A reply is weighted as 27× a like
The single most consequential number in X's algorithm. Replies signal genuine engagement; likes are passive.
Source: X open-source ranker (github.com/twitter/the-algorithm) · 2023 release, still confirmed in 2026
Algorithm · Data from the open-source ranker
X (Twitter) algorithm statistics 2026
X open-sourced its ranking algorithm in March 2023. These are the specific engagement weights, velocity curves, and reach-decay numbers from the code + observed performance — the data that explains why some posts go viral and others die.
Last updated 2026-05-17
Most 'X algorithm' content online is speculation. The numbers below come from X's open-source ranker repo (github.com/twitter/the-algorithm) and observed performance across thousands of tracked tweets. Every claim has a source.
Key takeaways
- ▸Replies = 27× a like in X's algorithm. The single most important engagement signal. Write tweets that provoke replies.
- ▸First 30 minutes determine ~70% of a post's eventual reach. Engagement velocity beats absolute engagement count.
- ▸External links get 30-40% less reach. Put links in the first reply, not the main post.
- ▸Threads (5-9 tweets) earn ~3× engagement of single tweets. Native video earns 40-60% more than text.
- ▸X Premium gives ~10-25% reach lift in replies. Modest but real, especially for accounts focused on reply-based growth.
Engagement weight constants (from the open-source code)
X's ranking algorithm assigns numerical weights to each type of engagement signal. Higher weight = more amplification.
A retweet (repost) is weighted as 1× a like
Often surprising — retweets aren't as algorithmically valuable as conventional wisdom suggests.
Source: X open-source ranker · 2023 release
A bookmark is weighted heavily (relative value undisclosed publicly)
Bookmarks signal 'I want to come back to this' — the strongest possible quality signal. Specific weight not in the open-source repo but observed performance confirms heavy weighting.
Source: Industry consensus + observed data · 2026
Dwell time (read duration) outweighs all click-based signals
Per X's algorithm leads, dwell time is the most heavily weighted signal of all. Long posts and threads that earn read-through compound this signal.
Source: X public statements + observed data · 2026
A 'profile click' is weighted ~12× a like
If a viewer clicks through to your profile from a post, the algorithm treats that as strong interest. Profile clicks drive follower conversion at much higher rates than direct engagement.
Source: X open-source ranker · 2023 release
Engagement velocity curves
X's algorithm scores posts based on engagement RATE in the first 30-60 minutes. Posts that earn fast engagement get amplified to far more users than posts that earn the same engagement slowly.
First 30 minutes determine ~70% of a post's eventual reach
If a post doesn't earn engagement in the first half-hour, it likely never will. The algorithm scores velocity and decides scale.
Source: AutoTweet research · 2026
Posts gain ~80% of their lifetime impressions in the first 2 hours
The recency cliff. By hour 6, reach has dropped 80%+. By 24 hours, near zero for non-viral content.
Source: AutoTweet research · 2026
Viral posts (>10× normal reach) typically break out within 60 minutes
If a post is going to go viral, it shows in the first hour. Slow-burn viral is rare.
Source: AutoTweet research · 2026
Posting during your audience's peak online window can 2-3× reach
Same content, same hook — posting at peak vs. off-peak windows produces 2-3× difference in total impressions due to engagement velocity.
Source: AutoTweet research · 2026
Content format performance
Different content formats earn different reach. The numbers below are relative — how each format performs compared to a baseline single-tweet text post.
Threads (5-9 tweets) earn ~3× the engagement of equivalent single tweets
Driven by dwell time + multiple opportunities for engagement signal. Threads that complete are massively amplified.
Source: AutoTweet research · 2026
Posts with external links get ~30-40% less reach than identical link-free posts
X actively down-weights posts that drive traffic off-platform. The workaround: link in first reply, not main post.
Source: AutoTweet research + creator-shared data · 2026
Posts with images earn ~15-25% more reach than text-only equivalents
Modest lift — images aren't a 'cheat code' but they help when relevant. Stuffing irrelevant images doesn't help.
Source: AutoTweet research · 2026
Posts with native video earn ~40-60% more reach than equivalent text
X aggressively boosts native video to compete with TikTok/Reels. The lift is real but requires actual video content.
Source: AutoTweet research · 2026
Polls earn the highest engagement variance — wins are 2-3× normal, flops are -50% normal
Polls are double-or-nothing — voting is high-signal engagement when it happens, but low vote counts signal 'low engagement content' worse than a non-poll.
Source: AutoTweet research · 2026
Account-level signals
Beyond per-post engagement, the algorithm scores your account as a whole. Recent quality signals affect your future posts.
Account-level engagement averages persist for 30+ days
Recent post performance affects future post reach. A flop today reduces the algorithmic lift on tomorrow's post.
Source: X open-source ranker + observed data · 2026
X Premium (verified) accounts get ~10-25% reach lift in replies
Verified accounts are explicitly ranked higher in reply threads. The boost is modest but real.
Source: X public statements · 2026
Accounts that post 3-5×/day outperform accounts that post 1×/day by 4-6× in follower growth
Frequency compounds. More posts = more opportunities for algorithm signal + more touchpoints with new audiences.
Source: AutoTweet research · 2026
Accounts with 'profile click rate' >5% see 30%+ follower conversion from each viral post
Profile optimization (clear bio + recent quality content) matters more than most people think. The post drives clicks; the profile drives follows.
Source: AutoTweet research · 2026
Common questions
Is the X algorithm public?+
Partially. X open-sourced the ranking algorithm in March 2023 (github.com/twitter/the-algorithm). The repo shows the engagement weights (replies = 27× a like, retweets = 1× a like, profile clicks = 12× a like) and basic ranking structure. It doesn't include the latest ML scoring layer or post-2023 updates, but the core dynamics — engagement velocity, weight tiers, account reputation — are confirmed.
How much weight does a reply have on X compared to a like?+
A reply is weighted as 27× a like in X's open-source ranking algorithm. This is the single most consequential number in X's algorithm — it explains why posts that provoke replies (questions, contrarian takes, polls) systematically outperform posts that only earn passive likes. The 27× number has been stable since the 2023 algorithm release.
Why do posts with links get less reach on X?+
X's algorithm down-weights posts with external links by roughly 30-40%. The reason: X wants to keep users on-platform; links drive them off-platform. The workaround that creators use: post the content WITHOUT a link, then drop the link in the first reply (which doesn't suffer the same penalty). This pattern is widespread enough that X's algorithm hasn't (yet) closed it.
How long do X posts stay relevant in the algorithm?+
About 2 hours of meaningful reach, with sharp decay after. ~80% of a typical post's lifetime impressions are earned in the first 2 hours. By hour 6, reach has dropped 80%+. By 24 hours, it's near zero for non-viral content. Viral posts can extend the curve to 24-48 hours but the median post is effectively dead after 4-6 hours. This is why posting cadence matters — each post is a fresh 2-hour campaign.