This study empirically investigates how YouTube’s recommendation algorithm shapes filter bubbles and media bias according to users’ topics of interest. Three experimental accounts, focused on political issues, environmental issues, and random interests, were created and programmed to view specific content between March 15 and March 27, 2025. The types, diversity, and bias of recommended content were systematically compared using both quantitative and qualitative analyses.Results showed that YouTube quickly adapts to users’ primary interests, reinforcing filter bubbles and limiting informational diversity. Accounts focused on political or environmental issues exhibited a strong tendency for repeated recommendations of similar content with notable bias, whereas the random interest account maintained broader variety with a weaker filter bubble effect. These findings suggest that YouTube’s algorithm responds acutely to early viewing behaviors, intensifying topic-focused exploration but risking increased information bias and confirmation bias. This research highlights the need for transparent recommendation algorithms and diversity-oriented design, offering practical insights for improving user information access and media literacy in digital environments.