Nextdoor's content filtering cuts views by 95% but doesn't change user behavior
Two large-scale trials with 200k users show filtering offensive content has zero effect on...
Researchers David J. Grüning and Matthew Katsaros conducted two large-scale randomized controlled trials on Nextdoor, each involving 100,000 users, to test the effectiveness of content filtering—hiding or downranking borderline offensive content. Study 1 (2022) used a report-triggered filter applied to comments, which achieved a modest 12% reduction in views of offensive comments, but across eleven behavioral metrics (including platform visitation, content consumption, and production), no significant effects were found.
Study 2 (2023-2024) addressed the weak manipulation by proactively scoring posts and comments at creation using Google Jigsaw's Perspective API and filtering them from the newsfeed. This produced a near-complete 95% reduction in views of offensive posts. Yet across thirteen additional metrics, the researchers again found no significant behavioral changes. The convergent null results—despite manipulation strength rising from 12% to 95%—provide rare field evidence that filtering changes visibility but not user behavior, underscoring the complexity of effective online moderation.
- Two RCTs with 200,000 users on Nextdoor tested content filtering interventions.
- Study 1 reduced offensive comment views by 12%; Study 2 reduced offensive post views by 95%.
- Neither study found any significant changes in user visitation, content consumption, or content production.
Why It Matters
Platforms spend billions on content moderation; this study suggests filtering alone doesn't alter user behavior.