Why posts go viral on X, line by line

A creator in our network sent me a post draft last spring and asked if it was ready. It was good. Sharp opening, real number in it, a screenshot that carried the argument. He posted it at 4:52pm his time, then closed his laptop and went to dinner, because he'd done the work and the work was done.

It got 1,100 impressions. Two weeks later he posted a worse version of the same idea, stayed on his phone for an hour, replied to the first eleven people, and it went to about 380,000.

I've watched some version of that happen enough times that I stopped calling it luck. Most of what people label "the algorithm" is really the gap between those two afternoons.

Why do posts go viral on X?

Posts go viral on X when a small early group engages hard enough that the ranking system shows the post to people who don't follow the author. Replies and profile clicks carry more weight than likes, so the first line and the author's own first-hour behavior matter more than posting time or hashtags.

That's the compressed version. I want to draw one line clearly before I say anything else, because the going-viral genre is full of people stating mechanics as fact that nobody outside X can verify.

Here's what's actually on the record. In March 2023, X open-sourced a chunk of its recommendation stack, including the home timeline ranker, on GitHub, alongside an engineering post explaining the shape of it. That release described a candidate-sourcing step pulling roughly 1,500 posts per request, split between in-network and out-of-network sources, then a heavy ranker scoring them with a set of weighted engagement predictions. The relative weights were sitting right there in the config.

Two things about that snapshot. It's from 2023, and X has shipped a lot since. And it only covered part of the pipeline. So treat those weights as evidence about how the company thought about the problem, not as a live spec.

X also documents, on its own help pages, that Premium subscribers get prioritized ranking in replies and search. That one is stated policy, not inference.

Everything else in this post that isn't linked is inference. I'm drawing it from campaign data across our creator network and from watching a few hundred posts go out with the author sitting next to me, which is a decent sample and still not a source. I'll flag it as I go.

The two lines that decide everything

Line one is the only line most people read

The single most useful thing I know about X is that the first line is not the beginning of your post. It's the whole ad for your post.

On a phone, a post with any real length gets truncated behind "Show more." What renders above that fold is usually two to three short lines, and the first one is doing almost all of the work, because a thumb moving down a feed makes its decision somewhere around the sixth word.

The failure mode I see most from founders is the windup. "I've been thinking a lot about onboarding lately, and I wanted to share something we learned." That's a first line that promises a first line. By the time the actual claim arrives on line three, the reader is two posts down.

Cut until the first thing you say is the most surprising thing you have.

Line two is where the post dies

Line one gets the tap. Line two decides whether the reader keeps going, and it's the line almost nobody edits.

The mistake is nearly always the same: line one makes a bold claim, and line two starts explaining it. Explanation is a downhill move. The reader's brain registers "ah, now the reasoning," and the tension that made them tap collapses.

Line two should escalate or complicate. Compare these two, both invented:

We turned off our onboarding email sequence. We'd spent six weeks writing it, and turning it off was the highest-impact thing we did that quarter.

versus

We turned off our onboarding email sequence. Activation went up 9%. I think the emails were doing the job the product should have done.

The first one is about the author's feelings. The second raises the stakes and hands you a claim you might want to fight about. Fighting is the point, which brings me to the next thing.

Replies count for more than likes, and the code said so

This is the one piece of the puzzle that isn't guesswork. In the 2023 open-sourced ranker configuration, the predicted-engagement weights were wildly asymmetric. A like was a small positive term. A reply was worth many multiples of it. A reply that the original author then replied to was worth the most of any positive signal in the list.

I'm not telling you to bait replies. Reply-bait posts get replies from people who will never buy anything, and if you're paying a creator for a launch, a thread full of "this" is a waste of the money. What I am telling you is that a post engineered to be liked and a post engineered to be argued with are structurally different objects, and only one of them travels.

The practical version: end on something contestable, and leave a gap the reader can fill. A post that answers every objection gets admired. A post that leaves one obvious objection unaddressed gets 40 replies raising it, and the author replying to each one is, per that config, the strongest positive signal available.

Quote tweets are the actual distribution engine

Now I'm off the record and into inference, so weigh it accordingly. I couldn't find a distinct quote-post term in the published weights, which means this next claim is mine, from campaign data, not from X.

A repost sends your post to someone else's followers with your framing intact. A quote post sends it with a new first line, written by someone whose audience already trusts them. Since line one is the whole ad, a quote post is a fresh ad for the same product, aimed at a room you couldn't reach.

Practically, this changes what we ask creators to do. A brief that says "please repost" gets you a repost. A brief that says "if you disagree with any part of this, quote it and say so" gets you a quote post with a real opinion attached, which outperforms the polite repost in nearly every campaign I've measured. Some creators refuse, which is fine, and honestly a creator who won't fake an opinion is one I trust more. That's roughly the standard we apply to the people in our network.

What a link costs, and how people get around it

Everyone on X believes posts with external links get less reach. I believe it too. I want to be careful here: I cannot point you at a document from X that says so, and I've seen X staff push back on the strongest versions of the claim. What I have is our own before-and-after, which is not proof, because we change several things at once and never run a clean holdout.

Here's the honest version of the workarounds and what each costs you.

For a launch, I default to the first reply, posted within about five seconds of the parent so it sits at the top of the thread before anyone else replies. And I set expectations with founders up front: if a post does 200,000 impressions, the reply with the link does not do 200,000 impressions. Plan for a fraction of it.

On media, my read is simpler than most advice you'll see. A screenshot of something real, cropped tall so it takes up more feed height, beats a designed graphic almost every time. Native video beats a link to video. And a post with no media at all is completely fine if the first line is strong, which is a thing plain-text posters have been proving for years.

The first 90 minutes

Here's my actual argument, the one I'd defend: for most posts, the author's behavior in the first 90 minutes changes the outcome more than the copy does.

The reasoning is straightforward even without knowing the mechanics. The system needs early evidence before it will show a post to people outside the author's followers. That evidence arrives in a narrow window. Replies weigh heavily, and author-replied replies weigh most. So the author, sitting there replying, is manufacturing the exact signal the ranker rewards, at the exact moment it's being measured.

So the brief we send creators now has a scheduling requirement in it, and it's the line founders push back on most: don't post unless you can be on your phone for the next 90 minutes. We've moved launch posts by a full day over this. A creator who posts at a mediocre hour and stays present beats a creator who posts at the perfect hour and disappears, and I'd take that bet with money.

The other thing that happens in that window is deletion. If a post is clearly not moving after 20 minutes, delete it and repost a better first line tomorrow. Founders hate this. It feels like admitting failure. It's the cheapest experiment available to you and the sunk cost is four minutes of typing.

The part I'd argue about

Most "why posts go viral on X" advice optimizes the wrong object. It treats the post as the unit. The post isn't the unit. The post plus the author's next 90 minutes is the unit, and if you only control one of them, control the second one.

I'll also admit the limit of everything above. This is pattern-matching against an undocumented system that changes without notice, by someone who reads a lot of campaign dashboards. If X ships a ranking change next month, half of this gets stale and I'd rather tell you that now than have you find out.

If you'd rather not learn this by burning your own launch on it, that's what we do. Browse the creator network, or send us the brief and we'll tell you which three people it's actually for. Creators who want in can apply here.