Adolescent AI Use and Evaluations: Prospective Bidirectional Associations with Internalizing Symptoms

Key Findings

• — Past-month AI use among the cohort rose from 62% to 75% across the five-month interval
• — Any AI use was associated with higher social anxiety (concurrent b=0.27, P<.001; prospective b=0.22, P<.001) and depressive symptoms (concurrent b=0.11, P<.001; prospective b=0.12, P<.001)
• — Frequency of AI use was associated with loneliness (concurrent b=0.03, P=.007), social anxiety (concurrent b=0.08, P<.001; prospective b=0.07, P<.001) and depressive symptoms (concurrent b=0.06, P<.001; prospective b=0.05, P<.001)
• — Of the three evaluative dimensions, Emotional Intimacy and Companionship prospectively predicted greater school loneliness (z=2.98, P=.003)
• — No internalizing symptom variable prospectively predicted AI evaluations, so within this sample the association ran in one direction only

Methodology Notes

Two-wave longitudinal cohort, approximately five-month interval; parents of all enrolled students with valid email addresses were invited, 2,570 consented and 2,342 students assented and started the survey. Grades 6-8, mean age 12.8, 47% boys; 70% White, 20% Black, 12% Hispanic/Latino, 5% Asian. Outcomes used validated instruments (PISA Sense of Belonging at School Index, Social Anxiety Scale for Adolescents, Short Mood and Feelings Questionnaire), but the AI use and AI evaluation items were developed by the research team and are not previously validated. Single US state, so generalisability is limited, and observational design means the prospective associations are not causal. Not peer reviewed: the manuscript states it is under review and asks that the journal version be cited once published. Authors are at the Department of Psychology and Neuroscience, University of North Carolina at Chapel Hill; funded by the North Carolina Collaboratory with state-appropriated funds, with support from the Winston Family Foundation. Declared conflicts are material to how the work is read and are quoted from the manuscript: Burnell 'received personal fees for serving as a paid consultant for ongoing social media litigation'; Telzer and Prinstein 'have been retained as an expert witness in U.S. social media litigation'; Maheux 'attended an unpaid convening on child safety at OpenAI'. Verified via the OSF API preprint record (title, authors, date_published 2026-08-03, preprint DOI) and by downloading and reading the deposited PDF title block, which supplied the affiliations, funding and conflict statements that the abstract omits.

Sources