Stanford University researchers have developed a new tool designed to reduce political polarization on social media platform X, formerly known as Twitter. The research, published in Science, introduces a web-based method that downranks posts containing antidemocratic attitudes and strong partisan hostility, without removing them or requiring cooperation from the platform itself.
The team tested their tool with approximately 1,200 participants during the 2024 election over a ten-day period. Participants whose feeds were reordered to display less antidemocratic content reported warmer feelings toward the opposing political party. This effect was observed among both liberal and conservative users.
“Social media algorithms directly impact our lives, but until now, only the platforms had the ability to understand and shape them,” said Michael Bernstein, professor of computer science at Stanford’s School of Engineering and senior author of the study. “We have demonstrated an approach that lets researchers and end users have that power.”
Bernstein added that this approach could lead to interventions promoting greater social trust and healthier democratic discourse. The study built upon earlier Stanford sociology research by identifying types of online content considered threats to democracy—such as calls for violence or rejection of bipartisan cooperation—and targeting those for downranking.
Study co-author Jeanne Tsai, professor of psychology in Stanford’s School of Humanities and Sciences, noted the emotional response triggered by polarizing content: “This polarizing content can just hijack their attention by making people feel bad the moment they see it,” she said.
The project included collaborators from University of Washington and Northeastern University across several disciplines. Tiziano Piccardi, first author and former postdoctoral fellow at Stanford (now assistant professor at Johns Hopkins University), developed a browser extension powered by artificial intelligence that scans for negative partisan sentiment before reordering posts in real time.
In experiments comparing groups exposed to more or less extreme political content versus a control group, researchers found clear impacts on polarization. “When the participants were exposed to less of this content, they felt warmer toward the people of the opposing party,” Piccardi explained. “When they were exposed to more, they felt colder.”
Participants’ attitudes improved by an average of two points on a scale measuring feelings toward opposing parties—a shift similar to changes seen nationally over three years.
Previous interventions aimed at reducing polarization online have yielded mixed results and often relied on blunt strategies like chronological ranking or encouraging users to leave social media altogether. In contrast, this new tool allows for more nuanced filtering while granting users increased control over their feeds.
The research team has released the tool’s code so others can create independent ranking systems outside existing platform algorithms. They are also exploring similar approaches for mental health interventions.
Other contributors include Martin Saveski (University of Washington), Chenyan Jia (Northeastern University), and Jeff Hancock (Stanford). Funding was provided by national science foundations in both Switzerland and the U.S., along with support from Stanford’s Institute for Human-Centered Artificial Intelligence.
