In 2026, X’s algorithm operates through Phoenix, a model that ranks each post according to each user’s probable actions rather than a raw total of likes. Link shares, replies, and direct messages carry more weight than likes, while reports, blocks, and mutes can sharply reduce distribution.
How does X’s algorithm work in 2026?
In 2026, X’s algorithm combines two groups of posts in the “For You” feed: recent posts from followed accounts and content discovered outside the user’s network. According to X’s official documentation published in 2026, Phoenix then assigns a personalized score to each candidate post.
The Phoenix model uses the reader’s recent interaction history. In particular, it predicts the likelihood of a like, reply, repost, share, click, reading time, follow, mute, block, or report.
The final score therefore does not correspond to the number of interactions already accumulated. It results from individual probabilities multiplied by coefficients. Two users can receive radically different rankings for the same post, even at the same minute.
The distinction from the “Following” feed is clear. According to X’s official help documentation in 2026, this non-personalized feed contains only posts from followed accounts, displayed in reverse chronological order. The Phoenix settings detailed here apply to the “For You” feed, whose personalization may also evolve with new X feed control filters.
Which signals provide the most visibility on X?
The signals Phoenix values most in 2026 are actions that indicate strong sharing or conversation. X’s official algorithm GitHub repository publishes a coefficient of 20.0 for copying a link, compared with 5.0 for a reply, quote post, or direct message share.
These values are not “points per interaction.” Since the clarification added by X in August 2026, they multiply the predicted probability that a user will take the action. A high coefficient therefore does not guarantee that a post will rank well if Phoenix estimates that the relevant reader is unlikely to act.
| Predicted action | Coefficient published in 2026 | Marketing take |
|---|---|---|
| Copying a link | 20,0 | Especially strong sharing signal |
| Reply | 5,0 | Conversation likely to extend the session |
| Quote post | 5,0 | Reusing the content with a comment |
| Direct message share | 5,0 | Direct distribution to another user |
| Following the author | 4,0 | Long-term interest in the account |
| General sharing | 2,0 | Circulation beyond the initial impression |
| Repost | 1,0 | Simple public redistribution |
| Like | 0,5 | Positive approval, but relatively light |
The trap many creators overlook involves reciprocal relationships. In 2026, the prediction of a reply receives an additional coefficient of 15.0 when the author and reader follow each other, according to X’s official ranking file. A small community that interacts regularly can therefore produce more relevant signals than a massive but distant audience.
Honestly, chasing likes alone is a dated strategy. A useful post that someone forwards to a colleague, quotes with an opinion, or replies to precisely aligns better with the signals X has published. Native formats, including video replies on X, can support this approach when they add information rather than an artificial reaction.
Why do some X posts disappear quickly?
A post may lose its visibility on X due to its age, repetition from the same author, negative signals, or a safety filter. In the pipeline published in September 2026, candidate posts older than 48 hours are removed before the “For You” feed is ranked.
The algorithm also applies author diversity. The public 2026 parameters indicate a default decay of 0.5 and a floor of 0.25 for repeated posts from the same account. Publishing several messages in a row therefore does not mean getting several identical chances to appear.
Content from accounts the user does not follow also receives a published multiplier of 0.75 in 2026. The same reduction may affect replies and reposts from followed accounts. Out-of-network recommendations remain possible, but they must offset this disadvantage through a strong match with the reader’s predicted interests.
X’s algorithm primarily penalizes likely negative reactions. In 2026, the published coefficients reach −234.0 for a report, −58.8 for muting the author, −43.2 for “Not interested,” and −31.2 for blocking. Again, these numbers weight personalized probabilities; they do not accumulate mechanically like points.
The official repository specifies in 2026 that coordinated blocks or reports made by accessing a post directly do not affect the Home Timeline ranking. To count in this mechanism, the eligible action must take place on a post served in that timeline. This distinction prevents confusion between moderation, reputation account and recommendation score.
