X algorithm: how the For You feed works in 2026

Short answer

The X algorithm is not one magic formula.

It is a recommendation system. X uses it to decide which posts show up in For You, which replies rise in a conversation, which accounts get recommended, and which posts appear in Search, Explore, notifications, and other surfaces.

X's own recommendation pages describe the same basic pattern: find content from inside and outside a user's network, rank it with many signals, filter out content that should not be recommended, and keep learning from behavior. The signals include accounts and Topics someone follows, posts they like, posts liked by people in their network, reposts, replies, watched media, and other interaction patterns.

The creator takeaway is blunt: X is trying to classify both the post and the person likely to care about it.

So the practical work is not "beat the algorithm." It is to make your account easier to understand, easier to recommend, and less likely to trip spam or quality filters.

Build a cleaner X growth system in PostFury if you want one place to collect ideas, draft posts, schedule a steady queue, and review what actually earned reach.

Why this article says X and Twitter

The platform is X. A lot of people still search "Twitter algorithm."

That is why this guide uses both terms. If you searched for "Twitter algorithm," you are probably asking about the current X For You feed, not the old chronological Twitter timeline.

That mixed language is also why so many algorithm guides age badly. They rewrite the same Twitter advice every year, then bolt on "X" in the title. The current topic needs current sources: X's recommendation docs, the public algorithm repositories, and the latest 2026 mutuals change.

This is not a place for folklore.

How X explains recommendations

X's recommendation policy page says recommendations appear in places such as the Home timeline, Explore, notifications, Topics, email, and other surfaces. It also says recommendations use many signals and that no one signal has a fixed permanent weight.

X's For You timeline page says For You includes posts from accounts and Topics a user follows, plus recommended posts. X says it selects posts using signals such as popularity and how people in your network are interacting with them.

X's For You recommender page is more direct. It says X finds content from inside and outside your network, then ranks relevance with a neural network trained on interactions such as likes, reposts, and replies.

That gives creators a better mental model:

You cannot control the model. You can control the signals you keep sending.

What the open-source algorithm tells us

X has two public repositories worth knowing:

• twitter/the-algorithm, the older public repository for X's recommendation algorithm.
• xai-org/x-algorithm, the newer For You feed algorithm repository.

The older repository describes the recommendation system as services and jobs that serve feeds across X product surfaces, including For You, Search, Explore, and notifications.

The newer xai-org/x-algorithm repository says it powers the For You feed. Its May 15, 2026 update added an end-to-end inference pipeline and new components for content understanding, ads, and candidate sourcing. The Phoenix README describes a recommendation system that handles retrieval and ranking with transformer-based architecture.

That does not mean every live ranking rule is sitting in GitHub for marketers to exploit.

It does mean the simple version is wrong. X is not just counting likes. It is retrieving candidates, understanding content, ranking predicted engagement, using network context, and applying filters.

For creators, the useful lesson is this:

The algorithm does not know your positioning unless your posts and interactions keep proving it.

The July 2026 mutuals change matters

The most recent practical change creators should know is the mutuals update.

On July 13, 2026, TechCrunch reported that X adjusted its algorithm to boost visibility for mutuals: accounts you follow that follow you back. X's head of product, Nikita Bier, described it as a fix for replies feeling too dominated by people you do not recognize.

PostFury already has a deeper guide to the X algorithm update around mutuals, replies, and reach.

The short version: relationships are part of distribution again.

Do not turn that into follow-back farming. The useful version is smaller and more boring:

• Follow people whose work you actually want in your feed.
• Reply when you can add context, proof, disagreement, or a useful question.
• Publish posts that people in your niche would recognize as specific.
• Track whether those interactions lead to profile visits, follows, and better replies.

Random mutuals make your graph noisy. Real mutuals make your niche clearer.

What helps posts get recommended

Nobody outside X can promise a fixed ranking formula.

