EDTok: Mapping the Hidden World of Eating Disorders on TikTok
Hashtags like #ednosheeran or #edr3c0very might look like typos to a casual observer scrolling TikTok. For a teenager struggling with their relationship with food, however, these are secret doorways into a digital subculture that has exploded in size since 2020. A new study has now mapped this hidden geography, unveiling "EDTok," a vast and complex ecosystem.
This research proves social media isn't just reflecting eating disorder (ED) trends—it is actively hosting a complex, high-stakes ecosystem. Here, 1 billion users, mostly adolescents, navigate the thin line between recovery and relapse.
The Scope of the Study
Researchers compiled a massive dataset to analyze this phenomenon over a five-year window.
The EDTok Dataset
- 43,040 Videos analyzed
- 577,071 Comments examined
- 537,329,790 Total Views across the content
Key Findings: A Polarized Landscape
Using Google Gemini AI to filter content with 99% accuracy, the study identified ten primary thematic clusters, revealing a starkly polarized emotional landscape.
The "Broken Heart" of the Data
Content volume spiked dramatically following the March 2020 lockdowns. This surge led to a pattern researchers termed "Radicalization by Thinness," where social isolation collided with internalized body ideals.
Emotional Extremes
- The Dark Side: The "Fear/Anxiety" topic was characterized by 80.00% fear.
- The Protective Side: "Gratitude" clusters were filled with 79.26% joy, and recovery content often sparked waves of supportive comments like "hugs" and "love."
Evasion, Engagement, and Moderation
The study uncovered how users actively evade platform safety measures, creating a persistent challenge.
Bypassing the Filters
To keep "fear food" challenges and triggering imagery alive, users act as linguistic chameleons, using intentional misspellings and coded language (like #ednosheeran for #eatingdisorder).
The Scale of the Challenge
The sheer volume of engagement—79,876,956 likes on this content—suggests TikTok’s current moderation is a game of cat-and-mouse that the platform is not always winning.
A Cautious Note of Hope & Study Limitations
Following a peak between 2020 and 2022, the data shows a decline in ED-related video volume by mid-2023. While this may point to more effective "shadow-banning," researchers remain cautious.
The study also acknowledges its own methodological hurdles.
Limitations of the Research
- Keyword Reliance: Using a specific set of keywords likely missed newer, more obscure slang used to evade filters.
- Data Loss: Technical glitches claimed about 16,000 potential data points during download.
- AI Classification: The AI used to classify videos lacked specific psychiatric training.
Nevertheless, the EDTok dataset provides a vital baseline for understanding a digital world where dangerous health advice is often just one swipe away.
Reference: Bickham, C., Ramirez-Gonzalez, B., Chu, M. D., Lerman, K., & Ferrara, E. (2025). EDTok: A Dataset for Eating Disorder Content on TikTok. arXiv:2505.02250v1 [cs.SI].