The era of claiming social media platforms are neutral town squares has definitively ended. New research demonstrates that companies can turn political polarization up or down through algorithmic adjustments, with effects that manifest in just seven days. The study found that subtle feed manipulations produced polarization changes that would normally require at least three years to develop naturally.
Researchers from multiple universities collaborated to conduct the experiment on X during the 2024 presidential election. They developed AI systems capable of analyzing post content in real-time and adjusting what appeared in users’ feeds based on levels of anti-democratic sentiment and partisan hostility. Over 1,000 participants received customized experiences without being told their feeds were modified.
The results challenge any remaining notion that online polarization is simply a reflection of offline divisions. When researchers increased divisive content by barely perceptible amounts, users reported significantly more negative feelings toward political opponents. The effect size—measured on a zero-to-100 degree “feeling thermometer”—showed changes of more than two degrees, matching decades of gradual polarization in the physical world.
This algorithmic power operates largely invisibly. Most experiment participants never realized their feeds had been manipulated, yet their political attitudes shifted measurably. This raises profound questions about informed consent and democratic autonomy. If citizens’ political views can be shaped without their knowledge by algorithms optimized for corporate objectives, traditional notions of free political discourse require urgent reevaluation.
The research also identified a potential solution. When divisive content was down-ranked, political animosity decreased by similar magnitudes to the increases seen when such content was boosted. While engagement-driven business models might resist such changes, the findings suggest that healthier algorithms wouldn’t necessarily crater user interaction—people just might engage more thoughtfully rather than more frantically.