YouTube Recommendations Reinforce Negative Emotions: Auditing Algorithmic Bias with Emotionally-Agentic Sock Puppets

YouTube's algorithm amplifies negative emotions: The study found that YouTube actively reinforces negative emotional preferences like anger and grievance by increasing their prevalence in recommendations, meaning users seeking out angry or grievance-focused content will see more of it—not less.

The problem gets worse over time: The reinforcement of negative emotions intensifies as users continue engaging with the platform, suggesting that YouTube's recommendation system creates a self-reinforcing cycle that deepens emotional biases rather than moderating them.

Context matters more than personalization: Surprisingly, recommendations based on video context (what's similar to what you're watching) were often more effective at reinforcing emotional alignment than recommendations based on your personal viewing history, indicating the problem may be structural to how the algorithm works.

Emotional filter bubbles pose real risks: By systematically amplifying negative emotions across different contexts, YouTube may be creating "emotional filter bubbles" that could affect user well-being and have broader societal consequences, particularly for vulnerable users.

Platforms need to prioritize diversity over pure engagement: The findings suggest that recommendation algorithms should be redesigned to balance personalization with content diversity and give users more control over their recommendations, rather than optimizing solely for engagement metrics.