In an unprecedented exploration of how digital platforms shape political dialogue, researchers have developed innovative custom algorithms to dissect the role of feed-ranking systems in influencing social norms and partisan perceptions. This groundbreaking study, conducted over eight weeks during the pivotal 2024 U.S. presidential election, involved the random assignment of 2,000 participants to different algorithmic feeds, exposing new dimensions of how social media may engineer civic discourse.
At the core of this investigation lies the engagement-based algorithm, widely employed across major social media platforms. These algorithms prioritize content that garners the most interaction, often magnifying intergroup, moralized, and emotional (IME) information. Prior studies have flagged such content as potent in exacerbating political polarization and animosity. By reconstructing this algorithm within a controlled environment, researchers could directly measure its impact on user perceptions, engagement behaviors, and social norm accuracy.
The study contrasted this approach with a more traditional reverse-chronological feed, a mechanism that presents posts in the order they are published, limiting algorithmic interference. This comparison enabled precise evaluation of engagement-driven algorithms’ influence on exposure to emotionally charged and morally framed political content, and how this shapes users’ beliefs about what is normative in political discourse.
A novel element of the research was the introduction of a ‘diversified extremity’ feed-ranking algorithm. This system deliberately attenuates the influence of users who exhibit extreme viewpoints or hyperactivity, a phenomenon increasingly recognized as a key driver of polarization on platforms like Twitter and Facebook. By moderating the visibility of these users’ content, the algorithm aims to foster a more balanced information ecosystem without compromising the richness or enjoyment of the user experience.
The findings provide compelling evidence that engagement-based algorithms do indeed amplify IME and toxic content when compared to chronological feeds. Notably, moral outrage and political messaging surged under the engagement-driven system, aligning with theories that such content provokes stronger user reactions and viral spread. These dynamics contribute to skewed social norm perceptions, where users inaccurately infer that extreme views represent a broader consensus.
Intriguingly, while engagement-based feeds distorted perceptions of prescriptive social norms—those dictating what behaviors are deemed acceptable—they did so in unexpected ways. Rather than uniformly inflating perceptions of divisiveness, the inaccuracies varied, suggesting complex cognitive interactions at play when users interpret algorithmically-curated content. Moreover, participants reported heightened perceptions of partisan animosity, underscoring the role of algorithms in deepening affective divides.
Despite these perceptual shifts, users’ own engagement behaviors remained surprisingly stable across feed types. This indicates that while algorithmic curation may reshape how users view the political landscape and other groups, it does not necessarily translate into immediate changes in their interactive behaviors. Such a dissociation invites further inquiry into the long-term psychosocial and behavioral consequences of curated digital environments.
Turning to the diversified extremity algorithm, the study revealed promising outcomes in mitigating the concentration of IME and toxic content exposures. By effectively curbing the visibility of hyperactive and extreme users, this approach improved the accuracy of social norm perceptions among participants, facilitating a more true-to-reality understanding of political discourse norms. Crucially, it maintained levels of platform enjoyment comparable to those of more traditional algorithms, suggesting that interventions aimed at reducing polarization need not sacrifice user satisfaction.
This research signals an important paradigm shift in algorithm design, advocating for intentional restructuring to combat the pernicious effects of polarization on social media. Rather than passively accepting engagement maximization—which tends to privilege sensational and divisive content—platforms might adopt diversified ranking mechanisms to enhance social cohesion and informational fidelity.
The implications are vast, encompassing the fields of political communication, cognitive psychology, and computer science. Modifying algorithmic feeds not only influences micro-level user perceptions but has potential ripple effects on national electoral dynamics, collective political attitudes, and ultimately democratic governance. As social media increasingly mediates public discourse, such evidence-based interventions could redefine how societies navigate digital public spheres.
Furthermore, the study’s temporal frame—covering pre- and post-election moments—provides critical insights into how algorithmic impacts evolve around high-stakes political events. Election cycles are moments of heightened exposure to political messaging and intensification of moral and emotional expression, further amplifying the stakes of algorithmic influences on social norms and group animosities.
While this experiment breaks new ground, it also opens avenues for future research. Understanding the psychological mechanisms through which algorithmically curated content affects users’ meta-perceptions about their social environments could guide more sophisticated interventions. Additionally, assessing long-term behavioral changes and societal outcomes remains a critical frontier.
In sum, this study confronts one of the most pressing challenges of our digital age: the unintended consequences of algorithmic curation on political polarization and social cohesion. By empirically validating the effects of engagement-driven feeds and proposing viable alternatives that temper extremity exposure, it charts a path toward healthier online political ecosystems that support more accurate social norm perceptions and reduce intergroup hostility.
The nuanced balance between content visibility, user experience, and social norm accuracy illuminated here underscores a pivotal frontier where technology design intersects with democratic values. Such work positions algorithmic redesign not merely as a technical problem but as a crucial endeavor in sustaining the integrity of public discourse in increasingly algorithmic democracies.
Subject of Research: The influence of social media feed-ranking algorithms on political discourse, social norm perceptions, and partisan animosity during the 2024 U.S. presidential election.
Article Title: Redesigning algorithms to intervene on social norm misperceptions during a national election.
Article References:
Brady, W.J., Doyle, M., Elnakouri, A. et al. Redesigning algorithms to intervene on social norm misperceptions during a national election. Nature (2026). https://doi.org/10.1038/s41586-026-10536-1
DOI: https://doi.org/10.1038/s41586-026-10536-1
Tags: 2024 U.S. presidential election social media studyalgorithm redesign for social mediaalgorithmic influence on political beliefsalgorithmic moderation in political communicationelection social norms influenceengagement-based feed-ranking systemsimpact of algorithms on civic discourseintergroup moralized emotional contentpartisan perception in social mediapolitical dialogue on digital platformsreverse-chronological feed effectssocial media and political polarization
