X just updated the algorithm.
It now scores your posts on 19+ signals before deciding whether or not people see your content.
Here's what changed, and what it means for your account:
Quick context on what's different:
The old algorithm had fixed rules.
Likes, retweets, clicks, etc.
Each one had a weight. You could reverse-engineer it.
That's gone.
X replaced the whole system with an AI model called Phoenix that reads your engagement history and predicts whether a specific user will care about your post.
No more shortcuts. Gaming it with a formula is over.
But before your post even reaches that scoring model, it has to clear a new gate first.
There's a content classifier called Grox that rates every post on two things:
A quality score. A slop score, built specifically to detect low-effort, repetitive, template-driven content.
If you don't clear the quality threshold, the engagement model never even sees your post.
This is the biggest change in the update.
If you're running the same hook structure every week, you're already losing here.
Here's every action the algorithm now scores:
The ones you already know:
Likes
Replies
Reposts
Quotes
Still matter. But they're not the whole picture.
The ones most people sleep on:
• Thread expansion clicks: someone clicking to read your full post.
• Profile clicks: the algorithm reads this as "this person was worth investigating more."
• Photo expands: Interesting media, graphics, etc.
• Quote clicks: someone clicking through on a post you quoted or viewing the quotes to your post.
The sharing signals:
DM shares and copy-link shares are tracked as two separate signals.
Someone liked your post enough to send it to another person.
That hits harder with the new algorithm than a regular like does. Both are underused levers.
The video signals:
There's a minimum video duration gate. Short clips below the threshold get zero weight regardless of view count.
If you're cutting clips to 15 seconds because "that's what gets views," this is worth a conversation.
The consumption signal most will miss:
Dwell time: how long someone stops on your post before scrolling.
The algorithm rewards this. Posts that make people stop and read perform better than posts that get a quick like and a scroll.
Not-dwelled is an explicit penalty in the code. Fast scrolling past your post actively lowers your score.
This is why generic, surface-level content kills accounts over time. It stacks up negative signals quietly, month after month.
The strongest positive signal on the list:
P(follow_author).
The probability that someone follows you after seeing your post.
Content that converts viewers into followers gets pushed harder. This is literally the metric that rewards authority-building content over viral bait.
If your content is getting views but not followers, the algorithm is telling you something.
The 4 signals that bury you:
"Not interested" is the lightest hit. Mutes are worse. Blocks carry their own, heavier penalty weight. Reports hit hardest of all.
Each has its own distinct weight parameter in the code. They are not the same bucket, and treating them like they are will cost you.
Polarizing content that generates blocks is not the move, regardless of the views it gets.
The part that surprised me most:
There's a built-in Author Diversity Scorer.
Its job is to reduce your score if you've already appeared in someone's current feed load.
This runs within a single feed refresh, not across days or weeks. It's not a daily post cap.
But the first post from you in a feed gets full score. The second gets a decay multiplier applied. The third decays further.
Quality over quantity is now written into the algorithm.
Based on our data across clients, 3 posts per day is the max effective frequency. We will continue to test this, so follow me @itsmarcosruiz if you're finding this article valuable and want future updates.
One more layer most breakdowns are missing:
Posts from accounts someone follows get their full score in the ranking system.
Posts from accounts they don't follow get multiplied down before final ranking, with two exceptions: topics they explicitly follow get a boost, and new accounts get a special out-of-network lift to help them get discovered faster.
What this means practically: a follower still gives you a structural scoring advantage that has nothing to do with engagement. Build your following like it's an asset, because it is.
And a big one no one talks about --> Optimize for custom timelines. We're past the age of only a 'for you' page.
So the algorithm pulls from two pools:
- Posts from accounts you follow
- Posts from the entire platform
One thing people will miss about pool #2:
Your post has to pass a retrieval gate before it ever reaches the scoring model.
If your content doesn't match a user's interest profile closely enough at retrieval, the engagement scoring model never even sees it.
There are two separate things to optimize for: retrieval relevance and engagement prediction.
This is why a 200-follower account can go viral. And why raw engagement metrics don't explain everything.
A few takeaways for your account:
- Write posts worth expanding and reading all the way through.
- Make content people share in DMs or copy the link on.
- End threads with a clear point of view.
- Follow probability is one of the top signals in the system. Keep your profile sharp. Profile clicks score, and your bio is part of your content's result.
- Stop posting generic content, copy-paste content.
- A block hits harder than a mute. A mute hits harder than a not-interested. They are not the same thing.
- Post 3x/day maximum.
If you learned something from this article, share with a friend (so I can rank high on the sharing score 🐦)
Cheers
Marcos
