Human–AI Collaboration in Online Content Moderation: An Empirical Study

Abstract

Online discussion platforms rely on a combination of automated filtering and human review to manage harmful or misleading content. As user populations grow, moderation workflows become increasingly complex. This study examines moderation logs, appeal records, and escalation paths from a large community platform over 18 months. Analysis reveals that automated classifiers handled the majority of routine cases, but human reviewers remained essential for ambiguous and context-dependent content. Several workflow optimizations, including priority routing and reviewer specialization, were introduced. Review consistency improved, although reviewer burnout remained a persistent concern. Effective moderation depends on organizational design as much as on algorithmic accuracy.