Mapping Content-Specific Emotional Effects of Social Media

Social networks expose users to a wide range of emotional experiences, yet the psychological mechanisms underlying perceived and actual emotional harm remain incompletely understood. Across three studies, we developed and validated a psychologically grounded, user-derived taxonomy of content dimensions, and examined their associations with negative emotional and cognitive outcomes. The pilot study, which quantified people’s perceptions about harmful content, revealed that social networks were perceived as more emotionally harmful, particularly in relation to political vitriol and social comparison. Study 1 experimentally exposed participants to real-world Twitter content and found that content features derived from language model analysis predicted increases in hostility, distress, and nervousness. Study 2 employed a controlled laboratory design with participants’ own Instagram feeds, showing that exposure to polarized content was associated with increases in distress, fear, and shorter viewing duration, while content related to social comparison caused a reduction in enthusiasm and alertness. Methodologically, the project uses large language models within a psychologically interpretable framework, moving beyond black-box classification approaches. The findings offer initial insights for content moderation, platform design, and interventions aimed at mitigating the psychological costs of digital environments, and highlight the value of users' subjective beliefs as meaningful signals for identifying harmful content.