For years, the conventional wisdom surrounding social media has held that untrustworthy content is a primary engine of political polarization and that simply restricting its visibility would cure the public’s deepening divisions. However, a landmark study conducted by a team of academics and researchers from Meta during the 2020 U.S. election has challenged this assumption. By reducing the volume of misinformation reaching the feeds of thousands of consenting users, researchers sought to determine if removing “bad” content would naturally shift public opinion. The results were stark: while the feeds themselves were successfully altered, the opinions and beliefs of the users remained stubbornly unchanged.
The study revealed a surprising baseline regarding the actual reach of misinformation. Far from being a ubiquitous presence for the average user, content from untrustworthy sources was found to be highly concentrated. For half of the users on Facebook, such content comprised only 1.1 percent of their total feed, while on Instagram, the figure was a mere 0.1 percent. Instead of a broad societal flood, exposure was clustered among a dedicated minority; roughly 80 percent of all untrustworthy content was consumed by just 23 percent of Facebook users and 11 percent of Instagram users. This supports the concept of “echo chambers,” where rumors flow through networks already predisposed to refine and reinforce specific narratives.
To test the impact of this content, the researchers implemented a three-month intervention that significantly scrubbed the feeds of nearly 16,000 study participants. By applying filters that targeted heavy consumers of misinformation, the team achieved a 70 percent reduction in exposure to untrustworthy sources on both platforms. Despite this drastic shift in the digital environment, subsequent surveys showed almost no movement across 10 key metrics, including trust in media, belief in false claims, and political animosity. For 17 of the 20 primary measures examined, the study found no meaningful effect, suggesting that exposure to misinformation is not the sole, or perhaps even the primary, driver of belief.
Experts suggest that this lack of change stems from the reality that beliefs are social products rather than individual responses to solitary posts. Tim Tangherlini, a researcher at UC Berkeley, likens the process of changing a deeply held belief to turning an oil tanker; it is a slow, structural shift rather than an immediate reaction to feed adjustments. Because these beliefs are embedded within existing social networks and community identities, an intervention that merely alters what flows into a digital feed does not dismantle the underlying infrastructure that sustains those perspectives.
The authors of the study, whose findings were published in Science Advances, are careful not to suggest that misinformation is a trivial issue. Instead, they warn that the lack of a “quick fix” necessitates a more nuanced approach to public discourse. Olivier Bergeron-Boutin, who led the analysis, emphasized that those offering simple solutions to complex problems like polarization should be viewed with skepticism. He argued that while minor digital adjustments might be beneficial, the true roots of these issues are deeply tied to the institutional features of American democracy, which remain largely untouched by algorithmic changes.
Ultimately, the study serves as a snapshot of a moment in time that may be impossible to replicate. The researchers acknowledge significant limitations, including the short three-month duration and the fact that social media platforms have evolved significantly since 2020. Most notably, Meta ended its U.S. fact-checking program in early 2025, stripping away the infrastructure required to conduct such an experiment today. As the tools for measuring and moderating content disappear, the ability to understand how misinformation functions—and whether it can be mitigated—remains a daunting and increasingly opaque challenge for society.
