AI Analysis Reveals How Partisan Posts Reshape Political Attitudes in Days

Artificial intelligence isn’t just being used to generate content—it’s also being deployed to understand how content shapes minds. Researchers used AI systems to analyze social media posts in real-time, identifying markers of anti-democratic attitudes and partisan hostility. They then manipulated what users saw based on these analyses, demonstrating how quickly political attitudes can shift when feeds are subtly altered.
The experiment took place on X during the 2024 presidential election cycle, a period marked by widespread viral misinformation. Fake images and AI-generated propaganda reached tens of millions of viewers, creating an information environment where distinguishing truth from fabrication became increasingly difficult. Against this backdrop, over 1,000 users unknowingly participated in research examining how feed composition affects polarization.
Researchers identified several categories of divisive content: posts supporting undemocratic practices, those advocating partisan violence, content opposing bipartisan consensus, and posts presenting biased interpretations of factual information. By increasing or decreasing such content in different users’ feeds, they could isolate the specific effects of exposure to these various forms of divisive messaging.
The results quantified algorithmic power with unprecedented precision. Changes in polarization measured on a “feeling thermometer” exceeded two degrees on a 100-point scale after just one week. To put this in context, American society typically experienced such shifts over three-year periods between 1978 and 2020. The acceleration represents a fundamental change in how political division develops.
Beyond polarization, the research documented emotional consequences. Users exposed to more divisive content reported increased sadness and anger, suggesting that algorithmic choices affect mental health alongside political attitudes. This dual impact on cognition and emotion makes social media’s influence particularly powerful and potentially harmful to both individual wellbeing and democratic functioning.