X (Twitter) algorithm: how the For You feed works

The X (Twitter) algorithm determines what millions of users see in their “For You” feed every day - and understanding its logic radically changes how you publish. Yet for most content creators and brands, this mechanism remains a frustrating black box: some posts take off for no apparent reason, while others disappear into the void despite obvious care in their writing.

X made a rare decision in the industry: to make the source code of its recommendation algorithm publicly available. This transparency initiative is documented in the official technical post published on X’s engineering blog. Since October 2026, this algorithm has been completely redesigned around a Transformer model based on the same architecture as xAI’s Grok model. This transparency nevertheless remains partial: some files (precise weighting rules and internal prompts) are deliberately omitted to prevent users from circumventing the system.

This article explains the algorithm’s documented mechanisms, the difference between the For You and Following feeds, and the practical levers you can use to improve your organic visibility on X.

For You and Following feeds: two opposing approaches

Since the redesign of the X experience under Elon Musk, the platform has distinguished between two main timeline tabs. Understanding the difference is the essential starting point for any serious content strategy.

The Following feed: chronological and predictable

The Following feed displays posts from accounts you follow in a broadly chronological order. A few promotion signals exist - posts with many interactions move slightly higher - but the principle remains simple: you follow an account, so you see its posts. This feed resembles the Twitter experience most users knew before the major algorithmic redesigns.

The For You feed: algorithmic and dynamic

The For You feed is radically different. It is not limited to accounts you follow - it includes posts from accounts you do not yet know, selected by the algorithm according to your behavioral profile. This is where new creator discovery and organic virality happen. A post can reach hundreds of thousands of impressions through this feed without its author having millions of followers.

The share of posts from unfollowed accounts in the For You feed is deliberately significant - X seeks to maximize time spent on the platform by offering relevant and surprising content. This editorial choice lies at the heart of feed personalization and explains why the algorithm deserves particular attention.

Ranking signals: what the algorithm really evaluates

The X algorithm combines several categories of signals to assign each post a score and decide where it appears in timelines. X’s official technical documentation confirms that these broad signal families exist. The ranking model - called Phoenix - is a Transformer trained to predict the likelihood that you will perform different actions on a given post. The exact weightings of each action are not all public and change during regular updates.

Direct engagement signals

These are measurable interactions with a given post. According to the structure documented by X’s engineering team, the Phoenix model predicts the probability of several different types of user actions, including:

• Likes: a quick approval signal indicating that the content satisfies readers.
• Reposts (retweets) and quotes: strong sharing signals that mechanically expand a post’s audience beyond the author’s followers.
• Replies: a conversation signal indicating that the post generates active discussion.
• Link sharing (copying the post link): according to the weightings made public in the technical documentation, this signal is particularly valued.
• Reading time (dwell time): X measures how long a user pauses on a post before scrolling. A prolonged pause is interpreted as a sign of interest.
• Negative actions: reporting, hiding, or muting an account are also predicted by the model - and heavily penalize a post’s score.

These signals are not all equal. Actions that require a marked voluntary effort from the user - replies, reposts, and link shares - appear to receive greater weighting than simple likes, although the precise values remain partly confidential and subject to change.

Signals related to the author’s account

The algorithm also evaluates the profile of the account publishing the post:

• The account’s age and consistency
• The follower-to-following ratio
• The average engagement history of previous posts
• Whether the account has an X Premium subscription, whose documented effect mainly concerns priority in reply rankings (reply prioritization), according to the different plan levels described in X’s official documentation on the Premium checkmark. The exact scale of this advantage in the For You feed overall has not been publicly quantified.

An account with a history of high interaction receives an initial advantage: its new posts are shown to a larger sample of users to test their early performance before potentially receiving wider distribution.

Post recency: a basic filter

Post recency is a systematic pre-filter. A post published several weeks ago has virtually no chance of appearing in a user’s For You feed unless it continues to generate fresh interactions. The algorithm favors recent content to maintain the perception of a lively, up-to-date platform. Publishing at the right time therefore remains a relevant lever - just as optimizing posting times on Instagram significantly influences post reach.

Personalization: how the algorithm learns your preferences

The For You feed is not the same for any two users. The algorithm builds an individual behavioral profile from your past interactions and browsing history on the platform.

What the algorithm learns from your behavior

Several types of behavior contribute to this profile:

1. The accounts you regularly interact with through likes, replies, and reposts
2. The topics and keywords in the posts you spend the most time reading
3. The types of content you prioritize when liking and commenting, such as videos, threads, images, and long-form text
4. The accounts you have explicitly muted or marked as irrelevant
5. Your search history on X

Based on this data, the algorithm groups users according to thematic and behavioral affinities. A post that performs very well among a subgroup of users with shared interests may be distributed to similar profiles that do not yet follow the author - this is precisely the mechanism that makes algorithmic discovery possible on X.

Negative signals to avoid

The algorithm does not optimize only for positive outcomes - it also penalizes certain behaviors. Negative signals documented in X’s authenticity policy include:

• A user hiding a post from their feed or reporting it as unwanted
• Muting an account after being exposed to its content
• A high rate of quickly scrolling past an account’s posts without pausing, indicating a lack of interest
• Recognized spam behavior, such as mass posting over a short period or publishing duplicate or nearly identical content

Understanding these negative signals is as useful as understanding positive ones: avoiding practices that reduce the algorithmic score is a prerequisite for any improvement in organic reach.

Timeline organization: how the algorithm builds your feed

Building the For You feed is not a simple sort by score. X uses several successive components to compose each reading session.

