The X algorithm, in 2026.

The one sentence summary

X For You ranks tweets on a composite of five signals weighted against follower count, content freshness, and account credibility, with bookmarks now weighted at roughly 2.5 times a like and Premium eligibility gated behind 500 followers plus 5 million impressions across three rolling months.

That single sentence covers about 80 percent of what most operators need to know. The rest of this guide is the detail behind each clause, because the implications change depending on which clause you are optimizing against.

The five signals For You actually uses

The 2026 For You model takes five primary engagement signals on every tweet and composites them into a single ranking score. The weights below come from our own regression analysis across 12 million tweets we have tracked since 2022, and they match what X engineering has described in public posts within roughly plus or minus 15 percent per axis.

• Bookmarks, weight 2.5. Re-weighted up dramatically in late 2025. The reasoning published by X engineering is that bookmarks indicate intent to return, which correlates with long term content value better than momentary reaction signals.
• Retweets, weight 2.0. Historically the strongest signal, now second. Retweets still carry secondary reach because they surface the tweet to the retweeter follower graph, which bookmarks do not.
• Replies, weight 1.6. Replies are a strong signal but come with a quality modifier. Replies from accounts with high authority on the topic weight higher than generic replies.
• Likes, weight 1.0. Still a signal, but the weakest of the five. Historically weighted higher, deliberately downweighted as bot generated likes became trivially cheap.
• Dwell time, weight 1.2. The time a user spent on a tweet before scrolling past or interacting. Heavily weighted for video and thread tweets, less for short text tweets.

Composite signals, not individual axes

The practical consequence of five weighted signals is that no single axis dominates. A tweet with 500 likes and zero bookmarks scores lower than a tweet with 200 likes and 100 bookmarks, even though the first tweet has 2.5x the aggregate engagement volume. The composite math is why we built the Engagement Suite as a bundled product rather than selling each axis separately.

Our regression analysis suggests the ideal viral signature for a new tweet has a ratio of roughly 10 bookmarks to 1 retweet to 5 likes to 2 replies across the first 60 minutes. Tweets that land inside 30 percent of that ratio graduate into For You distribution at roughly 4x the rate of tweets with lopsided signal profiles.

This is the core reason we do not recommend buying individual signals in isolation. A thousand likes on a tweet with no bookmarks screams purchased to the ranking model. A thousand likes combined with 100 bookmarks and 50 retweets looks like a viral tweet signature and is scored accordingly.

Follower count as a multiplier

Absolute signal volume is multiplied by a follower count modifier before the composite is compared against other tweets in the user For You queue. The modifier compresses extreme volumes on high follower accounts and amplifies modest volumes on lower follower accounts, so a 50 bookmark tweet on a 1,000 follower account scores comparably to a 500 bookmark tweet on a 100,000 follower account.

This normalization is why follower count still matters for growth but matters less than it used to. The old For You model before 2024 weighted absolute follower count as a primary ranking signal. The current model weights it as a normalization input for engagement velocity, which is a fundamentally different mechanic.

Content freshness and temporal decay

Every tweet carries a freshness multiplier that decays from 1.0 at the moment of posting to 0.3 at 24 hours, 0.15 at 48 hours, and 0.05 at 72 hours. After 72 hours the tweet is effectively out of For You candidate pool regardless of ongoing engagement, though bookmarks continue to accumulate on the tweet in perpetuity.

The practical consequence is that the first 60 minutes of a tweet is where almost all For You leverage lives. A tweet that crosses the viral signature threshold in the first hour gets distributed through For You for the following 24 to 48 hours and typically collects 80 percent of its lifetime impressions in that window. A tweet that fails to cross the threshold in the first hour rarely recovers even if it eventually accumulates high engagement totals.

This is also why drip pacing matters. Engagement arriving across 90 minutes to 3 hours after a tweet is posted matches the shape of organic viral lift and scores well in the ranking model. Engagement arriving in a 2 minute batch looks like a purchased push and the ranking model actively suppresses tweets with that signal pattern.

Premium monetization, the revenue share math

X Premium revenue share pays creators from the ad pool based on impressions earned from other Premium subscribers on their tweets. The eligibility thresholds as of April 2026 are 500 followers plus 5 million impressions across the previous three rolling months. Both thresholds must be met simultaneously. Miss either and the payout for that period is zero.

