The Prosocial Ranking Challenge: Reducing Polarization on Social Media without Sacrificing Engagement
A 6-month experiment with 9,386 users shows algorithms can reduce political division without major platform trade-offs.
A massive, multi-platform study led by Stanford researchers has delivered the first direct evidence that social media algorithms can be redesigned to reduce political polarization without sacrificing user engagement. The 'Prosocial Ranking Challenge,' involving 42 authors, deployed a browser extension to randomly assign 9,386 desktop users to either a control group or one of five experimental ranking algorithms during the 2024 U.S. presidential election. For six months, these algorithms altered the content shown across Facebook, X/Twitter, and Reddit. The key finding was a statistically significant reduction in the study's pre-registered index of affective polarization by an average of 0.03 standard deviations, which included a measurable 1.5-degree decrease in the difference between how users rated their own political party versus the opposing party on a 100-point 'feeling thermometer.'
Crucially, the experiment found this reduction did not come with a catastrophic drop in platform usage, a primary concern for social media companies. While active use time decreased slightly for Facebook (-0.37 minutes/day) and Reddit (-0.2 minutes/day), it actually increased for X/Twitter by 0.32 minutes per day. The research, published on arXiv, did note an increase in reports of negative social media experiences but found no significant effects on broader measures like user well-being, news knowledge, empathy for political outgroups, or perceptions of partisan violence. This implies a potential path forward: integrating 'bridging' content that exposes users to opposing viewpoints can improve societal health metrics without fundamentally breaking the engagement-based economic engine of major platforms.
- Reduced affective polarization by 0.03 standard deviations, including a 1.5-point shift on in-party/out-party feeling thermometers.
- Platform engagement saw mixed results: Facebook/Reddit use dipped slightly, but X/Twitter use increased by 0.32 minutes per day.
- Found no negative impact on user well-being or news knowledge, suggesting prosocial algorithms don't require major trade-offs.
Why It Matters
Provides concrete, scalable evidence that social platforms could algorithmically mitigate political division without destroying their core business model.