Research has found that small changes in the tone of posts fed to X users can increase feelings of political polarization in a week by more than it would historically take at least three years.
An unprecedented experiment to assess the ability of Elon Musk’s social platform to exacerbate political division found that when posts expressing undemocratic viewpoints and partisan hostility were promoted in the feeds of Democrat and Republican supporters, barely perceptibly, there was a major change in their unfavorable feelings toward the other party.
The degree of increased division – known as “effective polarization” – achieved by changes made by academics to X users’ feeds in one week was equivalent to what it took on average three years between 1978 and 2020.
Most of the more than 1,000 users who took part in the experiment during the 2024 US presidential election did not notice that the tone of their feed had been changed.
The campaign was marked by divisive viral posts on
They found that repeated exposure to posts expressing anti-democratic viewpoints and partisan hostility “significantly impacts” users’ feelings of polarization and increases sadness and anger.
Musk bought Twitter in 2022, renamed it
The extent to which more undemocratic posts make users feel more hostile toward political opponents “demonstrates the power of the algorithm,” said Martin Sevski, an assistant professor at the University of Washington’s School of Information, who co-authored the study published in the journal Science with colleagues at Stanford, Johns Hopkins and Northeastern universities.
“The change in their feed was barely noticeable, yet they reported a significant difference in how they felt about other people,” said Tiziano Picardi, assistant professor in the computer science department at Johns Hopkins University and co-author of the research. “Based on U.S. trends, this shift matches about three years of polarization.”
The study also found that relatively subtle changes to the content of users’ feeds could significantly reduce political animosity between Republicans and Democrats, suggesting that X had the power to increase political harmony if Musk chose to use it that way.
After newsletter promotion
“The exciting thing about these results is that there is something the platforms can do to reduce polarization,” Sevski said. “This is a new approach they can take in designing their algorithms.”
X has been contacted for comment.
According to Pew research, eight in 10 American adults say that not only can Republicans and Democrats not agree on policies and plans, but they can’t even agree on basic facts. More than half of people in Britain believe differences in people’s political views are so divisive that it is dangerous for society, a recent poll conducted by Ipsos found.
Changes in political polarization resulting from ex post exposure were measured using a new approach. First, academics used AI to analyze posts in X’s “For You” feed in real time. The system then showed more divisive posts to one group and less divisive posts to the other, a power that was typically the sole preserve of X. Divisive posts included those that showed support for undemocratic practices, partisan violence, opposition to bipartisan consensus, and biased assessment of political facts.
After a week of reading these subtly altered feeds, researchers asked users to rank how warm or cold, favorable or unfavorable they felt toward their political opponents. Changes in “effective polarization” were ranked over two degrees on a 0 to 100 degree “feeling thermometer”. This was the same amount of increased polarization that had generally occurred in the US in the four decades leading up to 2020. Feeding users fewer posts containing anti-democratic attitudes and partisan hostility reduced political division by a similar amount.
Social media platforms have long been accused of encouraging divisive content in order to increase user engagement and therefore advertising revenues. But the research found that when divisive content was down-ranked there was a slight decrease in overall engagement in terms of time spent on the platform and number of posts viewed, but those users tended to “like” or repost more often.
“The success of this method suggests that it can be integrated into social media AI to reduce harmful personal and social outcomes,” the authors wrote. “At the same time, our engagement analyzes indicate a practical trade-off: interventions that down-rank (democratic and partisan content) may reduce the amount of short-term engagement, creating challenges for engagement-driven business models and supporting the hypothesis that content that provokes stronger reactions generates greater engagement.”
<a href