Study reveals how X, Bluesky, and Mastodon reward different political speech patterns

Petter Törnberg's new research applies attention price analysis—an exploratory method using RoBERTa reward models trained on residualized engagement from X/Twitter, Bluesky, and Mastodon—to quantify how different platform environments reward rhetorical, emotional, epistemic, and relational features in public communication. The results reveal sharp differences: X/Twitter sharing strongly rewards antagonism while penalizing respect and nuance. Bluesky reposting favors neutral, lower-emotion language. Mastodon, by contrast, boosts reasoning, nuance, compassion, and collective expression. Toxicity is rewarded across all three platforms, but the relationship is bounded and nonlinear—meaning extreme toxicity doesn't necessarily yield more engagement than moderate forms.

This production-side approach shifts the focus from what content users see to what content political actors create under attention pressure. The study suggests that moving from engagement-optimized platforms like X to alternatives like Bluesky and Mastodon does not eliminate attention pressures but may reward less antagonistic and more deliberative political discourse. The findings make one dimension of platform media logic empirically visible, offering a data-driven tool for understanding how algorithmic incentives shape political communication in the age of social media.