What changed in January 2026:

What is the X algorithm update 2026?

In January 2026, X published its full recommendation system to GitHub under the xAI organization. The GitHub write-up notes that the system relies entirely on the Grok-based transformer to learn relevance from user engagement sequences, with no manual feature engineering for content relevance.

This was not a minor update. Before January 2026, X used a legacy ranking system based primarily on engagement metrics and social graph connections. The January release replaced it entirely with something far more intelligent and far more ruthless.

The X algorithm now processes approximately five billion ranking decisions every day, each completing in under 1.5 seconds. Every time someone opens their feed, the system scores every candidate post, predicts how likely that specific person is to engage, and ranks accordingly. It is not reading the room. It is running the math.

How the Grok-powered ranking system actually works

The system operates in three distinct stages. Understanding this pipeline is the foundation of everything else.

Stage 1: Candidate retrieval

The algorithm collects candidates from two distinct sources: in-network tweets from accounts you follow and out-of-network tweets from accounts you do not follow. The algorithm deliberately mixes both types to expose you to diverse content beyond your existing network.

The system pulls posts from two sources: accounts you follow (called Thunder) and random posts from across the platform (Phoenix Retrieval). Thunder handles your following feed. Phoenix handles discovery. Both run simultaneously, and their outputs get merged before scoring begins.

Stage 2: Scoring with Phoenix

Both sources are combined and ranked together using Phoenix, a Grok-based transformer model that predicts engagement probabilities for each post. The final score is a weighted combination of these predicted engagements.

The system has eliminated every single hand-engineered feature and most heuristics. The Grok-based transformer does all the heavy lifting by understanding your engagement history and using that to determine what content is relevant to you.

What the model is actually predicting for every post: probability of a like, reply, repost, quote, click, profile click, video view, share, dwell time, follow after seeing the post, and critically, probability of block, mute, and report. Positive actions have positive weights. Negative actions have negative weights, pushing down content the user would likely dislike.

Stage 3: Filtering

The algorithm removes duplicates, old posts, self-posts, blocked authors, muted keywords, and content deemed too violent or spam-like. Only after all three stages does your post enter the actual feed ranking.

The engagement weights exact numbers from the open-sourced code

This is the section most articles get wrong or skip entirely. The weights are not equal, and the gap between them is enormous.

The simplified scoring formula widely cited from the code: Likes × 1, Retweets × 20, Replies × 13.5, Profile Clicks × 12, Link Clicks × 11, Bookmarks × 10.

But there is a multiplier that changes everything.

If someone replies and the original poster responds back, that interaction jumps to a weight of 75. Translation: one reply where you engage back equals roughly 150 likes in terms of algorithmic value.

Put that in plain terms. If you post and walk away, a reply from someone is worth 13.5 likes. If you reply back to that person, the entire exchange is now worth 150 likes. The algorithm is not rewarding reach. It is rewarding conversation depth.

The most underrated signal is bookmarks. They carry a 5x multiplier weighted at 10 compared to 0.5 for likes. Creating content worth saving reference material, data, frameworks directly boosts algorithmic distribution.

The full weight breakdown in order of impact:

Time decay: why the first 30 minutes decide everything

How quickly your tweet gets engagement after posting is the strongest signal. The algorithm watches the first 30 to 60 minutes closely. A tweet that gets 10 replies in the first 15 minutes will dramatically outperform a tweet that gets 10 replies spread over 24 hours.

A tweet loses half its visibility score every six hours. After 24 hours, algorithmic distribution is minimal. However, if someone with a large following engages with your tweet later, the algorithm can briefly re-amplify it.

This means posting time is not about "optimal hours" in the old social media sense. It is about posting when your most engaged followers are online and available to interact. Your first wave of replies needs to hit fast. The algorithm interprets early engagement velocity as a signal that the content is worth expanding.

The link suppression problem in 2026

This is the change that most creators have not fully processed yet.

External links suffer severe algorithmic depression, especially for non-Premium accounts. Since March 2026, non-Premium accounts posting links receive zero median engagement their link posts are essentially invisible in the algorithm.

Elon Musk confirmed that posts containing links that take users off-platform often see a reach reduction of 50 to 90%. X wants to keep users on the app, so text-only posts or long-form articles generally perform better than posts with outbound links.

The workaround that still holds: post the link in the first reply to your own tweet, not in the main post body. The main tweet gets full algorithmic distribution. The reply carries the link for people who want to click through. This is not a hack, it is how the platform is designed to work now.

The TweepCred score: X's hidden account rating

Every X account carries a TweepCred score from 0 to 100, calculated using a weighted PageRank approach. Most creators have no idea this score exists, but it is one of the most important filters in the entire system.

The score sits below 65 by default for newer and lower-engagement accounts. Drop below that threshold and only a handful of your posts are eligible for broad algorithmic distribution regardless of how good the content is. The system treats you as unverified until your engagement patterns prove otherwise.

What moves your TweepCred score up: consistent posting, high reply-to-impression ratios, replies from high-TweepCred accounts engaging with your content, and low negative signal rates (low blocks, mutes, and not-interested clicks). What tanks it: posting content that generates blocks and reports, long posting gaps followed by sudden bursts, and bot-like behavior patterns.

