How the X (Twitter) algorithm works, and how to work with it

The short version

The For You feed is built by a recommendation system. It collects posts from people you follow and from people you do not, scores them for how likely you are to engage, and ranks them. X has published parts of its code, so you can read how it works in outline.

Early reaction matters

A new post is usually shown to a small group first. If they react, with replies, reposts or likes, it can be shown to more people. If they do not, it fades. That is why the first period after posting matters.

Habits that help

• Start conversations. Replies, especially back-and-forth ones, tend to count for more than a like.
• Open with the point. A weak first line is scrolled past.
• Mix formats. Text, images and video behave differently, so test.
• Be around after you post to answer replies.
• Avoid repeated spammy behaviour. Mutes, blocks and reports can lower your reach over time.

What about paid accounts?

Reports and public analyses suggest that paid accounts get more reach than free ones. That may change, so treat it as something to measure in your own results, not a rule.

Work with the algorithm, not for it

Algorithms change, so build around things that last: a clear point of view, useful posts and a rhythm. Content Postiz helps you keep that rhythm with a queue, and your AI can draft posts and threads for you to approve.

How the recommendation system picks posts

The For You feed is built in stages. First the system gathers candidate posts from people you follow and from accounts you do not follow. Then it scores them, using a model that estimates how likely you are to react, such as by liking, replying or reposting. Finally it filters and orders the results, mixing in things like ads and recommendations.

X has published parts of the code, so people outside the company have studied it. It is a useful reminder that the system is made of many parts, and that no single trick controls it.

What your own history changes

The feed you see is personal. It depends on who you follow, what you reply to and what you spend time reading. The same applies to your audience: a post is shown to people whose past behaviour suggests they might like it. That is why a clear topic helps. If your account is about one subject, the system can find the right people for it.

Write posts that invite replies

Replies are usually a stronger sign of interest than a like. You can invite them without begging.

• Ask a genuine question your audience can answer from experience.
• Share a view that is clear enough to agree or disagree with.
• Leave a gap for a story: “what is the worst advice you were given?”
• Reply to the replies you get, which keeps the conversation going.

Formats to try

Test text, images, video and threads, since the system may treat them differently and your audience may prefer one. Keep a note of how each does for you. A clear image or a short video can stand out in a busy feed, while a well-made thread can earn reposts and follows. There is no format that always wins. Your own results are the guide.

What can hurt your reach

• Repeating the same post or very similar posts again and again.
• Behaviour that looks automated or spammy.
• Posts that many people mute, block or report.
• Chasing arguments and rage, which may bring reactions but costs trust.
• Long gaps, which give the system less to learn from.

A weekly plan that works with the system

• Post a few times a day on your main topic.
• Reply to ten or fifteen people in your field each day.
• Publish one thread a week on something you know well.
• Review your best posts on Friday.
• Try one small experiment each week and note the result.

Be sceptical of algorithm advice

Every few months a new set of rules goes around, often with confident numbers attached. Be careful. The system changes, and advice built on one version goes stale. Rely on what your own account shows you, and on habits that have always worked: clear writing, real value and being a good conversation partner.

What the open code can and cannot tell you

X has published parts of its recommendation code, and engineers and researchers have written about it. That is useful, but it has limits. The published code shows the structure of the system and the kinds of signals it considers. It does not show the exact weights, which change over time, or the live data the models learn from.

So treat articles that claim to give a precise formula with care. They are usually reading a snapshot, and the numbers they quote may be out of date by the time you read them. The safe lessons are the broad ones: the system tries to predict what you will react to, it learns from behaviour, and it filters out content people dislike.

Writing for the first thirty seconds

Because a new post is tested on a small group first, the opening of the post matters a great deal. The first line should make clear what the post is about and why someone should care. Avoid starting with a greeting or a vague statement.

If you add an image or a short video, make sure it makes sense at a glance. Many people scroll quickly, and a clear picture can stop them. Add alt text, which helps people who use screen readers and gives the system more context about the image.

Using threads well

A thread lets you tell a longer story, and a good one can earn reposts and follows. Make the first post stand on its own, since it is the one most people see. Number the posts or use a clear structure so readers know where they are. Keep each post short and give each one a single point.

End with something useful: a summary, a link, a question or an invitation to follow. If you want to share a link, putting it in the final post or in a reply keeps the main post clean.

Content Postiz can schedule threads of up to 25 posts, so you can write the whole thing in one sitting and place it in your queue.

Measure what matters to you

It is easy to chase numbers that look good but do not matter. Decide what you want from X. If it is conversation, watch replies. If it is traffic to your site, watch link clicks. If it is a bigger audience, watch follower growth among the people you want to reach.

Then check once a week. Which posts did best against your goal? What did they have in common? Use the answer to plan the following week, and repeat. Over time, this habit teaches you more than any article about the algorithm.

Think about the people, not only the system

It is easy to become so focused on the algorithm that you forget the people it serves. Every post is read by a person with a few seconds to spare and a reason to be on the app. If your post is clear, useful or funny, they will react, and the system will notice.

Write for one reader you can picture: a customer, a colleague, a younger version of yourself. Ask what they would find useful and say it plainly. That habit produces better posts than any trick, and it is the part of the process that does not change when the system does.

Finally, be patient. Accounts rarely grow in a straight line. Give a new approach several weeks before you decide it has failed, and keep notes so you can see what changed.

Build an audience that would follow you anywhere

Algorithms are borrowed ground. They change, and an account that depends on one version of a system is fragile. The safer plan is to build an audience that has a direct connection to you: an email list, a newsletter, a community or a site people visit by name.

Use X to meet people and to share ideas, and give those who like your work a way to stay in touch outside the feed. Mention your list or your site now and then, and make it easy to join. If the system changes next year, the people who chose to hear from you will still be there.

This habit also changes how you write. When you aim to earn a long-term relationship, you write posts that are honest and useful, instead of posts that merely chase a reaction. That is better for your readers and, over time, better for your results.

Frequently asked questions