Reddit Bot Moderation Outperforms Humans: 11.8M Events Study

A large-scale study from researchers at the University of Southern California and the University of Pittsburgh examined content moderation on Reddit using 11.8 million events over four years. The team found that bot moderation (automated systems) consistently outperforms human-only or mixed modteam moderation in two key metrics: higher user compliance with rules and lower self-censorship after removal. This directly challenges the common belief that human agency cues are inherently better for community governance. Modteam moderation, where multiple humans act under a shared identity, produced the strongest self-censorship effects, suggesting institutional depersonalization drives behavioral withdrawal.

The study also analyzed 480 linguistic interaction features, with 33 surviving statistical correction. For routine violations, strategies like elaborated explanation, community-scale appeals, and direct personal address are effective. However, for serious violations, those same approaches backfire. Instead, prosocially framed messages with emotional emphasis become most effective when stakes are highest. The researchers extend the Human-AI Interaction Theory of Interactive Media Effects (HAII-TIME) by introducing violation salience as a critical moderator. This provides empirical grounding for context-adaptive moderation design on platforms like Reddit.