X is giving users and developers a much closer look at what happens behind the scenes when a post appears in the For You feed.
The social platform formerly known as Twitter has released the code behind its recommendation system, offering unusually detailed insight into how its Phoenix ranking model decides which posts users are most likely to see.
The newly available code shows that X’s feed is far more complicated than simply counting likes, reposts or replies. Instead, the system attempts to predict several different actions a user might take and combines those predictions to determine how posts should be ranked.
The official X For You Feed Algorithm repository on GitHub explains that Phoenix looks at a viewer’s recent engagement history and predicts how likely that person is to perform different actions on each candidate post.
For creators trying to understand why one post reaches a large audience while another disappears quickly, the release offers a rare look inside one of the world’s biggest social-media recommendation systems.
What Is Phoenix?
Phoenix is one of the key components behind X’s personalized recommendation system.
Rather than showing every user the same trending content, Phoenix attempts to determine which posts are most relevant to each individual.
According to the official open-source repository, Phoenix examines a viewer’s recent engagement history and predicts the likelihood that the viewer will take various actions on a particular post.
Those predictions are then combined into a score.
Posts with stronger predicted scores have a better chance of being placed higher in the For You feed, although ranking is only one part of the overall system.
This means two people following similar accounts could still receive very different For You feeds depending on how they previously interacted with posts.
X Is Predicting More Than Just Likes
One of the most interesting details revealed by the code is that not every interaction is treated equally.
The recommendation system considers multiple predicted actions rather than relying on a single engagement metric.
These can include actions such as liking a post, replying, quoting it, sharing it or sending it through a direct message.
A report from Business Insider notes that the released system assigns different importance to these interactions when calculating ranking scores.
This is important for creators.
A post that encourages meaningful sharing or conversation could potentially send stronger recommendation signals than one that simply collects passive likes.
However, creators should avoid treating individual weights as a guaranteed formula for going viral. Recommendation systems can change, and X uses multiple stages of retrieval, ranking, filtering and feed construction before users see the final result.
Negative Feedback Can Hurt Distribution
The released information also highlights the importance of negative user signals.
Actions such as blocking or muting accounts, reporting content or selecting options indicating that a user does not want to see similar posts can work against a post or account.
That makes engagement bait potentially risky.
A controversial post might attract a large number of replies, for example, but if it simultaneously generates reports, blocks or other strong negative signals, high raw engagement alone does not necessarily mean the algorithm will continue pushing it.
This gives creators another reason to focus on content people genuinely want to interact with instead of attempting to manufacture outrage purely for engagement.
The Feed Uses Two Major Pipelines
The open-source documentation also provides a clearer picture of how X builds the final feed.
The X algorithm repository describes two major pipelines: the Post Pipeline and the Blending Pipeline.
The Post Pipeline is responsible for finding, ranking and filtering posts.
The Blending Pipeline then combines those results with additional elements that are not ranked in exactly the same way, including advertisements, account recommendations such as Who to Follow, and other prompts.
This distinction matters because Phoenix itself does not decide every single item that appears on the screen.
It is a major ranking component inside a much larger recommendation infrastructure.
Ranking and Visibility Are Not the Same Thing
Another useful detail is the separation between ranking and visibility.
A post may theoretically receive a strong ranking score, but that does not automatically mean it is eligible to appear.
The official documentation says visibility decisions are handled separately through the system’s filtering components.
These filters can take into account actions such as blocks and mutes as well as labels applied to posts or accounts by other systems.
In simple terms, X first needs to determine whether content is eligible to be shown and then determine where eligible content should appear.
That helps explain why analyzing a single ranking score cannot provide the full story behind reach on X.
X Is Also Showing More About Account Restrictions
Alongside the increased algorithm transparency, X has also been experimenting with ways to help users understand restrictions affecting their accounts and posts.
Business Insider reports that X introduced a pilot called “Under the Hood,” which can give selected users more information about labels that may affect their visibility.
These could include classifications related to spam or other types of restricted content.
The feature could be particularly important because users have long debated whether their accounts have been “shadowbanned” when engagement suddenly declines.
Making some of these internal labels visible would give users more concrete information instead of forcing them to guess why distribution changed.
What the New Algorithm Details Mean for X Creators
The biggest lesson from the release is that simply chasing likes may not be the best strategy.
X appears to be evaluating a much broader collection of signals to predict whether users actually value a post.
For creators, publishers and brands, that means content designed to generate genuine interaction could become increasingly important.
Original reporting, useful information, thoughtful discussions, images, videos and posts that people actively share may provide stronger signals than low-effort engagement bait.
This direction also fits with X’s broader push toward rewarding original content.
X has separately confirmed on its Creator Revenue Sharing support page that its existing Creator Revenue Sharing program will be retired on September 7, 2026, with creators being directed toward its newer Original Content Rewards program.
The changes suggest X wants both its recommendation and monetization systems to place greater emphasis on content users find genuinely valuable.
Can You Now Predict Exactly What Will Go Viral?
Not really.
Opening the algorithm provides significantly more information, but it does not create a guaranteed recipe for viral posts.
Recommendation systems operate on enormous amounts of personalized data, and the same post can generate very different predictions for different users.
Phoenix evaluates posts in relation to an individual viewer’s behavior.
That means there is no universal ranking score that automatically pushes a post to everyone.
X can also update its models, ranking weights and filtering systems over time.
Creators should therefore treat the open-source code as insight into how the platform works rather than a permanent checklist for gaming the feed.
Why X Opening Its Algorithm Matters
Major social networks rarely provide this level of visibility into their production recommendation infrastructure.
For developers and AI researchers, the release offers an opportunity to examine how a large-scale personalized feed retrieves, scores, filters and rearranges enormous numbers of posts.
For everyday X users, it provides a clearer explanation of why the For You timeline can look dramatically different from one account to another.
And for creators, the message appears increasingly clear: X is looking beyond raw likes.
Shares, conversations, user satisfaction, negative feedback, content eligibility and personalized predictions can all contribute to what ultimately appears in the feed.
The release does not completely solve the mystery of going viral on X, but it makes the machinery behind the For You feed considerably easier to inspect.