Engagement, user satisfaction, and the amplification of divisive content on social media

Social media ranking algorithms typically optimize for users' revealed preferences, i.e. user engagement such as clicks, shares, and likes. Many have hypothesized that by focusing on users' revealed preferences, these algorithms may exacerbate human behavioral biases. In a preregistered algorithmic audit, we found that, relative to a reverse-chronological baseline, Twitter's engagement-based ranking algorithm amplifies emotionally charged, out-group hostile content that users say makes them feel worse about their political out-group. Furthermore, we find that users do <i>not</i> prefer the political tweets selected by the algorithm, suggesting that the engagement-based algorithm underperforms in satisfying users' stated preferences. Finally, we explore the implications of an alternative approach that ranks content based on users' stated preferences and find a reduction in angry, partisan, and out-group hostile content, but also a potential reinforcement of proattitudinal content. Overall, our findings suggest that greater integration of stated preferences into social media ranking algorithms could promote better online discourse, though potential trade-offs also warrant further investigation.