The small account boost: the most underreported change

One of the most significant algorithm changes in 2026 is the deliberate emphasis on surfacing content from smaller accounts. A tweet from an account with five hundred engaged followers can now outperform a tweet from an account with fifty thousand disengaged followers if the engagement quality signals are stronger.

This is a meaningful shift. The old system used follower count as a credibility proxy. The new Grok-powered system ignores follower count almost entirely in favor of engagement quality ratios. If your replies-to-impressions ratio is strong, the algorithm treats you as credible regardless of your size.

X algorithm cheat codes that still work in 2026

These are not hacks. They are tactics that align with what the algorithm is specifically designed to reward.

Cheat code 1: Reply to every reply within the first hour

Reply to replies. The author engaging with replies is worth 75 the single most powerful signal. Respond to every genuine reply within the first hour. This is not courtesy. It is the highest-leverage action you can take after posting.

Cheat code 2: Create bookmark-worthy content

Bookmarks are 10x more valuable than likes. Reference material, data, frameworks, and checklists drive bookmark behavior. Every post you write should have one element worth saving a stat, a framework, a list, or a tool. If someone bookmarks a post, they have signaled to the algorithm that this content has lasting value.

Cheat code 3: Post the link in the reply, not the tweet

The most effective link-sharing format in 2026: write the main post with a strong hook and full value, then add your link in a self-reply. The original post gets distributed normally. The reply is visible to everyone who clicks through. You lose nothing on reach and gain everything on click-throughs.

Cheat code 4: Use threads for extended algorithmic life

Threads have a slightly longer active window because new replies bump the score. A thread with five posts gives the algorithm five distinct opportunities to find a new entry point, and each new reply on any post in the thread refreshes the score calculation.

Cheat code 5: Spark conversations, not broadcasts

The algorithm rewards reply depth, not broadcast reach. Ask questions, share opinions, invite disagreement. Posts that end with a genuine open question outperform equivalent posts that end with a statement. The algorithm cannot tell the difference between a thoughtful reply and a low-effort one it just counts the signal.

Cheat code 6: Keep hashtags at zero or one

The Grok-powered algorithm reads your tweet content directly. It does not need hashtags to understand what your tweet is about. Excessive use is penalized. Using three or more hashtags signals spam. The algorithm already knows what your post is about from the content itself.

Cheat code 7: Never delete and repost

Deleting and immediately reposting the same content can trigger spam detection. If a post underperforms, let it age out naturally. Reposting the same content fast is one of the clearest spam signals the system watches for.

What the algorithm penalizes in 2026

Worth knowing as clearly as what it rewards.

Obvious engagement farming gets detected. Tweets like "Like if you agree, RT if you disagree" are now recognized as low-quality engagement and may be actively suppressed.

Posting ten tweets in five minutes looks like spam. Space your content out. Spread tweets across the day with natural-looking intervals.

Grok's sentiment analysis reduces distribution for content with a consistently negative or combative tone. You can share controversial opinions. You cannot build a feed that is purely confrontational without the system flagging the pattern.

X Premium vs free reach: the real gap

The algorithm treats Premium and free accounts differently at the distribution level, not just the feature level.

Being paid or verified does not magically guarantee reach, but multiple analyses have found measurable differences in distribution and engagement patterns between Premium tiers and non-paying accounts.

The most concrete advantage is the link suppression gap: Premium accounts posting links see dramatically reduced reach compared to their native content, but they still see some reach. Non-Premium accounts posting links since March 2026 see near-zero median engagement on those posts. If you are sharing external links regularly, Premium removes the hardest penalty even if it does not eliminate it.

Frequently asked questions

Q: What changed in the X algorithm in January 2026?

X replaced its legacy ranking model with a fully Grok-powered transformer called Phoenix. The new system eliminated all hand-coded ranking rules and now learns entirely from user engagement patterns replies, bookmarks, dwell time, and conversation depth.

Q: What are the engagement weights in the X algorithm 2026?

The highest-value action is an author reply to a reply (weight: 75). Reposts are worth 20, profile clicks 12, bookmarks 10, and plain likes only 0.5. One full reply exchange where you respond back is worth approximately 150 likes algorithmically.

Q: Why do tweets with links get fewer impressions?

X actively deprioritizes content that sends users off-platform. Since March 2026, non-Premium accounts posting links in the main tweet body receive near-zero distribution. The fix is to post the link in a self-reply while keeping the main tweet link-free.

Q: How does TweepCred affect my account reach?

TweepCred is a 0 to 100 account quality score. Accounts below a 65 threshold have only a small number of posts eligible for broad distribution in any given period, regardless of content quality. Consistent posting, strong reply ratios, and low negative signals (blocks, mutes, reports) push the score higher over time.

Q: Does the first 30 minutes really matter that much?

Yes. The algorithm uses early engagement velocity as the primary signal for whether to expand distribution. A post that earns replies in the first 15 minutes is treated as higher quality than a post that earns the same replies spread across 24 hours. Post when your engaged followers are online, not at generic "peak hours."

Q: Do hashtags help in 2026?

No. The Grok model understands your content directly without needing hashtags as topic signals. Zero or one targeted hashtag is fine. More than two is actively penalized as a spam signal.

Q: Does X Premium guarantee more reach?

It removes the harshest link suppression penalty and provides some baseline distribution boost, but it does not override content quality signals. A Premium account posting low-engagement content will still underperform a free account posting high-reply-rate content.