For years, the consensus among policymakers, tech watchdogs, and the general public has been that social media algorithms serve as the primary engines of political polarization and radicalization. The prevailing theory suggests that when users are funneled content from unreliable or hyper-partisan sources, their beliefs are systematically warped, ultimately eroding trust in democratic institutions. However, a landmark, large-scale study conducted during the tumultuous 2020 U.S. presidential election challenges this assumption, suggesting that the link between social media exposure and ideological shifts may be far more tenuous than previously believed.
The research, led by UC Berkeley Ph.D. student Olivier Bergeron-Boutin in collaboration with Meta’s internal research teams, involved a rigorous experiment where voluntary Facebook and Instagram users had their exposure to “untrustworthy sources” reduced by approximately 70%. Published in Science Advances, the study aimed to see if stripping these sources from a user’s feed would result in measurable changes to their political attitudes. While the intervention successfully altered the content users saw, the researchers found, strikingly, that there was virtually no corresponding shift in the participants’ political beliefs or core attitudes, even among those who had previously consumed the highest volumes of misinformation.
A central insight of the study is that misinformation consumption is not a universal experience; it is highly concentrated within specific subsets of the population. The data revealed that roughly 80% of all exposure to dubious content on Facebook and Instagram was concentrated among just 23% and 11% of their respective users. This finding supports the broader, long-standing academic observation that misinformation operates within “echo chambers.” Experts like computational folklorist Tim Tangherlini note that beliefs are reinforced through social networks—communities where information is retold and refined—rather than being passively absorbed from a single feed. Therefore, an algorithm can control the information flow, but it cannot fundamentally dismantle the social scaffolding that sustains these beliefs.
Comparing the shift in beliefs to “steering a large oil tanker,” the researchers emphasized that entrenched political viewpoints are not easily redirected by a temporary filter on a newsfeed. Because political identity is deeply rooted in personal history, community, and social belonging, removing a specific source of information is merely a marginal adjustment that fails to address the underlying psychological and structural forces of polarization. The study cautions against believing in a “quick fix” for political division, suggesting that the mechanisms of democracy and social interaction are far more resilient—and complex—than simple content-moderation strategies account for.
The study carries sobering implications for the future of transparency in the digital age. Since 2020, the landscape of social media monitoring has degraded; Meta has dismantled the third-party fact-checking programs that enabled this research, and platforms have generally become less accessible to outside scholars. Co-author Brendan Nyhan of Dartmouth College noted that because of this declining transparency, this type of definitive, large-scale academic inquiry is becoming increasingly difficult to replicate. The irony is that as we enter future election cycles, the ability to empirically measure the “harms” of online content is shrinking exactly when it is most needed to inform public policy.
Ultimately, the findings serve as a call for a more nuanced approach to the misinformation crisis. While the researchers warn that this does not mean misinformation is harmless, it does dispel the notion that algorithmic manipulation is solely responsible for contemporary political instability. Instead of focusing on sweeping changes to the average user’s experience, policymakers should look toward those most intensely exposed to bad-faith content and recognize that the roots of polarization lie deeper than the screen. Without better access to data and a more sophisticated understanding of how human beliefs are formed, society risks fighting the symptoms of misinformation while ignoring the structural realities of a modern, divided electorate.
