X Tests Downvotes for Premium Users to Combat AI Spam

X is experimenting with comment downvotes, restricting the feature to Premium subscribers as part of its ongoing efforts to curb AI-generated spam and low-quality replies. The update was previewed by X Head of Product Nikita Bier, who bluntly described the current reply algorithm as “the worst product in the company” – lacking meaningful signals and often prioritizing “garbage” like crypto spam.

The new thumbs-down icon appears in the reply function bar. Tapping it opens a menu with five selectable reasons: Not Interested In This Post, Incorrect or Misleading, AI Generated, Spam, or Report Post.

Unlike Reddit’s system, downvotes do not directly reduce a reply’s visibility, alter like counts, or broadly influence trending content. Instead, Bier explained that the primary goal is to collect user preference signals to refine personalized reply ranking and provide a streamlined path for reporting problematic content. The feature aims to give paying users more control over their feed while feeding valuable data back into X’s algorithm improvement efforts.

To prevent abuse, such as coordinated brigade attacks or mass downvoting of dissenting views, the downvote option is exclusive to X Premium subscribers – approximately 1% of the user base. This gating mechanism also helps block bot manipulation, a growing concern as AI tools enable increasingly sophisticated automated accounts.

This is not X’s first attempt at comment downvotes, but previous trials focused mainly on simplifying reporting rather than ranking. The current iteration appears more data-focused, aiming to improve reply quality through aggregated user feedback. f successful, it could become a model for a larger-scope system, or a more refined system that remains exclusive to paying users.