X/Twitter Algorithm Changes: The Complete Guide (2024-2026)
Every major X algorithm change from 2024 to 2026 explained. Engagement weights, Premium boosts, link penalties, Community Notes impact, and the new Grok-powered ranking system.
How the X Algorithm Actually Works
The X algorithm determines which posts appear in your "For You" feed, how they are ranked, and how far they travel beyond your existing followers. Understanding it is not optional if you are serious about growing on the platform. Between January 2024 and January 2026, the algorithm underwent dramatic changes — from incremental policy tweaks to a complete architectural overhaul powered by Grok AI.
This guide covers every significant change, explains the current state of the algorithm, and shows you exactly how to adapt your strategy.
The Engagement Weight System
The foundation of X's ranking algorithm is a numerical weight system that scores every interaction differently. These weights were first revealed when Twitter open-sourced its recommendation algorithm in March 2023, and they were confirmed as largely unchanged in the September 2025 algorithm documentation update.
Current Engagement Weights (Out of 100)
Reply where the author re-engages: +75
This is the single most important signal in the entire algorithm. When someone replies to your post and you reply back, creating a genuine conversation, the algorithm registers this as the highest quality interaction possible. One back-and-forth exchange is worth 150 times more than a like.
Reply: +13.5
Even without the author responding, a reply is the second strongest signal. Posts that generate replies get significantly more distribution than posts that only get likes.
Profile click followed by engagement: +12
When someone sees your post, clicks through to your profile, and then likes or replies to something else, the algorithm treats this as deep interest in you as a creator.
Conversation click followed by engagement: +11
When a user clicks into the full conversation thread and then engages (replies or likes within the thread), this signals high-quality, substantive content.
Conversation dwell time of 2+ minutes: +10
Even without any visible engagement, the algorithm tracks how long users pause to read your content. Two or more minutes of dwell time is a strong positive signal — proof that your content held attention.
Retweet: +1
Surprisingly low. Most people assume retweets are the most valued action, but the algorithm weights them at just 1 out of 100. They still matter for distribution (they put your content in front of new audiences), but they carry far less algorithmic weight than conversation.
Like: +0.5
The baseline positive interaction. Likes are easy to give and the algorithm values them accordingly — as a mild positive signal, nothing more.
Video watch 50%+: +0.005
Extremely low individual weight, but video gets distribution advantages through other mechanisms (see the video section below).
Negative Signals
The algorithm also tracks negative interactions, and these carry enormous weight:
- "Not interested" click: -100x penalty — A single user marking your content as irrelevant outweighs 200 likes
- Block or mute: -1000x penalty — The strongest negative signal. Multiple blocks from different users can crater your reach
- Report: triggers manual review — Can result in temporary or permanent reach restrictions
- Offensive content detection: -80% reach — The algorithm scans for offensive text, ALL CAPS abuse, and other quality indicators
What This Means for Your Strategy
The engagement weights reveal a clear hierarchy: conversation is everything. Your primary goal should be creating content that generates replies, and then actively responding to those replies. A post with 50 thoughtful replies will dramatically outperform a post with 500 likes in terms of algorithmic distribution.
Major Algorithm Changes: 2024
Q1 2024: Premium Subscriber Advantages Solidified
Throughout early 2024, the premium subscription tiers became a meaningful factor in content distribution. X Premium subscribers at all tiers receive confirmed algorithmic advantages:
- Initial reach multiplier of 2-4x: Premium accounts get shown to 40-80% of their followers initially, compared to 10-20% for free accounts
- Reply ranking priority: Premium replies appear first in all threads, regardless of the reply's engagement metrics
- For You feed priority: Premium users have a higher probability of appearing in non-follower feeds
- Engagement from Premium users carries more weight than engagement from free users
The practical implication: free accounts need roughly 50% higher engagement to compete for equivalent placement against a Premium account. For serious creators, Premium ($8/month) is now a near-required investment.
Q2 2024: Video Content Priority Expanded
X significantly boosted the algorithmic reach of video content throughout mid-2024:
- Video posts receive approximately 2-4x more reach than equivalent text or image posts
- Vertical video (9:16 format) performs better than horizontal
- Video completion rate became a ranking factor — videos watched past 50% get additional boost
- The sweet spot for length is 15-60 seconds with captions
- According to X's own data, 4 out of 5 user sessions now include video watching
This was a strategic move to compete with TikTok and YouTube Shorts. Creators who added video to their content mix saw immediate reach improvements.
