this study is that people with an eating disorder—and their clinicians—may be unaware or under-aware of the potential impact of their social media algorithms in exacerbating their eating disorder and/or impeding their recovery. To address this, the authors recommend social media literacy training, and taking active steps to tweak algorithms such as by using the “not interested button” on TikTok. Ultimately, however, the authors underscore the necessity of systemic changes at the legislative and policy levels, for example to ensure that TikTok takes greater responsibility in protecting its users.