AutoTweet Logo
AutoTweet
Direct answer · Updated 2026-05-17

How does the X (Twitter) algorithm work?

Short answer

The X algorithm in 2026 ranks posts on engagement velocity (first 30 minutes), engagement weight (replies are 27× a like, retweets are 1× a like, dwell time matters most), account reputation (recent post quality, follower engagement, policy history), and content type (long-form > short, native > linked, threads > singletons). The single biggest signal: replies in the first hour.

The four pillars of X's ranking

1. Engagement velocity — likes/replies/RT per minute in the first 30 minutes. Posts that earn early engagement get amplified to far more people than posts that earn the same engagement slowly. 2. Engagement weight — not all engagement is equal. Per X's open-source code: a reply is worth 27× a like. A retweet is worth ~1× a like. Dwell time (how long users read the post) outweighs all of them. A post with high dwell time but low likes can outrank a post with the reverse. 3. Account reputation — recent post-quality history. Accounts whose last 30 posts had high engagement-per-post get a tailwind on new posts. Accounts whose last 30 posts flopped get a headwind. 4. Content type — native content (text-only, no external links) outranks linked content. Long posts (250+ chars) outrank short ones, all else equal. Threads outrank single posts because they generate more dwell time per visit.

Replies = 27× a like (the most consequential rule)

This number comes from X's open-source ranker repo. A single reply on your post is weighted in the algorithm's scoring like 27 likes would be. Practical implication: posts that provoke replies (questions, contrarian takes, polls) systematically outperform posts that provoke only likes (jokes, observations). The fastest growth tactic in 2026: write replies on other people's posts that are themselves provocative enough to earn replies — those reply-threads compound your own reach via X's reply prioritization.

Dwell time — the invisible signal

X measures how long users spend reading each post. A 1-tweet 'great take' earns dwell time of 2-3 seconds. A 9-tweet thread earns dwell time of 90+ seconds across the same user. Threads outrank singletons not because of the format itself but because of the cumulative dwell time. If you're optimizing for reach, write posts that demand reading time — long posts, threads, posts with multiple distinct ideas that earn re-reads.

The 'recency cliff' at 2 hours

X's algorithm aggressively discounts posts older than 2 hours. By 6 hours, a post's algorithmic reach has dropped 80%+. By 24 hours, near zero. The implication: every post is essentially a 2-hour campaign. Front-load engagement in the first 2 hours. If a post is going to work, it works in that window; if not, repost a variant rather than waiting for slow burn.

People also ask

Is the X algorithm public?+

Partially. X open-sourced the ranking algorithm in March 2023 (github.com/twitter/the-algorithm). The repo shows the weights (replies = 27× like, retweets = 1× like, etc.) but not the full ML scoring layer or the latest 2024-2026 updates. The core dynamics — engagement velocity, weight tiers, account reputation — are confirmed from the public code.

Does the algorithm hate links?+

Not 'hate' but systematically down-weights posts with external links to non-X domains. Posts with links get ~30-40% less reach than identical link-free posts. The workaround creators use: post the content WITHOUT a link, then drop the link in the first reply (which doesn't suffer the same penalty).

How fast does the algorithm change?+

Major changes every 3-6 months, smaller tuning constantly. The big-picture mechanics (engagement velocity, reply weight, dwell time) have been stable since 2023. The fine-tuning shifts often enough that very narrow optimization tactics decay within 6-12 months. The durable plays: hook craft, reply frequency, posting cadence — these stay relevant across algorithm versions.

Why does my older content sometimes resurface?+

X has a 'For You' feed that surfaces older posts when a user's recent activity suggests the topic. A post from 6 months ago can suddenly get 1,000 new impressions because the algorithm spotted a relevant signal in your audience's behavior. This is rare but happens — it's why evergreen content (definitions, frameworks, examples) outperforms timely content over a 6-month horizon.

Related questions

How do you go viral on X (Twitter)?How often should you tweet?Why am I shadowbanned on X?

Skip the manual work

AutoTweet's AI Autopilot generates a week of voice-matched tweets, schedules them at peak times, and tracks per-tweet performance. 14 tweets queued the moment you connect X.

Cancel anytime