Scientific research has unveiled the mechanism by which social media platforms can rapidly manufacture political animosity. An experiment involving X users revealed that imperceptible adjustments to content feeds generated as much political polarization in seven days as would have taken three years to develop naturally, demonstrating unprecedented algorithmic power over democratic societies.
The study was conducted by researchers from multiple leading universities and published in the prestigious journal Science. They developed an innovative approach using artificial intelligence to evaluate posts in real-time, then modified what appeared in the feeds of more than 1,000 participants. Some users received slightly more content featuring antidemocratic attitudes, partisan aggression, opposition to bipartisan solutions, and distorted political information, while others saw less. Most participants remained unaware of these modifications despite their psychological impact.
The experimental period coincided with the 2024 presidential election campaign, which was marked by viral spread of manipulated and AI-generated political content on X. Since the platform’s acquisition and transformation, its algorithmic curation has prioritized engagement-maximizing content, raising concerns about effects on political culture. The “for you” feed replaced the simpler chronological display of posts from followed accounts with algorithmically selected content designed to keep users engaged.
Martin Saveski, who helped lead the research, noted that the algorithm’s power stems from its subtlety—barely perceptible changes to feeds resulted in significant shifts in users’ feelings toward political opponents. Tiziano Piccardi emphasized that the observed changes correspond to roughly three years of polarization based on historical trends. The measurement approach involved participants rating their feelings toward opposing parties on a 0 to 100 degree “feeling thermometer,” with those exposed to more divisive content showing increased hostility of more than two degrees.
The research carries important messages for platforms and policymakers. Polling indicates that democratic societies face crisis-level political division, with majorities unable to agree even on basic facts. The study proves that platforms could actively reduce this polarization through algorithmic redesign. While there may be trade-offs in terms of overall engagement volume—which could pose challenges for advertising-dependent business models—the research found that users exposed to less divisive content actually engaged more meaningfully through likes and reposts. This suggests that platforms could pursue social responsibility while maintaining viable business operations, though it would require prioritizing societal well-being over maximizing every possible engagement metric.