But the public documentation and repository architecture point in a consistent direction. The algorithm needs to answer a few questions:

• What is this post about?
• Who might care about it?
• Is the author trusted or relevant to that audience?
• Are people interacting in useful ways?
• Is the content safe enough and high quality enough to recommend?
• Has this user shown interest in similar topics, accounts, media, or conversations?

For a creator, that turns into five operating habits.

1. Make the account easy to classify

If one day you post founder lessons, the next day politics, the next day memes, the next day crypto, and the next day personal productivity, X may still show some posts to people. But your account becomes harder to place.

Clear accounts have recurring topics. They repeat useful language. They build a recognizable promise.

That does not mean every post sounds identical. It means a stranger can look at your last 20 posts and understand why someone would follow you.

2. Write for replies, not just likes

Likes are easy. Replies are richer.

A useful reply tells X and you that the post created enough context for someone to join the conversation. It also gives you new language, objections, examples, and follow-up ideas.

If your posts get likes but no replies, try sharper prompts:

The goal is not bait. The goal is a post with enough shape for someone to answer.

3. Use analytics as feedback, not a scoreboard

X Analytics exists to show how an audience responds to content and what is working. X's developer docs also separate public metrics from private owner-level metrics such as clicks, and note that some non-public metrics are only available for recent posts.

That split matters. You can study public likes, replies, reposts, quotes, and views from the outside, but your own strategy should use owner-level signals too.

If you need the deeper framework, read PostFury's guide to X analytics metrics for creator growth.

For algorithm work, review these weekly:

Do not overreact to one post. Look for patterns.

4. Keep the queue steady, not robotic

Scheduling does not hurt reach by itself. Robotic posting does.

A healthy queue gives you enough surface area to learn without flooding people. A weak queue turns the account into recycled hooks and dead templates.

PostFury's guide to growing on X without posting like a bot covers the workflow in more detail. The algorithm-friendly version is simple:

• Plan topics before hooks.
• Keep a mix of short posts, examples, opinions, replies, and proof.
• Leave room for live context.
• Review old queue posts before they publish.
• Do not schedule stale takes into breaking conversations.

Consistency helps when it makes your account clearer. It hurts when it makes you sound manufactured.

5. Avoid behavior that looks manipulative

X's reach-limited content page says X uses behavior-based signals such as who you follow, what conversations you join, and whether people mute or block you. Its authenticity policy also warns against unauthorized automation and inauthentic behavior.

Creators should take that seriously.

Avoid:

• duplicate posts across accounts;
• burst posting from a new or quiet account;
• empty reply automation;
• follow/unfollow tactics;
• engagement pods;
• misleading media;
• excessive mentions to people who do not respond;
• queues that publish long after the context has expired.

If your reach is already down, read the Twitter shadowban test guide before assuming the algorithm is punishing every post.

X algorithm checklist for creators

Use this before blaming reach on the algorithm:

This is also why "best time to post" advice is incomplete. Timing can help, but it cannot save a vague account, weak post, or noisy relationship graph. Use the X posting schedule guide for rhythm, then use analytics to tune the rhythm to your own audience.

A simple weekly workflow

Here is a practical weekly loop:

This is boring on purpose. Boring systems beat algorithm panic.

If you publish through tools or automation, also read X rate limits for posting in 2026. A good growth workflow should respect platform limits instead of trying to sneak around them.

What to do next

If you want to work with the X algorithm in 2026, start here:

1. Pick three topics your account should be known for.
2. Draft posts that prove those topics with examples, opinions, and lived context.
3. Reply to relevant people before you need them to notice you.
4. Schedule a steady queue, but keep human review in the loop.
5. Review analytics weekly and keep the posts that attract the right people.

The algorithm is not your audience. It is the sorting layer between your work and your audience.

Treat it that way.

Use PostFury to plan a steadier X queue when you want drafting, scheduling, queue review, and analytics thinking in the same workflow.