Post sources: In-Network and Out-of-Network

According to X’s official technical documentation, the For You feed combines two broad categories of content:

• In-Network posts: from accounts you follow - the “Thunder” component in the documented architecture - retrieved in real time according to your subscriptions.
• Out-of-Network posts: discovered by the “Phoenix” component through similarity search in a pool of hundreds of millions of daily posts. These are posts from accounts you do not follow but that the algorithm deems relevant to your profile.

The two streams are merged and then ranked together by the final ranking model. An author-diversification filter is also applied: if the same author appears several times within a single session, the scores of later occurrences are reduced to prevent one account from being overrepresented in your feed.

The final ranking model

Once the candidates have been selected, the Phoenix model assigns a final score to each post. This score is the weighted sum of the estimated probabilities that you will perform each predicted action - like, reply, repost, link share, prolonged pause, and so on - or react negatively through reporting, muting, or rapid scrolling. The final list shown in your feed results from this multi-criteria calculation. The logic is similar to that used by other algorithmic platforms - YouTube and other discovery services apply comparable principles to maximize engagement in their own discovery feeds.

Practical strategies to improve your visibility in the For You feed

Understanding the X (Twitter) algorithm has value only if that understanding leads to action. Here are the best-documented levers for improving the ranking of your posts.

Maximize engagement in the first hour

The first hour after publishing is a critical window. The algorithm uses a post’s initial performance to determine how widely it will subsequently be distributed. A post that quickly generates replies and reposts generally receives broader distribution. Replying to your own posts to sustain the conversation, publishing when your audience is active, and inviting interaction within the text are practices that can positively influence this evaluation phase.

Refine the format and readability

Post reading time is a direct signal in the scoring model. Structured, readable posts that encourage users to pause are favored. A few guidelines:

• Threads - numbered sequences of posts - hold attention longer than standalone posts
• Embedded media, including images and short videos, increases the time users pause on a post
• Open questions in the text encourage replies, a strong signal for the algorithm
• An engaging opening is essential because the truncated timeline display determines whether users click “show more”

Build thematic consistency

The algorithm associates user profiles with topics. An account that consistently publishes about a given subject is more easily recommended to audiences interested in that subject. Excessive thematic variety can dilute this relevance signal. Gradually specializing an account in two or three related areas can strengthen its perceived algorithmic authority.

Control posting frequency

Contrary to popular belief, posting in large volumes is not an advantage on X. The author-diversification mechanism built into the algorithm mathematically reduces the scores of posts from the same account when they appear several times in one user’s reading session. Quality and thematic consistency matter more than raw volume.

Amplify the social signal with complementary services

Initial engagement shapes algorithmic distribution: posts that start strongly receive more exposure, which can generate further organic engagement. To provide an initial boost when launching an account or a new editorial direction, some creators explore options such as buying X (Twitter) retweets or buying X (Twitter) likes to strengthen a post’s social perception. To develop a follower base over time, buying international X (Twitter) followers can expand the reach of each post.

Important: in its authenticity rules, X explicitly prohibits the use of artificial engagement to manipulate platform metrics. These practices may expose an account to reach restrictions or even suspension. Each user must assess these risks and ensure that any approach remains compatible with the applicable terms of service. These levers make sense only when combined with quality content and a consistent editorial strategy.

Frequently asked questions

What is the difference between the For You and Following feeds on X (Twitter)?

The Following feed mainly displays posts from accounts you follow in a roughly chronological order. The For You feed is algorithmic: it includes posts from accounts you do not follow, selected according to your behavior and past interactions to maximize your engagement. This is where new-content discovery and organic virality occur.

How does the X (Twitter) algorithm decide which posts appear first?

The algorithm relies on a Transformer model called Phoenix, based on the same architecture as xAI’s Grok. It predicts the likelihood that you will perform different actions on a post - like, reply, repost, link share, prolonged reading pause, as well as reporting or muting. A weighted score is calculated for each candidate post before the personalized timeline is assembled. The official technical documentation is available on X’s engineering blog.

Do reposts carry more weight than likes in X’s algorithm?

According to information made public in the official technical documentation, actions that involve marked voluntary effort - replies, reposts, and link shares - appear to receive higher weighting than simple likes. The exact weightings of each signal are not fully published and may change during regular model updates.

Does posting frequently on X really improve algorithmic visibility?

Posting frequency is not a direct ranking signal, and excessive posting can even be harmful: the algorithm includes an author-diversification mechanism that reduces the scores of posts from the same account when they appear several times in a single user reading session. Quality and thematic consistency matter more than raw volume.

Does X Premium influence ranking in the For You feed?

According to X’s official documentation, X Premium subscribers receive ranking priority in replies and conversations (reply prioritization), with increasing priority according to the subscribed tier - Basic, Premium, or Premium+. The exact scale of this advantage in the For You feed overall has not been publicly quantified, and X states that priority levels are being tested.

Making the most of the X algorithm: next steps

The X (Twitter) algorithm rewards accounts that generate meaningful interactions, publish consistent content, and hold their readers’ attention. The For You feed is not arbitrary: it follows logic learned from each user’s behavior, and thoughtful editorial choices can influence that logic.

To make tangible progress, start by auditing your posts from the last 30 days: which generated the most replies? Which formats held attention? This data is available directly in X’s native analytics. Identify two or three formats that work, refine your thematic consistency, and optimize your posting times around peaks in your audience’s activity. If you are new to the platform or relaunching an account, exploring the services available on BoostFollowers may provide a way to get started - provided that you have carefully reviewed X’s authenticity rules mentioned above.