X Recommendation Algorithm Overhaul! It's not your imagination that no one is viewing your posts: how does the new system "guess your mood" to determine exposure?

Starting from 2026, many X (formerly Twitter) users began to notice: Why are all the accounts I usually follow disappearing? Why can’t I see my friends’ posts? My posts clearly have likes, but they all seem to be “drowned out”? The main reason behind this is that X officially launched a brand-new recommendation algorithm at the beginning of 2026: the Phoenix system. This update is not just a “minor revision,” but a complete overhaul of the platform’s operational logic. From who sees your posts, to why no one sees them, and even which people you see, everything is now determined by this AI algorithm.

What is Phoenix? X no longer relies on popularity to determine post exposure

In the past, social media post visibility depended on “popularity metrics”: who gets more likes, who gets retweeted more, and what topics are trending.

But the Phoenix system completely abandons this approach.

This new system is more like a large AI model similar to GPT. It predicts “what you want to see next” based on your recent micro-behaviors such as scrolling, clicking, blocking, and dwell time, and then decides what content to push.

In simple terms: whether a post becomes popular isn’t about how hot it is itself, but whether it can attract the right people to stop and look.

The world each person sees becomes a “private channel”

This kind of algorithm makes each user’s feed more personalized.

Even if you and your friends follow the same accounts, the content you see can be vastly different because:

The model decides what to push based on your own behavior

“Common trending” content becomes less and less

The overlap between similar interest groups diminishes

This means you and your friends might be scrolling through X in completely different informational bubbles.

The key to a post is no longer “how many likes,” but “who liked it”

That’s also why many creators now say: “Even though I got likes, why am I still not popular?”

The answer is: Phoenix doesn’t care about numbers at all; it cares about the characteristics of the interactors.

If your posts attract a group with common interests who regularly watch your similar topics, the algorithm will consider “your content has a clear audience” and will recommend it to more similar users.

But if you randomly post a cat photo today, talk about politics tomorrow, and AI the day after, the algorithm will get confused about who you are and whom to assign you to, making exposure unstable.

Each post is “individually scored,” no longer competing with each other

Phoenix also introduces a new mechanism called Candidate Isolation.

Traditional algorithms compare all posts together to decide who gets priority exposure. But Phoenix does not.

It scores each post in an “independent room”:

Your posts won’t be crowded out by celebrities or major events

Small accounts also have a chance to go viral, as long as you target the right audience precisely

Each post is a fresh evaluation

For users, this is both fairer and more difficult to master.

The first 10 seconds after posting are critical, deciding life or death

Phoenix is extremely fast. After you post, it observes within 10 to 30 seconds:

Who stops to look

Who clicks in

Whether the engaged users match your “typical audience”

Whether these users have similar behavioral profiles

If the model thinks the post didn’t generate the right interest, it will immediately reduce your exposure.

So many people feel “my posts sink as soon as I publish”—not because your content is bad, but because you haven’t attracted the “right people” as defined by the algorithm.

What behaviors cause the algorithm to lower your post’s exposure?

Phoenix’s algorithm is somewhat “picky.” The following behaviors make it hard for the algorithm to determine your account’s attributes, leading to unstable exposure:

Topic fluctuates wildly, content style jumps too much

Posting too many external links (e.g., directing traffic to websites, YouTube)

Frequent cross-domain posting, like writing about AI today and food tomorrow

Attracting an audience completely different from your usual followers

In simple terms: “The harder you are to categorize, the harder it is to get stable exposure.”

How to let the algorithm “understand who you are”? Two essential strategies for creators

Based on open-source documents and observations, two strategies can significantly improve the algorithm’s understanding of you and ensure stable recommendations:

1. Maintain consistent content themes and unified style

Allow AI to categorize you into a specific interest group.

For example: Main focus on cryptocurrency, secondary focus on sharing life anecdotes

Avoid jumping between topics like politics today, travel tomorrow, pets the day after

2. Cultivate a “fixed interaction circle”

You need to attract a small, stable group of engaged followers.

The model will determine your category based on this group

And automatically recommend you to other similar potential audiences

These two things are much more important than “liking” or “following trending topics.”

Does writing style and format also affect exposure?

Phoenix observes your language style, structure, and emotional tension in your posts.

Recommendations:

Start with a “hook” (question, surprise, conclusion, contrast)

Use a consistent structure: for example, “Opinion → Example → Breakdown”

Maintain stable tone and format, avoid fluctuating styles

Don’t make every post wildly different in style (AI gets confused)

It’s not that you’re not popular; it’s that you haven’t been understood yet

X’s recent algorithm update fundamentally changes the logic of post exposure.

It’s no longer a “popularity contest,” but a test of “can you attract the right people.”

Every post you make is telling X:

Who I am

Who I want to attract

Who likes me

If you write in the right direction, the algorithm will guide you toward stable growth; if you go the wrong way, your content may never surface.

This is a new information game under AI orchestration. The algorithm is no longer an enemy but a partner—first, let it “understand you.”

This article: X’s recommendation algorithm overhaul! If your posts aren’t getting views, it’s not a coincidence: how does the new system “guess your mood” to decide exposure? Originally published on ABMedia.