Abstract

In this talk, I present two complementary lines of work that together illuminate how information spreads across platforms and is evaluated through human–AI interaction. First, drawing on over 10 million news-link posts shared across seven major platforms including mainstream, alternative, and decentralized systems, I show that relationships between political orientation, engagement, and news quality are highly platform-specific rather than universal. While political engagement follows an “echo-platform” pattern that depends on a platform’s dominant ideology, the tendency for lower-quality news to receive higher per-post engagement is strikingly consistent across platforms, even in the absence of ranking algorithms, pointing to user preferences rather than algorithmic bias. Second, I turn to the rapidly growing role of large language models embedded in social media, analyzing nearly 1.7 million real-world fact-checking requests to AI systems on X. I show that human–AI fact-checking is already operating at scale, exhibits partisan asymmetries in both usage and trust, and produces belief updates comparable to professional fact-checking, while also becoming entangled with polarization and model identity. Together, these findings highlight the need to study social media as a multi-platform ecosystem, where both information exposure and judgment emerge from interactions among users, platforms, and increasingly, AI systems themselves.

Host: Dr Jon Roozenbeek

This talk will be recorded and uploaded to the Zangwill Club Youtube channel in due course.