X Heavy Ranker optimizes clicks and replies, not whether users found a link useful

Afinetheorem · x · 2026-07-29

The post argues that the misleading part is the claim that links are not penalized directly. In practice, ranking is still heavily influenced by downstream engagement, and posts that send users away usually collect fewer likes and replies.

The attached screenshot of X’s Heavy Ranker documentation is used to show what the model optimizes: prediction of engagement actions such as favorites, retweets, replies, profile clicks, and conversation clicks. The author’s point is that the system measures proxy signals, not the user’s actual satisfaction with reading an informative post and then clicking out.

Related event: X's Heavy Ranker: External Links Indirectly Hurt Post Distribution(2 posts)→

Original post →

More from Companies & People

Companies & People channel →