Algorithmic Audit of Personalisation Drift in Polarising Topics on TikTok

A team of 11 researchers from Slovak University of Technology published a comprehensive algorithmic audit of TikTok's recommendation system, examining how it handles polarizing topics like US politics, climate change, vaccines, and conspiracy theories. Using controlled accounts designed to simulate users with aligned or opposing interests, they systematically measured content exposure drift over time across three dimensions: preference alignment, topic polarization, and stance reinforcement.

Their findings reveal complex and topic-dependent recommendation patterns. For US politics, TikTok's algorithm showed strong personalization toward user interests while simultaneously amplifying opposing political stances—creating a paradoxical effect of both reinforcing and challenging user viewpoints. In contrast, for misinformation-themed topics, the platform demonstrated a neutralizing effect, steering users away from extreme content. The study provides concrete evidence that recommendation trajectories differ markedly across topics, with some pathways amplifying polarized viewpoints more strongly than others.

This research represents one of the most systematic audits of TikTok's personalization mechanisms to date, employing rigorous methodology with controlled accounts and longitudinal tracking. The findings have significant implications for platform governance, suggesting that blanket approaches to content moderation may be insufficient given the topic-specific nature of recommendation drift. The study also highlights the need for greater transparency in how social media algorithms handle politically sensitive content versus other polarizing topics.