How the X Algorithm Works in 2026: Weights, Grok, Payouts

Understanding how the X algorithm works in 2026 takes one afternoon, because X told us. The company open-sourced its ranking code in 2023, the engagement weights were sitting right there, and the changes since have all pushed the same direction. This is the short technical tour, and what each piece means for a small account.

The weights: conversation beats applause

The open-sourced heavy ranker scored engagement types explicitly:

Read that last row twice. A reply that pulls the original poster back into the thread is worth roughly 150 likes of ranking signal. This single number explains most credible growth advice: why the reply guy method works, why "engagement bait" questions outperform statements, and why answering your own commenters matters as much as posting.

The candidate pipeline, in one paragraph

When someone opens the For You feed, X pulls candidate posts from accounts they follow (in-network) and accounts they don't (out-of-network), scores each candidate with the ranking model for predicted engagement, applies filters and diversity rules, and serves the top of the list. The out-of-network half is why small accounts can grow at all: your post or reply competes on predicted engagement, not on follower count. It's also why early signal matters so much; a post that earns replies in its first minutes gets promoted into more feeds, which earns more replies. Momentum compounds in both directions.

What Grok changed

X's recommendation stack now uses Grok-based content understanding to score quality: is this post original, informative, entertaining, or is it engagement-bait sameness? Two practical consequences. First, template content (the same hook shapes every AI writer produces) gets classified as the noise it is. Second, rage-bait stopped being a strategy: manufactured outrage still triggers replies, but the quality layer discounts it. The full argument about what replaced it is in how to grow on X in 2026.

The April 2026 payout change

X started cutting creator revenue-share payouts for clickbait and recycled content: aggregator accounts were dropped to 60% of their share in one cycle with another 20% cut queued, after the head of product said stolen reposts and bait were crowding out real creators. This matters even if you don't monetize yet, because payout policy telegraphs ranking policy: the behaviors X stops paying for are the behaviors it's learning to detect. Evergreen-repost automation and viral-template libraries aged badly for exactly this reason; the trade-offs are covered in best X growth tools in 2026.

So what do you actually do with this?

Five moves fall straight out of the mechanics:

1. Spend most of your effort on replies, where the 27x and 150x multipliers live. Ten sharp replies beat three mediocre posts every day of the week.
2. Respond to everyone who replies to you. You're handing each of them the author-engagement multiplier, and teaching the model your posts produce conversation.
3. Be early. Both on others' posts (first minutes ride the whole curve) and on niche news, where the first good take wins the distribution.
4. Say something only you can say. Original specifics survive the Grok layer; recycled shapes don't.
5. Keep links out of the main post when you can; deliver value natively.

None of this requires gaming anything. The 2026 algorithm is unusually honest: it pays for being a person worth talking to, quickly and consistently. The tools that help with exactly that, and the ones optimized for the old game, are ranked in the tools comparison.