Bot Moderation Makes Online Community Discourse Less Toxic

Toxicity persists on social media. It intensifies psychological stress among users, exacerbates intergroup hostility, and undermines the democratic potential of digital spaces. How can people effectively learn civil social norms in technology-mediated environments? In this study, we examine bot moderation as a potential intervention to reduce toxic behavior. Beyond content removal, we propose that bots guide users toward norm compliance by issuing prescriptive norm reminders. In a naturalistic field study of 144 subreddits, we find that, in the short term, bot moderation operates primarily through immediate content removal. Over the long term, prescriptive norm reminders become increasingly effective, fostering cumulative social learning in norm compliance. Furthermore, we identify key heterogeneities: bot reminders are more effective in communities with higher reminder saliency and higher prior levels of toxicity. These findings suggest that a strategic and targeted deployment of bot moderation can help foster more sustainable and civil online discourse.