Follower count is measured at the end of the rolling period. Impressions are counted across all tweets posted during the three month window, summed. Impression quality matters: only impressions from Premium subscribers count toward the 5M threshold, not impressions from free tier users. In practice this ratio runs around 60 to 70 percent of total impressions on most accounts, which means operators need to aim for 7 to 8 million total impressions to clear 5 million Premium eligible impressions.

Typical payouts for newly eligible accounts range from $40 month one on niche technical accounts to $2,000 month one on general culture or crypto accounts. Payouts compound as impression velocity sustains, often reaching $500 to $4,000 monthly by month three for accounts that maintain weekly posting consistency. Our Impressions product is designed specifically to clear the 5M threshold on a 90 day drip, paired with follower shipments for the 500 gate.

Account credibility signals

Beyond follower count and engagement history, the ranking model also scores accounts on a credibility vector that influences how aggressively a tweet enters For You distribution. The credibility score is derived from roughly eight inputs visible to the ranking model:

• Account age in days
• Posting frequency variance (consistent posting scores higher than sporadic)
• Profile completeness (picture, bio, banner, pinned tweet all filled)
• Follow graph quality (following accounts with their own follow graphs, not dormant accounts)
• Engagement rate variance (sudden spikes trigger a suspicion flag)
• Platform penalty history (past warnings, appeals, restrictions)
• Premium subscription status (subscribed accounts get a modest credibility bonus)
• Reply graph authority on topic clusters (replies received from authoritative accounts boost score)

The credibility score compounds slowly. Accounts that post consistently for 90 plus days and maintain clean variance patterns accumulate credibility that meaningfully lifts For You eligibility even on modest engagement. Accounts that game the system through sudden engagement spikes without corresponding credibility growth get suppressed regardless of raw signal volume.

What the algorithm does not care about

A few common operator beliefs that the ranking model in 2026 does not actually care about, based on our regression analysis:

• External links. Historically penalized for taking users off platform. In 2026, neutral weighting. Tweets with links score equivalently to tweets without, adjusted for link click through which is a separate signal.
• Hashtag count. Zero weight either way. Hashtags are indexed for search but do not influence For You ranking.
• Posting time of day. Zero direct weight. Posting time matters only through the indirect effect of audience availability, which is captured by engagement velocity in the first 60 minutes.
• Capitalization, punctuation, line breaks. No direct weight. Content quality is measured entirely through engagement response, not style signals.
• Emoji usage. Zero weight either way. Emoji heavy tweets and plain text tweets score identically per engagement signal.

Practical implications for operators

The composite signal model in 2026 has three practical implications for anyone trying to grow or monetize on X.

First, pair signals. Buying a thousand likes on a tweet with nothing else underneath produces worse algorithmic lift than buying 100 bookmarks, 50 retweets, and 250 likes together. The ratio beats the volume.

Second, drip. Any engagement pattern that arrives faster than organic viral lift physically could is downweighted by the ranking model. Deliver signals across 30 minutes to 3 hours matching the shape of real viral pickup.

Third, if Premium monetization is the goal, follower count plus impressions matters more than raw viral outcomes on individual tweets. A steady account hitting 200,000 impressions per week across twenty tweets outperforms a sporadic account with one million impression viral tweets but gaps between.

Everything Twitterz ships is built against this model. Engagement Suite ships the composite signal. Impressions and Bookmarks ships the monetization fuel. Followers Engine ships the account credibility floor. The products map 1:1 to the algorithm.

How we update this guide

When X engineering publishes a change to the For You ranking model, we run the regression analysis on our campaign data to confirm or update our understanding, then revise this guide within 14 days of the change taking effect. Revisions are versioned and the change log lives on this page is footer.

If a change to the model materially affects how our products perform, we proactively email affected customers with guidance on how to adjust. Major changes since 2022 include the bookmark re-weighting of November 2025, the Premium threshold adjustment of August 2024, and the credibility signal re-tuning of March 2023. All three were covered in customer outreach within two weeks of deployment.

Further reading

For the operator playbook shaped around this algorithm, see How to grow on X in 2026. For the Premium revenue share math specifically, see the creator solutions page. For the mechanics of how our products pair against the signal model, see the individual product pages at /products.