Platform-independent experiments on social media.

Social media is an important source of political information, yet there is little external oversight of platforms' ever-changing algorithms and policies. This opacity presents a major problem: Conducting a real-world experiment on the causal effects of platform features generally requires the collaboration of the platform being studied, which rarely happens, and even when it does, future platform changes may invalidate prior findings. On page 903 of this issue, Piccardi et al. (1) report one possible solution to this challenge. The authors introduce a methodological paradigm for testing the effect of social media on partisan animosity without platform collaboration by reranking users' existing feeds using large language models (LLMs) and a browser extension. They find that changing the visibility of polarizing content can influence people's feelings about opposing partisans. [ABSTRACT FROM AUTHOR]