How the X (Twitter) Algorithm Works in 2026

X open-sourced a large portion of its recommendation algorithm in 2023. Most people skimmed the headlines. Few actually read the weights. Here is exactly how the algorithm scores content, what signals move the needle, and how to work with it rather than against it.

The algorithm runs in three stages: candidate sourcing (which posts are even considered), ranking (how each post is scored), and filtering (removing content the user has signaled they do not want). Most content never makes it past candidate sourcing. Ranking determines the order of what does.

Understanding these mechanics gives you a concrete playbook. Instead of guessing what works, you can optimize directly for the signals the algorithm rewards most.

The Two Feeds: For You vs. Following

X surfaces content through two primary feeds. The Following feed is chronological: you see posts from accounts you follow in order of time. The For You feed is algorithmic: X ranks content from both accounts you follow and accounts you do not, based on predicted engagement.

For most accounts, the For You feed drives the majority of impressions. It is the mechanism through which content reaches people who do not yet follow you, which makes it the primary growth lever.

The Engagement Weights That Actually Matter

Each action a user takes on a post contributes a different score toward its ranking.

• Reply that generates replies: +75 (150x a like)
• Reply: +13.5 (27x a like)
• Profile click: +12 (24x a like)
• Retweet: +1 (2x a like)
• Like: +0.5 (baseline)
• External link click: -0.5 (negative signal)

A reply is worth 27 likes. A reply that sparks further conversation is worth 150 likes. External link clicks actively hurt your score because X wants users to stay on the platform. This is why threads that end with a product link consistently underperform threads that end with a question.

Where the Algorithm Finds Content to Rank

Before ranking begins, the algorithm assembles a candidate pool of roughly 1,500 posts for each user. These come from two main sources.

In-network posts account for roughly half the pool: posts from accounts the user follows, accounts they interact with frequently, and posts that accounts they follow have engaged with.

Out-of-network posts make up the other half, surfaced through graph-based signals (accounts followed by people you follow) and embedding-based signals (content semantically similar to what you have engaged with). This is the mechanism that allows your content to reach entirely new audiences.

Getting your posts engaged with by accounts whose followers match your target audience is the most reliable way to enter new candidate pools. When a 30,000-follower account in your niche replies to your post, your content becomes a candidate for a significant share of their followers feeds.

What Boosts Reach

Three categories of signals consistently increase reach: engagement velocity, account reputation, and content format.

Engagement velocity in the first 30 minutes: Posts that accumulate replies, likes, and retweets quickly after posting receive a distribution multiplier. This is why posting time matters: whether your core audience is online and likely to engage immediately.

Account reputation score: X maintains an account-level reputation score based on accumulated engagement history. An account that consistently generates high-quality interactions receives a higher baseline distribution multiplier. This score compounds over time.

Content that generates replies: Given the weight table, posts that generate replies are the single most powerful type of content for algorithmic reach. Questions, controversial takes, and incomplete thoughts all outperform declarative statements.

What Kills Reach

• External links in the post body: X explicitly deprioritizes posts that send users off-platform. Put links as a reply to your own thread after it starts getting engagement.
• Low engagement on recent posts: A streak of low-performing posts lowers your near-term distribution score. Quality matters more than raw volume.
• High block or mute rates: When users actively block or mute you, X interprets this as a strong negative signal about content quality.
• Engagement bait: Posts that explicitly ask for likes or retweets without substantive content are demoted, even when they generate surface-level engagement.

The Reply Advantage and Practical Playbook

Replies are both a content type and a distribution mechanism. When you reply to another account post, your reply is shown to a subset of their followers in their For You feed. If your reply generates likes or further replies, it enters more candidate pools. Profile visits contribute +12 per visit toward your account reputation score.

The compounding effect: consistent high-quality replies build your account reputation score, which means your original posts start with better distribution, which means higher early engagement velocity, which means more reach per post. The flywheel starts with replies, even though the payoff shows up on your original content.

Practical steps: post when your audience is active, move external links out of post bodies, end posts with questions rather than summaries, and build a reply list of 20 to 30 accounts in your niche. Tools like XreplyAI help you hit 30 to 50 quality replies per day in your own voice without the blank-page friction.

The X algorithm rewards replies because replies keep users on the platform longer. It rewards account reputation because consistent engagement signals content worth distributing. Once you understand the weights, the strategy writes itself: reply more, link less in post bodies, post when your audience is active, and build the daily habit that compounds over months.

If you want to scale your reply game without sacrificing quality, XreplyAI drafts replies in your voice using your own AI API key. Try it free and see how many more conversations you can start per day.