The Twitter Algorithm Explained: How the X For You Feed Ranks Posts in 2026

When people talk about the Twitter algorithm, they usually mean the system that decides which posts appear in the For you feed on X, and in what order. It is not one formula. It is a set of systems that collect candidate posts, predict how likely you are to engage with each one, filter out what you should not see, and blend the result into a feed. The details change often, and X does not publish all of them.

This guide sticks to what is publicly known, avoids made-up numbers and turns it into habits you can act on this week, with notes on where GrowX, a Chrome extension for X (Twitter), can help. If someone quotes you an exact multiplier for a like or a reply, be sceptical: X has not published its current weights, and any precise figure you see online is at best a snapshot of an older version.

X has two main timelines, and they work differently:

• For you is the recommended feed. It mixes posts from accounts you follow with posts from accounts you do not follow, chosen because the system predicts you will find them interesting. This is the feed people mean when they talk about the X algorithm.
• Following shows posts from the accounts you follow. It is much closer to a plain timeline, although X has changed how it can be sorted over the years, so check what your app currently offers.

For growth, the For you feed is where new readers come from. Your followers can see you in Following, but people who have never heard of you mostly find you when the system recommends your post to them, or when someone they follow reposts or replies to you.

In 2023, X (then still called Twitter) published parts of its recommendation code, together with a write-up of how the For you feed is built. At a high level, the pipeline it described has three stages:

1. Candidate sourcing. Gather a large pool of recent posts that might interest you, both from accounts you follow and from accounts you do not, based on things like who you interact with and which topics and communities you engage with.
2. Ranking. A machine learning model scores each candidate by predicting how likely you are to engage with it in different ways, such as liking, replying, reposting or spending time on it.
3. Filtering and mixing. Remove posts from accounts you blocked or muted and content that breaks the rules, keep the feed varied so one author does not dominate, and blend in other items such as ads and suggestions.

Two caveats matter. First, not everything was published, and what was published was a snapshot: X has changed the system since and keeps adjusting it. Second, code shows how signals are combined, not what the live model has learned from data. So the useful lessons are about the general shape of the system, not specific numbers.

Putting the public material together with how recommendation systems usually work, a few principles hold up well. None of them comes with a published weight, and X can change any of them:

• Engagement from interested people. Replies, reposts, likes, bookmarks and time spent reading all suggest a post was worth showing. Engagement from people likely to care about the topic matters more than random attention, because the system is trying to predict what each reader will value.
• Conversation. Replies, especially when the author answers back, show that a post started a real discussion rather than collecting a passive tap.
• Relevance. Posts tend to reach people who already engage with similar topics and accounts. A clear niche makes it easier for the system to find your readers.
• Recency. The For you feed favours fresh posts, so the first hours after posting are when a post has its best chance to be picked up.
• Good standing. Accounts that follow X's rules, look authentic and are not flagged for spam tend to do better. X's enforcement approach, which it describes as "freedom of speech, not freedom of reach", can limit the visibility of posts that break certain rules.

The flip side of engagement is negative feedback. When people mute or block you, mark a post as not interesting, report it or unfollow you after seeing it, that tells the system your content was not welcome. X has not published how much each signal counts, but it is safe to assume a pattern of negative feedback works against you.

Behaviour that looks automated or manipulative is risky too, both for reach and for your account. X's rules on platform manipulation and spam cover patterns like these:

• Following and unfollowing large numbers of accounts in bursts.
• Posting the same or nearly the same text again and again, or sending identical replies.
• Stuffing posts with unrelated or trending hashtags to borrow attention.
• Buying engagement or joining groups that trade likes and reposts.

You do not need to know the exact cost of each one to avoid them. These are the patterns spam filters exist to catch.

• "There is a perfect posting hour." Timing helps, but only relative to when your audience is online. See the best time to post on Twitter for how to find yours.
• "Hashtags boost reach." One or two relevant tags can help people find a post about an event or community. Piling them on mostly makes a post harder to read.
• "Every dip means a shadowban." Reach rises and falls with topics, timing and whatever else is happening on X. Look at several weeks of data before concluding something is wrong.
• "Links get zero reach." Many creators believe posts with external links reach fewer people, but there is no published rule you can rely on. If it matters to you, test a link in the post against a link in a reply on your own account.

You cannot control the ranking model, but you can control what you give it. These habits follow from everything above:

1. Post consistently at your best times. A steady rhythm gives the system regular, fresh posts to test, and posting while your followers are awake gives each post a fair first hour.
2. Write for replies. Ask specific questions, take a clear position or share something people want to add to. A post that invites a thoughtful answer starts the kind of conversation ranking rewards.
3. Answer the replies you get, especially in the first stretch after posting, to keep the conversation going.
4. Reply to others early. A useful reply on a fresh post in your niche puts you in front of an audience that already cares about your topic. Our Twitter reply strategy guide covers how to do it well.
5. Avoid anything that looks like spam: no mass follow or unfollow bursts, no repeated posts, no hashtag stuffing, no bought engagement.
6. Measure your own results and keep what works for your audience rather than what a viral thread about the algorithm promised.

Because the details are not public, your own data is the most reliable guide you have. Track a few numbers for each post: views, replies, reposts, likes, bookmarks and the engagement rate, which is engagements divided by views. Engagement rate adjusts for reach, so you can fairly compare a post seen by a few hundred people with one seen by thousands. The free engagement rate calculator does the arithmetic.

Look for patterns over weeks, not single posts. If questions keep earning more replies than links, or evening posts keep beating morning ones, that is a signal worth acting on. One viral post is a nice surprise, not a strategy.

GrowX does not change how X ranks posts, and no tool can. What it does is make the habits above easier to keep:

• Scheduler and best time to post. The post scheduler has queue slots and a week by 24-hour heat map built from your own posts' engagement and from when your followers are active.
• Reply targets. The reply targets view collects fresh posts (younger than 120 minutes by default) from your sources, VIPs and anyone you add, ranked by freshness and how fast they are taking off. GrowX never replies or likes by itself; you write the reply and press the button.
• Analytics. Post analytics show views, likes, reposts, replies, bookmarks and engagement rate for your recent posts (your latest 30 on the free plan), and What works compares topics and formats.
• Safe pacing. Auto follow uses random delays, breaks and hourly and daily caps, and posting has a daily cap (20 by default) and a 5-minute minimum gap. Both slow down or pause when X shows a warning.

GrowX is free to install from the Chrome Web Store, and you can create a free account to try every Premium feature free for 30 days, no credit card needed.