William Brady's team at Northwestern University and the University of Chicago just proved something social media critics have suspected for years: the algorithms that run major platforms are deliberately funneling users toward outrage, and they don't have to.
Over eight weeks spanning the 2024 U.S. presidential election, researchers conducted a rare, large-scale experiment on Bluesky Social, a platform with an open architecture that allowed them to test three different feed algorithms side by side. What they discovered matters to anyone who's felt their feed grow increasingly toxic and polarized: the engagement-based algorithms that power X and Meta systematically amplify moralized, emotional, and toxic political content—but a relatively simple redesign can reduce that amplification without destroying the user experience.
The study, published in Nature, recruited 2,000 U.S. citizens who identified as Democrats or Republicans and were active Bluesky users. Researchers assigned them randomly to view content through one of three custom-built feeds: an engagement-based algorithm designed to mirror major platforms, a basic reverse-chronological feed showing only the newest posts, and a "diversified extremity" algorithm that downranked super-posters and reduced the probability of toxic content.
The numbers are striking. The engagement-based feed amplified toxic and morally outraged political content by roughly 37 percent compared to the reverse-chronological baseline before the election—and by nearly 80 percent after it. This amplification didn't happen by accident. "These algorithms are doing exactly what critics have long argued: They're selectively pushing content that grabs attention by appealing to outrage, moral conflict and negative emotion," Brady said. The engagement-based feed also skewed how users perceived their networks, making them see their political opponents as more hostile than they actually were.
But here's where the research breaks the conventional wisdom about social media design. The diversified extremity algorithm—which simply limited the outsized influence of a small number of extreme users who account for a disproportionate share of toxic posts—consistently reduced exposure to toxic content while preserving what users actually value about the platform. In fact, users in the diversified condition reported greater overall enjoyment of Bluesky after the election compared with those on the engagement-based feed.
"There's a common assumption that reducing toxic content will inevitably hurt the user experience because we like to click on it," Brady explained. "Our results push back on that."
This experiment matters because it's rare. For years, independent researchers have lacked the tools to study how algorithms actually shape political discourse—tech companies don't hand over their code to outsiders. Bluesky's open architecture finally allowed Brady's team to build, test, and measure algorithms in the real world. They analyzed roughly 20 million posts over two months and surveyed participants weekly about their perceptions of dialogue, partisan animosity, and platform satisfaction.
The findings suggest that the toxicity many people experience on social media isn't inevitable—it's a choice baked into platform design. Whether major platforms decide to follow this research remains to be seen, but the experiment proves that a less toxic feed doesn't require sacrificing user engagement or satisfaction. Sometimes, the algorithm that's best for people isn't the one that's best for attention and profit.