Q3 2024: Follower Quality Detection
X launched systems to detect artificial follower inflation:
- The algorithm now analyzes follower-to-view ratios. An account with 10,000 followers averaging 100 views per post gets flagged
- Suspected follower buying triggers a 50-70% reach reduction
- Flagged accounts are excluded from the For You feed entirely
- Recovery takes 30-60 days of clean, organic engagement
- This made buying followers not just useless but actively harmful to your account
Q4 2024: Thread Treatment Updated
X changed how threads are processed by the algorithm:
- Threads are now recognized as a single content unit rather than individual tweets
- Thread completion rate became a new ranking factor
- High completion rate boosts the entire thread in distribution
- The first tweet gets extra reach if the overall thread performs well
- Optimal thread length: 3-5 tweets (achieving 70-80% completion)
- Threads get 40-60% more total impressions than equivalent standalone tweets
Major Algorithm Changes: 2025
January 2025: Community Notes Impact on Ranking
X integrated Community Notes directly into the ranking algorithm:
- Tweets tagged with Community Notes receive a 60-80% reach reduction
- Noted tweets are removed from For You feed recommendations entirely
- Distribution is limited to chronological follower feeds only
- Accounts with multiple noted tweets receive an overall authority score reduction
- All future posts from repeatedly flagged accounts start with lower baseline reach
- Recovery after correction takes 7-14 days
This was a significant change because it means factual accuracy now directly affects your algorithmic reach. Sharing misleading information is not just reputationally harmful — it mechanically reduces your visibility.
May 2025: External Link Penalty Peaks
The external link penalty, which had been increasing since 2023, reached its maximum severity:
- Single external link: up to 50% reach reduction
- Multiple links in one post: up to 70% reduction
- Links to competitor platforms (Instagram, Facebook): up to 60% reduction
- Shortened URLs (bit.ly, etc.): additional -10% as spam signal
Creators adapted by posting links in the first reply rather than the main tweet, or by using "link in bio" approaches. This workaround period was frustrating but temporary.
September 2025: Updated Algorithm Documentation
X Engineering (@XEng) published an updated overview of the recommendation codebase on GitHub. Key confirmations:
- Engagement weights had remained largely consistent since the 2023 open-source release
- The boost/penalty factors list was expanded to include:
- Boosts: Video (especially 10+ seconds watch time), dwell time, quotes, bookmarks, verified/Premium status, and momentum for creators whose recent posts performed well
- Penalties: Offensive text (-80%), offensive username, ALL CAPS abuse, external links (at the time), and negative user reactions (show less often, block, mute, report)
- Elon Musk noted during this update that Grok AI assessments would "soon replace" the existing code
October 14, 2025: External Link Penalty Removed
In one of the most celebrated changes of the year, X announced the complete removal of algorithmic penalties on posts containing external links.
- Early data showed approximately 8x increase in link post reach compared to the penalized period
- Click-through rates improved significantly
- Creators could once again share articles, products, and resources without reach punishment
- The "link in first reply" workaround became unnecessary
This change reflected a philosophical shift: instead of trapping users on-platform, X moved toward being a content distribution hub.
November 30, 2025: Following Feed Becomes Algorithmic
This was one of the most controversial changes. X began using Grok AI to sort the Following feed:
- The Following feed now ranks posts by predicted engagement and relevance instead of showing them in chronological order
- Users are defaulted to the For You feed when opening the app, even if they previously closed on the Following tab
- A chronological view is still accessible but is no longer the default
- The change is less impactful for accounts following fewer than 200 people
Critics warned this could undermine X's strength as a real-time platform, since breaking news and live events benefit from chronological feeds. For creators, the change means your content competes for attention in every feed, not just For You.
January 2026: The Grok-Powered Algorithm
What Changed
On January 20, 2026, X released an entirely new recommendation algorithm on GitHub (xai-org/x-algorithm) under the Apache 2.0 license. This was not an update to the existing system — it was a complete replacement built on xAI's Grok transformer architecture.
How It Works
The new algorithm uses an end-to-end machine learning approach:
-
No hand-engineered features. The previous system used manually coded rules and feature engineering. The new system is a pure ML model that learns ranking entirely from user behavior patterns.
