Among the most disturbing findings, content advocating partisan violence receives algorithmic amplification when it generates engagement, potentially normalizing political violence through repeated exposure that makes once-unthinkable ideas seem increasingly acceptable.
The research identified posts supporting partisan violence as one category of divisive content manipulated in feeds for over 1,000 X users during the 2024 presidential election. When such content appeared slightly more frequently, users showed increased polarization and animosity toward political opponents, suggesting that exposure to violence advocacy affects attitudes even when users don’t explicitly endorse violence themselves.
The normalization mechanism operates through repetition. Ideas that initially shock become familiar through repeated exposure. What once seemed extreme gradually shifts toward seeming normal or at least debatable. Algorithms amplifying violence-advocating content accelerate this normalization by increasing exposure frequency beyond what would occur through organic spread alone.
This pattern proves particularly dangerous given contemporary political tensions. Survey data shows concerning percentages of Americans already believe violence might be justified to achieve political objectives. Algorithmic amplification of content supporting such views could push these concerning minorities toward dangerous majorities by normalizing violence advocacy through constant exposure.
Platform responsibility for violence-related content extends beyond simple removal of direct incitement. If algorithms systematically amplify subtler violence advocacy that doesn’t quite violate content policies, platforms may bear responsibility for normalization effects even when individual pieces of content seem borderline acceptable. Addressing this requires coordinating content moderation with algorithmic amplification policies rather than treating them as separate issues.