What X actually published
In March 2023, X open-sourced a large part of the system that builds the For You timeline, in the twitter/the-algorithm repository on GitHub. It covers how posts are gathered, scored and filtered before you see them.
Two caveats matter. The release was a snapshot, and X has changed its ranking many times since. And some parts, such as the trained models' exact behavior and safety rules, can't be fully read from the code. So treat it as a reliable map of how the system is shaped, not a list of current settings. Anyone quoting precise weights as today's rules is guessing.
The three stages of the For You feed
Every time you open the For You tab, the system runs roughly the same pipeline:
- Candidate sourcing. Out of hundreds of millions of posts, it gathers a pool of maybe-relevant ones: posts from accounts you follow, and posts from accounts you don't follow that similar people engage with.
- Ranking. A machine-learning model scores each candidate on how likely you are to engage with it in different ways, such as replying, reposting, liking or clicking into the profile.
- Filtering and mixing. Rules remove posts from accounts you blocked or muted, posts you've already seen and content that breaks policy. They also balance the feed so one author or topic doesn't take over, then mix in ads and other modules.
The practical takeaway: your post first has to get into someone's candidate pool, then win the ranking against everything else in that pool.
In-Network and Out-of-Network
Candidates come from two places, and they work differently.
| Source | What it is | How you get there |
|---|---|---|
| In-Network | Posts from accounts the reader follows. | Earn follows, and give followers a reason to engage so the system keeps showing you to them. |
| Out-of-Network | Posts from accounts the reader doesn't follow. | Get engagement from people whose interests overlap with the reader's. The system uses who-interacts-with-whom graphs and interest communities (the code calls them SimClusters) to find these. |
Out-of-Network is how small accounts grow. A reply from someone in a niche community can carry your post to others in that community. That's why posting clearly within a topic tends to work better than jumping between five unrelated ones.
The signals that shape ranking
The ranking model predicts several kinds of engagement at once. Without quoting weights, here is how they compare in practice:
| Signal | What it tells the system |
|---|---|
| Replies, and replies you answer | The post started a conversation. Generally treated as among the strongest positive signals. |
| Reposts and quotes | People put their own name on it in front of their followers. |
| Bookmarks | Worth saving and coming back to, common for tips and threads. |
| Profile clicks and follows | The reader wanted more from you, not just this post. |
| Likes | Positive but cheap, so on their own they say less. |
| Time spent | People stopped scrolling to read or expand it. |
| “Show less often,” mutes, blocks, reports | Negative feedback. These can hurt your reach to that reader and similar readers. |
What this means for how you write
- Write for replies, not likes. End with something people can answer: a specific question, a clear position, a gap they can fill.
- Reply to your replies. A real back-and-forth is the kind of conversation the system rewards, and it's good manners.
- Stay in a lane. Consistent topics help the system learn which interest communities to show you to.
- Make the first line carry the post. Ranking models learn from whether people stop. See how to write hooks.
- Be careful with links. Posts with external links are widely reported to get less reach. Put the link in a reply, or make the post useful without the click.
- Avoid spammy patterns. Mass-tagging, repetitive posts and engagement bait invite mutes and filters.
- Y
Your name@you
We removed the free trial and replaced it with a 20-minute onboarding call. Signups went down. Paying customers went up. I expected the opposite. Has anyone else tried this? Curious whether it holds outside B2B.
How a post travels, step by step
Putting the stages together, a typical post's life looks something like this:
- You publish. The post becomes a candidate for some of your followers' For You feeds and appears in the Following tab of anyone who follows you.
- Early readers react. Replies, reposts, bookmarks and time spent tell the ranking model the post is worth showing to more of your followers.
- People outside your network engage. Each engagement from someone with known interests makes the post a candidate for similar readers who don't follow you.
- The post competes in more pools. In each one it is scored against everything else those readers could see.
- Interest fades. As engagement slows, the post drops out of candidate pools and reach levels off.
This is why the first hour matters and why replying to early comments helps: each step depends on the one before. It's also why one strong post can bring weeks of new followers while a dozen average ones barely move.
Common myths
- “Deleting a tweet resets your reach.” No evidence for this. Delete posts you regret, not posts that underperformed.
- “Hashtags are required.” Not for ranking. One relevant hashtag can help discovery at events; a string of them looks like spam.
- “There's a magic posting time.” Timing affects early engagement, not ranking directly. See the best time to post on X.
- “You're shadowbanned.” Most reach drops have ordinary causes. See why your reach dropped.