Abstract:
Moderation changes what gets created, not just what gets removed or seen: creators anticipate rules and adjust behavior - so outcomes depend on incentives, not only takedowns. Neutral moderation may not have neutral effects: stricter toxicity controls encourage the ideological majority while silencing the minority. Average wins can hide widening gaps: platform-wide positive engagement may improve while minority readers and creators fall behind. Detection, flagging, and appeals must distinguish toxicity from disagreement and avoid mislabeling civil, cross-cutting speech.
