# How the X algorithm works in 2026 You don't need the internal ranking code to grow. You need to know what the algorithm is trying to do, because everything else follows from that. After the Grok switch, X's ranking reads your posts more like a person would and less like a keyword-matcher. Here is what that actually changed, and what it means for the account you're trying to grow. ## The one thing it optimizes for X wants people to stay and to talk. Every signal it rewards is a proxy for one of those two things. A reply keeps a conversation alive, so replies are weighted heavily. A save means someone wants to come back, so saves count for more than a like. Time spent reading, a profile visit, a follow: all of it says "this was worth it, show more like it." Once you see the ranking through that lens, the tactics stop being a list of tricks and start being obvious. ## What actually moves a post Not all engagement is equal, and the gaps are enormous. A like is close to free and worth almost nothing. A reply is worth a lot. A reply the author writes back to is worth more than everything else on the post combined. If you optimize for one thing, optimize for sparking a real reply, not for collecting likes. - **Early velocity decides the ceiling.** The first 30 to 60 minutes are the single biggest factor. X tests a post on a small slice of people, and how they respond decides whether it gets a bigger slice. A slow first hour caps the post no matter how good it is. - **Replies > reposts > saves > likes.** Roughly. The higher-effort the action, the more it signals. A like costs nothing; a thoughtful reply costs attention, and the algorithm knows the difference. - **Author reply-backs are the multiplier.** When you reply to someone and they respond, the system reads it as a real conversation and pushes it. This is why replying to accounts that actually engage back beats shouting under accounts that never do. *[Diagram: reply timing, showing how an early reply on a rising post beats a late one on a peaked thread.]* ## Grok reads tone now This is the change most people haven't adjusted to. The ranking understands whether a post is constructive or combative, and it treats them differently even when the raw engagement looks the same. A sharp, additive take gets pushed. A dunk or a pile-on gets throttled, because anger keeps people on the thread but doesn't make the platform a place they want to return to. I learned this the expensive way. A takedown reply of mine did about 58,000 impressions and earned exactly zero follows. The reach was real; the algorithm let it travel because it was getting engagement, but the tone meant none of that reach converted, and the reputation cost was mine to keep. Constructive isn't a moral preference here. It's the setting that actually performs. ## Links still die in the body X makes its money keeping you on X, so a post with an external link in it gets suppressed. This hasn't changed and probably won't. The workaround that works is simple: put the link in your first reply, not the post. The post earns its reach, and anyone who wants the link scrolls one tap to find it. ## Reach is not the goal The hardest thing to internalize is that a big number can mean nothing. I've had a post reach tens of thousands of people and convert two follows. I've had a plain open question do about 19,500 impressions and pull 59 profile visits and a couple of real follows, which is a far better day even though the number looks smaller. Impressions measure how many people saw you. They say nothing about whether those were the right people, or whether you gave them a reason to stay. The number that actually predicts growth is downstream of reach: of the people who land on your profile, how many follow? Mine sits around 14.9%, which means the profile does its job and the real constraint is getting enough of the right people to it. That's a reach-quality problem, not a reach-quantity one, and chasing raw impressions makes it worse, not better. *[Diagram: the profile-visit-to-follow funnel, showing where reach leaks.]* ## What this means for tomorrow The algorithm hasn't made growth harder, it's made it more legible. Do the things it actually rewards: - Reply early, into rising conversations, before the thread crowds. Velocity is most of the game. - Aim replies at accounts in your lane that engage back, not at the giants who never will. - Stay constructive. The clever dunk costs more than it earns now. - Keep links out of the post body. First reply if you need one. - Judge every post by saves and follows, not impressions. [Grade a post](https://argushq.cc/tools/tweet-grader) on what compounds. ## Frequently asked questions ### What does the X algorithm reward most in 2026? Replies that pull the original author back into the conversation, and strong engagement in the first 30 to 60 minutes. A reply the author responds to outweighs a stack of likes. Raw impressions are not a ranking goal on their own. ### Do external links hurt your reach? A link in the body of a post still gets suppressed because X keeps people on-platform. Put the link in your first reply instead and the post keeps its reach. ### Why does a post with huge impressions get no followers? Impressions measure reach, not fit. If a post reaches people who don't care what you post next, none of them follow. Follows come from reaching the right audience with something worth keeping, which you measure with save rate and follow rate, not impression count. ## Where Argus fits Knowing how the ranking works is half of it. Acting on it every day is the other half, and that's what [Argus](https://argushq.cc/how-it-works) does. It reads your real funnel so you know whether reach or conversion is your actual constraint, scores live posts by reply opportunity so you get in early on the right conversations, and drafts each reply in your voice so staying constructive and consistent is survivable. You always press send. Next, read the [reply-first playbook](https://argushq.cc/guides/grow-on-x-with-replies) or [the rest of the growth guides](https://argushq.cc/guides). Grow on the algorithm you actually have.