Growth and the algorithm · Article

How the X Algorithm Ranks Posts in 2026

Short answer

X's For You feed gathers candidate posts from accounts you follow and from the wider network, scores them with machine-learning models that predict how you'll engage (replies and conversations weigh far more than likes), filters them, and mixes the result with ads.

Key takeaways

  • The 2023 open-source code showed replies, conversations and profile clicks weighing far more than likes, and negative feedback weighing heavily against a post.
  • X has said it is moving ranking toward its Grok AI models, so exact weights change.
  • What works stays the same: posts people reply to, stay on and don't hide.

"The algorithm" on X mostly means the For You tab: the feed that mixes posts from people you follow with posts from people you don't. The Following tab is simpler, showing accounts you follow in reverse time order. Here's what we know about how For You works.

What are the steps?

Find candidates, rank them, filter them, then mix in ads and other content.

  1. Candidate sourcing. From hundreds of millions of daily posts, X pulls a pool of maybe a thousand or so for you. Roughly half came from accounts you follow (in-network) and half from accounts you don't (out-of-network), found through your interests, similar users and engagement graphs, according to the 2023 description.
  2. Ranking. A neural network scores each candidate by predicting the chance you'll like, reply, repost, click, watch, or react negatively.
  3. Filtering and heuristics. Blocked and muted accounts, posts you've seen, too many posts from one author in a row, and content that breaks rules are filtered or pushed down.
  4. Mixing. Ads, follow suggestions and other modules are slotted in.

Which signals matter most?

In the 2023 open-source code, predicted replies and conversations weighed far more than likes, and negative feedback counted heavily against a post.

Predicted action (2023 config) Relative weight
Like 0.5
Repost 1
Reply 13.5
Open the profile and engage 12
Reply that the author engages with 75
Negative feedback (show less, mute, block) −74
Report −369

Treat these as a snapshot. X has retrained and reweighted since, and in 2025 it said ranking would increasingly run on its Grok models. In April 2026 it launched Grok-labeled Custom Timelines for Premium users. But the direction has been stable: conversation is rewarded, annoyance is punished.

What does that mean for posting?

Write posts people want to reply to, reply to the people who do, and avoid things that make readers hide you.

  • Ask real questions and answer the replies; an author replying to a reply was the strongest signal in 2023.
  • Keep people on the post: images, video and complete thoughts beat teaser links.
  • Avoid spammy tactics (mass tagging, repeated links, engagement bait) that trigger "show less" and mutes.
  • Post when your audience is active; see best time to post.

Does a checkmark help?

Somewhat, especially for replies.

Premium replies are prioritized in conversations, and X's creator payouts count impressions from verified users, which nudges the system toward subscriber engagement. If your reach dropped suddenly, check for limits first; see shadowbans explained.

Questions people ask

Does X Premium boost my posts?

X prioritizes Premium accounts' replies in conversations and has said subscribers get some ranking benefit. It doesn't make a weak post do well.

Do links hurt reach on X?

X leadership has said posts that keep people on X do better, and many creators observe lower reach for posts with external links. Putting the link in a reply is a common workaround.

Is the X algorithm still open source?

X published its recommendation code on GitHub in March 2023 and has posted updates since, but the production models and weights change and aren't all public.

Sources

  1. X engineering: Twitter's recommendation algorithm (Mar 2023)
  2. GitHub: twitter/the-algorithm
  3. TechCrunch: Hands-on with X's AI-powered custom feeds (Apr 2026)