X’s Algorithm Reveal Shows How Posts Reach the For You Feed

X released source code for the system that determines how posts are recommended to users, according to Business Insider. The company said the goal is to help people understand whether reach is being limited, whether the system is fair, and why they see particular content. Elon Musk said the move was intended to improve fairness and gather feedback.

The algorithm, called Phoenix, ranks posts for a user’s For You timeline partly by predicting how that user will interact with posts from accounts they do not follow. A predicted URL-copy share carries about 40 times the weight of a predicted like, while replies, quotes, and direct-message shares each carry a weight of 5 versus 0.5 for a predicted like. Follows count as the equivalent of eight likes, and reposts count as two likes.

Phoenix also applies negative weights when it predicts a user will show displeasure with a post. A predicted report counts 468 times in the opposite direction of a predicted like, while a predicted mute is about 118 times the size of a like in the negative direction. A “not interested” interaction is about 86 times as large in the negative direction, and a block is about 62 times as large.

X also announced a pilot feature called “Under the Hood,” which lets participating users see aggregate information about labels applied to their account and posts that can affect visibility. The source code shows X classifies content and accounts across categories including spam and adult material, and those labels can be used by another system when deciding whether a post is shown. X said the pilot is active only for a randomized test group of eligible accounts, with any broader rollout depending on feedback.

The code described by Business Insider shows that X’s For You feed is shaped by predicted behavior, weighted engagement, negative feedback, and account or content labels. For creators and publishers, the practical lesson is that not all interactions are treated equally. Posts that inspire strong positive distribution signals may receive a different ranking boost than posts that mainly draw low-value or negative reactions.