For years, we have been told that social media platforms are essentially massive engines of radicalization, flooding the average person’s feed with a toxic, inescapable diet of conspiracy theories and lies. It is a compelling narrative that has fueled congressional hearings, aggressive policy debates, and constant calls for Big Tech to “clean up” the digital environment. However, a massive new study published in Science Advances—which tracked nearly 16,000 Facebook and Instagram users during the turbulent 2020 presidential election—has challenged this conventional wisdom. When researchers systematically slashed the amount of unreliable content users saw by roughly 70%, they expected to see a shift in public opinion. Instead, they found that even a dramatic reduction in “fake news” had almost no measurable impact on users’ political beliefs, their trust in institutions, or their adherence to false claims about the election or COVID-19.
The study’s findings are particularly striking because of how they reshape our understanding of who actually sees misinformation. It turns out that for the vast majority of people, the internet is not the chaotic misinformation machine we often imagine. Most users rarely encounter content from unreliable sources; for the average person, such content is statistically negligible. However, there is a “long tail” to this problem: a small, concentrated minority of users accounts for the lion’s share of exposure. Roughly 23% of Facebook users and 11% of Instagram users were responsible for nearly 80% of the total exposure to unreliable content on their respective platforms. These heavy consumers were effectively living in a different information ecosystem than the rest of the user base, yet even for them, removing that content failed to change their core attitudes.
The experiment was conducted with a high degree of precision, separating participants into control and treatment groups to see if “starving” them of untrustworthy content would influence their worldview. While the intervention was a massive technical success—successfully stripping away thousands of views of inflammatory material—the psychological outcome was a wall of silence. Whether it was feelings toward opposing political parties or beliefs about election integrity, the needle simply did not move. This suggests that while we can successfully clean up a feed, changing a mind is a much more complex, deeply rooted human process. People do not seem to become radicalized simply by scrolling; their views are likely formed by a much broader, more stubborn array of social, cultural, and offline influences.
One of the most revealing aspects of the research is where this content originates. On Facebook, the primary culprits were not individual posts, but “Pages”—broadcast-style hubs that effectively curate and amplify misinformation to massive, captive audiences. On Instagram, the dynamic was different; users were primarily seeing misinformation from accounts they had consciously chosen to follow rather than content forced upon them by opaque algorithms. This distinction is vital for future policy, as it suggests that simply tweaking recommendation algorithms may be far less effective than addressing the underlying issue of why users choose to follow these accounts in the first place. Furthermore, these unreliable sources were rarely just spreading misinformation; they were also hubs for hate speech and bullying, suggesting that the “misinformation problem” is really a broader “bad actor” problem.
We must be careful not to misinterpret these findings as a sign that misinformation is harmless. Misinformation is still a major issue that impacts the quality of our public discourse and the safety of our communities. However, the study serves as a necessary reality check for policymakers and platform executives who have long banked on the idea that “turning down the volume” of bad content would solve political polarization. The reality is that the problem is highly localized to a specific, heavy-consuming demographic, and the link between what we see online and what we believe is far more tangled and resistant to intervention than we once assumed. Relying solely on technical “de-ranking” or content suppression is clearly not the silver bullet many hoped it would be.
As we look toward the future, these results highlight the limits of platform engineering in a democracy. While social media companies continue to evolve their moderation strategies—shifting from third-party fact-checkers to community-based notes—the fundamental question remains unanswered: if removing content doesn’t change minds, what will? This research reminds us that political and social divides are deeply entrenched, and while tech companies can manage the hygiene of their platforms, they cannot easily engineer their way out of deep-seated human disagreement. We are left with an urgent, lingering question: if our digital feeds aren’t the sole culprit for our current polarization, then we must turn our attention elsewhere to understand why our society has become so divided.