X Open-Sources the Code Behind Its Recommendation Algorithm

X, the social network run by American entrepreneur Elon Musk, open-sourced (open source — releasing software code publicly for free use) on Thursday the system that determines which posts get recommended to users. The move marks Musk's latest step toward greater platform transparency.

In a post on X, the company explained: "The goal is simple — people should be able to see for themselves whether the platform is limiting their reach, whether the system is fair, and why they're seeing this particular content." Musk said the decision to open-source the code was made to "increase fairness" and gather feedback for further improvements.

The company's algorithm, called "Phoenix," analyzes a user's recent viewing history to predict how they'll react to a post from an account they don't follow, and ranks posts in the personalized "For You" feed accordingly. If Phoenix predicts a user will copy and share a post's link, that post gets a weight roughly 40 times greater than a predicted "like." A reply, a quote, and a share via direct message each carry a weight of 5 points — 10 times the 0.5-point weight of a "like." A follow carries the weight of 8 likes, and a repost the weight of 2 likes.

Conversely, if Phoenix predicts a user will react negatively to a post, negative weights sharply curb its visibility. A predicted "report" carries a negative weight roughly 468 times that of a like, "mute" about 118 times, "not interested" 86 times, and "block" 62 times. These figures suggest that content designed to provoke angry reactions — so-called "ragebaiting" (deliberately posting divisive content to anger users and spark more replies and arguments) — is not an effective way to grow followers, a somewhat unexpected finding given X's reputation under Musk.

The announcement also introduced a pilot feature called "Under the Hood" (a test feature showing restrictions applied to an account). It lets participating users see a summary of the labels applied to their account and posts that affect visibility. X's code reveals that the system sorts content and accounts into categories such as spam and adult material, and a separate system then uses these labels to decide whether to show a post. The company says the feature is currently active only for a "randomly selected test group," with wider rollout depending on user feedback.

The move is the latest in a string of steps by X to open-source parts of the platform. In January, Musk called the algorithm "dumb" and promised to make the system more transparent. "At least you can see we're trying to improve it in real time and transparently — no other social media company does that," Musk wrote on X at the time.