Social media platforms possess the technological capability to reduce political polarization, according to research demonstrating that algorithmic choices can decrease animosity as effectively as they currently increase it. When divisive content was down-ranked in users’ feeds, political hostility declined by amounts matching the increases seen when such content was boosted.
The study involved over 1,000 X users during the 2024 presidential election. Researchers created experimental conditions where some participants saw slightly more posts containing anti-democratic attitudes and partisan animosity, while others saw fewer such posts. The changes were subtle enough that most users never consciously noticed their feeds had been altered.
Results showed symmetrical effects. Just as increasing divisive content made users feel more negatively toward political opponents, decreasing such content made them feel less hostile. This symmetry suggests that current high levels of online polarization aren’t inevitable consequences of digital communication but rather reflect specific algorithmic choices that could be modified.
The business implications deserve serious consideration. Platform revenue models typically prioritize engagement metrics like time spent scrolling and number of posts viewed. Divisive content tends to provoke strong reactions that boost these metrics, creating financial incentives for algorithms that increase polarization. However, researchers found that down-ranking divisive content only slightly reduced overall engagement while increasing meaningful interactions like likes and reposts.
This finding challenges the assumption that healthier algorithms would necessarily devastate platform profitability. Users might spend marginally less time scrolling but engage more thoughtfully with content they do see. Whether platforms will voluntarily adopt such changes remains uncertain, but the research demonstrates that technical solutions exist for those willing to prioritize democratic health over maximum engagement optimization.