Instagram Study Links Body-Image Harm to Targeted Content

• A Reuters-reviewed internal Meta study found teens who felt worse about their bodies saw substantially more eating-disorder–adjacent posts than their peers.

Research Findings

Meta researchers manually sampled content seen by 1,149 teens during the 2023–2024 school year and compared what groups encountered over three months. The analysis shows that the 223 teens who frequently reported feeling bad about their bodies were exposed to eating-disorder–adjacent content at a rate of 10.5 percent of their feed. By contrast, other teens in the sample encountered such material at a 3.3 percent rate. The posts flagged by researchers commonly emphasized chest, buttocks or thighs, included explicit judgement of body types, or contained reference to disordered eating and negative body image.

Researchers also found that teens reporting the most negative feelings saw more content Meta tags as mature themes, risky behavior, harm and cruelty, and suffering. That set of categories made up 27 percent of what vulnerable teens viewed, compared with 13.6 percent for their peers who did not report similar feelings. Meta’s internal report stops short of claiming causation and explicitly notes it cannot determine whether teens seek harmful content or are driven to it by the platform’s recommendations. The document nonetheless highlighted that Meta’s own advisors and external experts have urged limits on the types of content examined.

Platform Detection Limits

Meta reported that its rule-based screening tools missed roughly 98.5 percent of the sensitive content the researchers were studying. The company has only recently begun developing an algorithm specifically to detect the types of material flagged in the study. Until those tools are mature, detection will remain uneven and depend heavily on manual review or ad hoc methods. The research team described that limitation as unsurprising but operationally significant.

Meta spokesperson Andy Stone said the study reflects the company’s ongoing efforts to understand young users’ experiences and to improve safety features on its platforms. Stone pointed to earlier steps Meta announced to align content shown to minors with PG-13 standards and to cut age-restricted content shown to teenage accounts roughly in half since July. The company faces ongoing investigations and legal challenges in the United States over Instagram’s effects on children and allegations of harmful product design. Those external pressures have repeatedly referenced prior leaked internal research suggesting algorithmic recommendations can amplify harmful content for vulnerable users.

Expert Take

Independent reviewers consulted by Reuters described the study’s methodology as robust and its findings worrying. University of Michigan pediatrician Jenny Radesky said the results support the idea that Instagram profiles vulnerable teens and amplifies harmful material into their feeds. Other child-development experts who have reviewed similar internal documents have warned that feed-driven consumption patterns differ from deliberate search behavior and can produce repetitive exposure. Those patterns are especially concerning for young people already reporting body dissatisfaction.

The researchers included examples of material they considered borderline or sensitive, such as images of very thin bodies in minimal clothing, fight videos, and posts depicting self-harm or severe emotional distress. Some samples showed graphic or disturbing imagery that did not clearly violate platform rules yet prompted a “sensitive content” flag from the study authors. Advisory groups inside and outside Meta, including the company’s Eating Disorder & Body Image Advisory Council, have recommended restricting how much of that content teens see. The study’s internal summary documents these recommendations while noting technical and policy challenges to acting on them immediately.

Router-style or network-level protections cannot address recommendation-driven exposure because the issue lies primarily with what the algorithm surfaces rather than with device-level threats. Public-health approaches that combine content moderation, age verification, parental controls and mental-health resources may reduce harms more effectively than single interventions. Research into algorithmic transparency and independent audit trails for recommendation systems is expanding, and future regulation could require firms to publish impact assessments for youth-facing algorithms. Monitoring the gap between detection capability and advisory demands will be essential for any policy or engineering response.