How can you improve the reach of your posts on X?
To improve reach on X in 2026, a creator must prompt a useful action without manufacturing artificial engagement: a reasoned reply, private share, contextualized quote, or link copy. X’s algorithm rewards a personalized probability of action, which means publishing for a specific audience rather than for the entire network.
A simple method is to treat each post as a self-contained unit. The reader should understand the topic, identify the value, and know why to reply or share without opening a long thread. A vague hook may generate a few impressions, but rarely the deeper signals Phoenix values.
Here are five actions you can apply to your next post:
- state a verifiable main idea within the first two sentences, without hiding the information behind a promise;
- add an example, dated data point, or method that the reader can forward as is;
- ask a precise question only if a reply genuinely develops the topic;
- reply to comments with additional information in order to build a reciprocal conversation;
- space out closely timed posts to limit the reduction related to author diversity.
Timing should come from your X Analytics data, not a universal formula. The primary sources verified in September 2026 disclose no magic posting time, ideal length, perfect number of hashtags, or fixed formula for the first hour. On X, it is better to test time slots by audience segment and compare impressions, replies, and shares across several comparable posts.
Automation deserves the same discipline. Automatically publish WordPress to X can speed up distribution, but a title and URL alone provide few reasons to reply. Add a thesis, a dated figure, or a concrete consequence to turn the technical relay into editorial content.
Are ranking and visibility filtering identical?
Ranking and visibility filtering are two distinct operations on X in 2026. Phoenix ranks candidate posts according to their predicted relevance, while safety systems may retain a post, remove it from certain surfaces, or display it behind a warning related to spam, abuse, violence, or adult content.
A post with limited reach is therefore not necessarily “shadowbanned.” Low affinity with the audience, older content, excessive repetition, or negative predictions may be enough. Conversely, a visibility label can restrict content independently of its Phoenix score.
In August 2026, X began testing “Under the Hood” with eligible users to display aggregated information about labels affecting accounts and posts. In September 2026, the reports also incorporated restrictions applied to comply with a law, indicating the country where the post is withheld.
This transparency remains partial. The algorithm repository indicates that its default parameters were synchronized with X’s configuration on September 18, 2026, and that experiments affecting significant traffic must appear there. Smaller or unpublished tests may nevertheless exist. X’s algorithm should therefore be understood as a documented, but still experimental, system.
To assess the issue at the campaign level, compare these mechanisms with data on X’s audience in France in 2026. Strong algorithmic reach can never compensate for a poor fit between the platform, the target audience, and the marketing objective.
ValueYourNetwork supports content and influence strategies with a practical, on-the-ground understanding of platforms. Whether you are influencer or an advertiser, grow your social media presence with us and contact us to build a strategy tailored to your goals.
FAQ about X’s algorithm
Do likes still increase visibility on X in 2026?
Likes contribute to ranking on X with a published coefficient of 0.5 in 2026. However, Phoenix weights the personalized probability of a like, not the raw number of likes the post has already received.
How long can a post remain in the For You feed?
Posts older than 48 hours are removed from the candidate pool for the “For You” feed in the ranking system published by X in September 2026. This limit does not mean that a post remains visible for exactly 48 hours to every user.
Does reading time influence X’s algorithm?
Ordinary watch time has a published coefficient of 0.05 in September 2026, while continuous watch time has a coefficient of 0.004. The published parameters then assign a zero coefficient to profile clicks, qualified video views, and continuous watch time after a click.
Should you post multiple times a day on X?
No universal daily frequency has been published by X in 2026. Author diversity gradually reduces repeated posts from the same account, with a default decay of 0.5 and a floor of 0.25, which justifies spacing out very closely timed posts.
Do hashtags improve ranking on X?
No ideal number of hashtags appears as a direct ranking factor in the primary sources verified in September 2026. Hashtags can clarify a topic or make it easier to find, but their presence does not guarantee a better position in the “For You” feed.