-
Two-tower retrieval model. A User Tower encodes each user's features and engagement history. A Candidate Tower encodes all posts. The system performs similarity search to find relevant out-of-network content.
-
Multi-action prediction. The Grok model predicts the probability of every possible user action simultaneously: P(favorite), P(reply), P(repost), P(quote), P(click), P(profile_click), P(video_view), P(photo_expand), P(share), P(dwell), P(follow_author), P(not_interested), P(block_author), P(mute_author), P(report).
-
Weighted scoring. Final Score = Sum of (weight × P(action)), where positive actions have positive weights and negative actions (block, mute, report, not interested) have negative weights.
-
Author diversity. Repeated posts from the same author receive attenuated scores to ensure feed diversity.
What This Means for Creators
The shift to a pure ML system has several practical implications:
- No more exploitable rules. With hand-coded features, creators could game specific signals (posting at certain times, using certain formats). With ML-based ranking, the system adapts to identify and devalue gaming patterns.
- Quality content genuinely wins. The system learns from actual user behavior. If users engage with your content meaningfully (reply, dwell, bookmark), the algorithm will distribute it further — regardless of whether you followed any "algorithm hack."
- Consistency matters more than ever. The model builds a profile of each creator based on sustained behavior. One-off viral posts matter less than consistent engagement patterns over time.
- Musk committed to updating the open-source code every 4 weeks, which means unprecedented transparency into how the algorithm evolves going forward.
How the For You Feed Pipeline Works
Understanding the technical pipeline helps explain why some posts succeed while others stall:
-
Candidate Sourcing: The algorithm pulls approximately 1,500 candidate tweets from two sources — accounts you follow (in-network) and accounts you don't follow but might like (out-of-network). The Grok model handles the out-of-network retrieval.
-
Ranking: The ML model scores each candidate based on predicted engagement (using the multi-action prediction system described above). Posts likely to generate meaningful engagement rank higher.
-
Filtering: The system removes content from blocked or muted accounts, balances the ratio of followed vs. recommended content, and limits posts from any single author to ensure diversity.
-
Mixing: Organic content is blended with sponsored posts to create the final feed.
The most important stage for creators is Stage 2: ranking. Your content needs to generate strong predicted engagement to rise in the candidate pool. This is why building a history of high engagement matters — the model uses your track record to predict how your new posts will perform.
Actionable Strategy for the Current Algorithm
Based on everything above, here is what you should prioritize:
1. Optimize for Replies Above All Else
The single highest-value action is a reply where you engage back (+75 weight). Structure your content to invite responses:
- End posts with genuine questions
- Share opinions that invite respectful disagreement
- Ask your audience to share their experience
- Always respond to replies within the first hour after posting
2. Invest in X Premium
The 2-4x reach multiplier for Premium accounts is too significant to ignore. At $8/month, it is one of the most cost-effective growth investments on any social platform.
3. Add Video to Your Content Mix
Video content gets 2-4x algorithmic reach over text. You don't need production quality — screen recordings, talking head clips, and simple explainer videos all count. Keep them 15-60 seconds, vertical, with captions.
4. Write Threads of 3-5 Posts
Threads now get 40-60% more total impressions than standalone posts. Keep them tight (3-5 tweets for optimal completion rate) and make sure the first tweet works as a standalone hook.
5. Build Authority Through Consistency
The Grok-powered algorithm builds a model of each creator over time. Consistent posting with sustained engagement trains the system to distribute your content more aggressively. There is no shortcut — profile authority accumulates through months of showing up.
6. Share Links Freely Again
With the October 2025 link penalty removal, you can share articles, resources, and product links without worrying about reach suppression. Pair links with compelling commentary for best results.
7. Avoid Negative Signals
A single block outweighs 2,000 likes in algorithmic impact. Focus on content that does not generate negative reactions:
- Be accurate with facts (Community Notes penalties are severe)
- Avoid unnecessary antagonism
- Don't buy followers (the detection system is active)
- Don't spam DMs or replies (reported accounts lose authority)
Looking Ahead
With the algorithm now powered by Grok and the codebase being updated every four weeks, we are entering an era of unprecedented transparency and rapid evolution. The fundamental principle remains constant: create content that generates genuine, meaningful engagement from a targeted audience. The algorithm is designed to find and amplify exactly that. Make your content worth amplifying, and the algorithm becomes your greatest growth tool.