Redesigning Algorithms to Intervene on Social Norm Misperceptions During a National Election

For the first time in history, civic discourse commonly occurs in digital environments where algorithms influence how we are exposed to social information. It is increasingly important to understand whether and how these algorithms impact our politics. Here, we build custom feedranking algorithms with full control over their features, and randomly assign 2000 participants to use one over the 8 weeks before the 2024 U.S. presidential election. We test whether an engagement-based algorithm (used on major social media platforms) amplifies ingroup, moralized and emotional information in ways that skews perceptions of social norms around political dialogue, and whether it leads to increases in perceptions of partisan animosity (compared to a reverse-chronological feed). We also build and test a ‘diversified extremity’ algorithm feed that aims to simultaneously improve the accuracy of social norm perception and reduce perceptions of partisan animosity compared to the engagement-based feed.