X's Algorithm Weights One Conversation at 150 Likes

A closer look at X’s recommendation system suggests the real driver of reach is not the post itself, but what happens after it is published. Growth researcher Charles Floate analyzed the company’s open-sourced Phoenix algorithm and found that conversations, especially back-and-forth replies, carry far more weight than passive actions like likes.

Floate identified an engagement hierarchy in which a two-way conversation carries approximately 150 times the algorithmic weight of a like, the highest-leverage action available on the platform. According to the analysis, a simple like carries the lowest weight in the system.

The data comes from X's Phoenix codebase, which the company open-sourced in January 2026. Phoenix replaced X's previous hand-engineered recommendation system with a transformer architecture powered by Grok, trained to predict probabilities across 15 distinct engagement signals simultaneously. Floate's analysis of the codebase surfaces the weight assigned to each signal.

How X's Algorithm Scores Engagement

Floate’s breakdown of the Phoenix codebase shows how each action contributes to ranking in the feed. Likes sit at the bottom with a base value of 1. Bookmarks follow at 10 times that value. Link clicks come in at 11, while profile clicks reach 12. Replies are slightly higher at 13.5, and reposts jump to 20.

The biggest shift appears in direct interaction. When a post author replies to someone, that action carries a weight of 75. The gap between the highest-weighted action and the most common one, a like, is the central finding. A post author responding to a reply generates 75 times the algorithmic signal of a like. A two-way conversation, where a user replies and the post author responds, generates approximately 150 times the signal of a single like.

Why Conversation Outperforms Content

The weight structure reflects a deliberate design choice. Phoenix prioritizes mutual engagement over passive content consumption. Likes and bookmarks score near the bottom of the hierarchy. Author-initiated replies to replies score highest by a wide margin.

The practical consequence is direct. Publishing content that generates likes produces a fraction of the algorithmic signal that a single responded-to conversation generates. In Floate's framing, the core implication is not about producing better content but about what happens after publishing. A brand that generates a reply and responds to it is worth more algorithmically than one that collects dozens of likes on the same post, regardless of content quality or posting frequency.

This also affects how brands handle links. Floate notes that including external links in the main post may reduce distribution. A common workaround is placing the link in a reply, allowing the main post to circulate without restriction while still providing access to the destination.

The Phoenix Algorithm and For You Feed

Phoenix is a transformer architecture that scores 15 engagement signals simultaneously, replacing X's previous system, which evaluated signals individually via hand-engineered rules. X open-sourced Phoenix in January 2026. The system uses Grok to make probabilistic predictions across signals rather than applying fixed rule weights to individual actions.

Floate's analysis also notes that X's "For You" feed splits approximately 50% between followed accounts and 50% unfollowed. That ratio makes the recommendation algorithm a constant factor in distribution regardless of follower count. A smaller account with high conversational engagement can reach audiences beyond its existing followers more reliably than a larger account generating primarily passive engagement.

X also incorporates a TweepCred score (0-100) that accounts for account age, follower ratios, and engagement quality. Floate identifies accounts scoring below 65 as reportedly facing distribution limits, though X has not publicly confirmed the threshold.

Competitive Context

No other major social platform has published its engagement weights at this level of detail. Instagram and TikTok both weight saves and replays more heavily than likes, but neither has disclosed specific multipliers. X's 75x weighting on author-to-reply responses is unusually steep by comparison, and its explicit nature gives marketers a directly actionable framework not available on other platforms.

The broader pattern across platforms is consistent: passive engagement is deprioritized as platforms seek signals indicating genuine interest. X's open-source codebase makes that principle quantifiable for the first time at this level of detail, and Floate's analysis makes those weights accessible without requiring marketers to parse the codebase directly.