X / Twitter · Algorithm-native writing

How X's Algorithm Decides What Gets Reach — And How We Use It

Most writing advice for X is guesswork. We built write better on something different: X's actual open-source algorithm. Here's what that means, why it matters, and how it changes every article the tool produces.

What is the X open-source algorithm?

X (formerly Twitter) released its recommendation algorithm as open-source code in 2023. The code — published on GitHub — reveals how the platform decides which posts get promoted in the For You feed and which ones die with zero reach.

At the core of the algorithm is a system called the Heavy Ranker. It's a neural network that predicts the probability of a user engaging with a post in specific ways — replying, liking, retweeting, bookmarking, clicking your profile. Each of these engagement types carries a different weight in the final score.

The signals that matter most: replies are weighted the highest (far above likes and retweets), early engagement in the first 30–60 minutes drives the largest score multiplier, and content that keeps users on the platform (no external links in the main post) gets a significant ranking boost.

In short: X has told us exactly what it rewards — most creators just aren't writing to it.

Why most X writing advice gets this wrong

The advice you find on X — and in most writing tools — is pattern-matched from popular posts. It tells you to “write good hooks” and “be controversial.” That's not wrong, but it's working backwards. It's observing outcomes and guessing at causes.

The open-source code gives you the actual causal mechanism. It tells you not just what performed — but why, and with what exact weight. A post that generates 50 replies outperforms a post with 500 likes, because replies carry a multiplier that most creators don't know about and no generic writing tool accounts for.

In short: the difference between writing for virality and writing for the algorithm is the difference between guessing and knowing.

How write better applies the algorithm to your article

write better is the only writing tool built directly on X's open-source algorithm. When you submit your topic, audience, and raw material, the model doesn't just generate text — it scores every sentence against the signal weights from the algorithm.

The output is structured to maximize reply generation (highest-weight signal), sustain dwell time through the full article, and avoid the patterns X's algorithm actively penalizes — like external links in the main post or generic content that triggers “not interested” clicks.

The training data for write better's prompting is sourced from X competition winners — articles that generated 4M to 48M views — cross-referenced against what the algorithm says should have performed. The tool writes in your voice. The structure is built for reach.

In short: you bring the idea; write better brings the algorithm.

What to do before you publish

Algorithm-native writing improves reach. What you do after generating matters too:

1. Review for voice. The tool mirrors your input — the more specific your voice notes and reference posts, the closer the output matches how you actually write.
2. Post natively — no link in the main post. If you're promoting a link, put it in the first reply. X's algorithm reduces reach on posts with external URLs by a significant margin.
3. Post at the right time. Early engagement in the first 30–60 minutes determines whether your article gets pushed to out-of-network accounts. Know when your audience is active.
4. Reply to every comment in the first hour. Replies chain into each other and multiply your engagement score. This is the highest-leverage thing you can do after posting.

In short: write better handles the structure; your timing and engagement in the first hour determines how far the algorithm carries it.