by Leon Donadoni, Donatella Marazziti, Federico Mucci

Recent developments in the study of highly visual social media have highlighted that, far from being neutral conduits, adolescents’ online environments might be curated by recommender systems that learn from behaviour. On this view, risk and benefit may depend less on time spent online than on patterns of exposure ‒ what reaches young people, how they meet it, and when in development contact occurs. For instance, in large adolescent cohorts, only 10–15% of users account for most of the association between use and distress, underscoring the relevance of heterogeneity. Similar feeds might support connection and care while, in susceptible users, amplifying appearance-centred comparison, compulsive use and contact with harmful material. A public-health perspective treats personalisation as an upstream determinant and invites proportionate responsibilities, namely transparency, independent audit and safety-by-design. Clinically, brief assessments that separate time from type of engagement, paired with low-burden, mechanism-focused interventions, could offer practical steps. Taken together, a precautionary and balanced stance should aim at keeping what is valuable online while reducing foreseeable harm through accountable design.

Key words: social-media, recommender systems, algorithmic exposure, mental health, public health, ethics, safety-by-design