How Clicks Affect X Reach
Clicks are one of the most interesting engagement signals in the X recommendation architecture because a user can show interest without liking, replying or reposting a post. The current Phoenix model separately predicts several click-related actions, including post clicks, profile clicks, link clicks and other click/expansion behaviors.
(GitHub)
A Click Is Not the Same as a Like
Both users demonstrated interest, but in different ways. The recommendation model can represent those behaviors separately.
Weights Scale Predicted Probability, Not Raw Counts
The latest X repository update specifically addresses a common misunderstanding: The weights scale the predicted probability of an action, not the number of times that action has already happened.
X explicitly warns against interpreting a weight ratio as a statement that one raw action cancels another number of raw actions. This is very important because it prevents your guides from turning open-source code into a misleading “X points calculator.”
Final Under The Hood Summary
Likes: A positive predicted action that contributes to ranking.
Replies: A separately predicted conversational action.
Reposts: A separately predicted redistribution action.
Quotes: A distinct prediction from ordinary reposts.
Clicks: A family of interest signals—including post, profile and link clicks.
The most important clarification from the latest X source is: The weights apply to predicted behavior, not raw engagement counts.
How X Ranks and Recommends Posts
Official Open-Source References
- • X / xAI — For You Algorithm repository — current architecture, Phoenix, Thunder, filtering, scoring and candidate pipeline. (GitHub)