Stanford researchers analyzed 2.2 billion social media posts in December 2025 and found that only about 3% of users have ever posted severely toxic content. Yet these prolific few produce roughly one-third of all content across platforms. The findings reveal a stark disconnect between what users believe about online discourse and the actual composition of social media communities.

The algorithm amplifies this distortion. On Twitter/X, toxic tweets receive 86% more retweets than non-toxic ones. Just 0.3% of users shared 80% of contested news, while 6% of users generate 73% of political tweets. Both Democrats and Republicans overestimate the other side's extremism by 3 to 4 times, fueling mutual misperception that silences moderate voices and inflames partisan hostility.

The proposed remedy is "Community Check"—an open-source design layer that could display how the broader population actually feels beneath contentious posts. Studies show that publicly correcting these misperceptions can reduce partisan hostility for up to a month. The concept addresses what Steven Pinker calls a "common knowledge problem": knowing the truth doesn't change behavior until you know others know it too.