Image: Adobe Stock
Research suggests that the political toxicity many users encounter on social media is a design choice that can be reversed
A new Stanford-led study is challenging the idea that political toxicity is simply an unavoidable element of online culture.
Instead, the research suggests that the political toxicity many users encounter on social media is a design choice that can be reversed. Researchers have unveiled a browser-based tool that can cool the political temperature of an X feed by quietly downranking hostile or antidemocratic posts. Remarkably, this can occur without requiring any deletions, bans, or cooperation from X itself.
The study offers the takeaway that algorithmic interventions can meaningfully reduce partisan animosity while still preserving political speech. It also advances a growing movement advocating user control over platform ranking systems and the algorithms that shape what they see, which were traditionally guarded as proprietary, opaque, and mainly optimized for engagement rather than civic health.
The research tool was built by a multidisciplinary team across Stanford, Northeastern University, and the University of Washington, composed of computer scientists, psychologists, communication scholars, and information scientists. Their goal in the experiment was to counter the engagement-driven amplification of divisive content that tends to reward outrage, conflict, and emotionally charged posts, without silencing political speech.
Using a large language model, the tool analyzes posts in real time and identifies several categories of harmful political subject matter, including calls for political violence, attacks on democratic norms, and extreme hostility toward people from the opposing party. When the system flags such content, it simply pushes those posts lower in the feed so they are less noticeable, like seating your argumentative uncle at the far end of the table during the holiday dinner.
“Social media algorithms directly impact our lives, but until now, only the platforms had the ability to understand and shape them,” said senior author Michael Bernstein, a Stanford computer science professor.
The project gives both users and researchers a way to actually influence the ranking system rather than being subject to it. “It’s a small algorithmic change,” said Bernstein, “but one that puts meaningful power back into the hands of users and researchers instead of leaving it solely with platforms.”
To test the tool, the team asked roughly 1,200 participants to install the tool and use X as usual for 10 days during the heated final run-up to the 2024 US election. This was a period marked by intense political emotions, and over the 10 days, participants viewed modified versions of their X feeds. Some users received the version of their feeds that downranked hostile content, others saw certain inflammatory content upranked, and a control group experienced no changes.
The differences in the results were striking. Participants who encountered less hostile content reported feeling, on average, two points warmer toward the opposing political party on a 100-point scale. While that may seem small, the study notes that it matches the shift typically seen across the entire US population over three years. The effect applied to both liberals and conservatives, confirming that neither side was uniquely susceptible or resistant.
The emotional impact was also significant, as users who saw less incendiary political content reported feeling less angry and less sad overall, suggesting that constant exposure to political hostility carries psychological costs that may quietly wear people down.
“When participants were exposed to less of this content, they felt warmer toward the people of the opposing party,” said first author Tiziano Piccardi, now an assistant professor at Johns Hopkins University. “When they were exposed to more, they felt colder.”
Until now, efforts to reduce polarization online have mostly revolved around blunt solutions that fail to yield consistent results, such as chronological feeds, fact-checking prompts, or pleas for users to spend less time on social media. This study’s success is significant because it preserves full access to political content while just subtly reshaping the ranking signals that determine what migrates to the top of a feed.
This aligns with a movement in the tech research community toward “algorithmic self-determination,” the idea that users should have a say in how ranking systems shape their online experience. Because the team released the code publicly, other developers can build their own versions that help people avoid harassment, manage anxiety, or filter out misleading content.
While the tool is still experimental, its success offers a more optimistic view of social media than we’re used to hearing, showing that a feed’s design plays a considerable role in shaping our political and emotional climate, not solely the posts themselves. And small changes to ranking logic can produce meaningful benefits that shape political attitudes and emotional responses.
Perhaps most importantly, the work demonstrates that polarization isn’t just an inevitable byproduct of life online. Instead, it suggests that with thoughtful design and user-driven control, social media feeds can promote healthier political engagement rather than incite division.
A fight over who gets to post on social media, who gets protected, and who gets fined has landed in Australia’s highest court.
Madeline is a content writer specializing in copywriting and content creation. After studying Art and earning her BFA in Creative Writing at Salisbury University she applied her knowledge of writing and design to develop creative and influential copy. She has since formed her business, Clarke Content, LLC, through which she produces entertaining, informational content and represents companies with professionalism and taste.
