Study: AI Tool Reduces Political Polarization on X
A multidisciplinary team of researchers from Stanford, the University of Washington, Johns Hopkins, and Northeastern universities has developed a browser extension that uses AI to detect and reorder polarizing and partisan content on X in real time, without requiring the platform's cooperation.
The tool employs large language models (LLMs) to assign scores to posts based on anti-democratic attitudes and partisan animosity, including content that advocates violence against opposing parties, rejects bipartisan cooperation, or questions facts that favor the other party's views.
Researchers conducted a 10-day field experiment during the 2024 U.S. presidential election campaign with 1,256 participants who consented to have their X feeds modified, randomly assigning them to feeds with increased, reduced or unchanged levels of polarizing content.
Narrative A
Reducing exposure to antagonistic political content on social media dramatically decreases partisan hostility and negative emotions. Algorithmic interventions that limit divisive posts produce changes equivalent to three years of polarization shifts, offering platforms a clear solution to heal democratic discourse and rebuild social trust.
Narrative B
While algorithms can control content visibility, they neither reshape beliefs nor decrease ideological division in a sustained or meaningful way. With the role of echo chambers and filter bubbles often far overstated, we must instead focus on the root causes of social tension to improve social discourse.
Nerd narrative
There's a 95% chance that an AI system will be reported to have independently gained unauthorized access to another computer system before 2033, according to the Metaculus prediction community.
© 2026 Improve the News Foundation.
All rights reserved.
Version 7.2.2