X just quietly fixed something that had been quietly driving people nuts for a long time.
On Monday, Nikita Bier, head of product at X, announced a small but pointed algorithm tweak: posts will now get a visibility boost among your mutuals, the people you follow who also follow you back.
The signal was simply missing from the ranking system before.
Without it, friends and familiar accounts ranked lower in reply threads and feeds, so those sections filled up with strangers.
The result felt less like a conversation and more like a public square fight with people you had never seen before.
Bier said the change should also make it easier for interest-based clusters to form naturally, something users had been asking for.
To understand why this matters, X's algorithm, which resides in the For You feed, isn't a simple chronological list of people you follow. It runs a three-stage process that starts with candidate sourcing.
The system pulls roughly 1,500 possible posts from two main pools: in-network content from accounts you follow (via a system sometimes called Thunder) and out-of-network recommendations based on similarity to what you have engaged with recently (via Phoenix, a Grok-powered transformer).
Those candidates then go through heavy ranking.
Then, a neural model predicts the probability of many different actions you might take: reply, quote, repost, like, bookmark, click a profile, watch a video past certain thresholds, or even negative actions like mute or report.
Positive predictions get weighted and summed into a final score; negatives subtract heavily. Replies are king.
A direct reply can be weighted dozens of times higher than a like.
Early engagement velocity in the first 30 to 60 minutes matters a lot. Dwell time, author reputation, and content quality signals also play in. Finally, filters and mixers remove spam, duplicates, low-quality posts, and enforce some author diversity so one account does not flood your feed.
The whole thing is designed to maximize predicted positive engagement for you specifically. The open-source code and public explanations make clear that the system learns from your recent behavior rather than relying on rigid hand-tuned rules or pure follower counts.
What was missing until this week was a clean mutual-follow signal inside that ranking.
The algorithm knew who you followed and who followed you in isolation, but the reciprocal relationship itself was not properly scored. So a reply from a mutual friend carried no extra weight over a reply from a random high-engagement account or a bot.
In reply threads especially, the ranking treated everyone more or less the same once basic engagement signals kicked in.
That is why your own friends often disappeared under a pile of strangers arguing at full volume.
The tweak simply injects that missing mutual data as a positive ranking feature.
Posts and replies from people you follow back now get a gentle lift when the system is deciding what to show you (and what to show them of you). It is not a hard override or a chronological switch. It is one more useful feature in the big scoring model.
Early metrics he shared the next day showed the first results: replies up 3.15%, original posts up 1.8%, small account reach up 1.19%, overall time spent down a negligible 0.13%, and what he called "unregretted time spent" effectively infinite.
In other words, people are talking a bit more with the people they actually know, and they do not seem to mind the extra time.
This change matters for a few practical reasons.
First, it restores some of the social graph that people actually experience offline. Social media works better when conversations feel local and familiar rather than like walking into a random stadium.
Second, it rewards genuine relationship-building over pure broadcast or follow-for-follow spam. Creators who have spent time cultivating real mutuals now get a small distribution edge inside those circles.
Third, it should reduce the “battleground” feel of replies without needing heavier-handed moderation. When the people you know show up first, the tone of the whole thread tends to stay higher.
Fourth, it helps interest clusters form organically. People who already follow each other around a niche topic will see more of one another’s posts, which can tighten those communities without X having to invent new group features.
Early data already hints at more replies and better small-account reach, which is good news for anyone who is not a mega-influencer.
User reactions so far have been overwhelmingly positive, with a few predictable notes of caution. Across thousands of replies and quote posts, the dominant vibe is relief.
People are posting screenshots of long-lost mutuals reappearing in their feeds and reply sections. Creators are saying their actual communities are finally seeing their work again instead of only random engagement farmers. Many are treating it as confirmation that building real relationships beats chasing vanity metrics.
In short, X noticed a basic social signal had fallen out of its ranking model, put it back in, and the platform is already a little less alien.
The algorithm still rewards content that sparks real engagement, especially replies, but it now remembers that some of those replies come from people you actually chose to follow back.
For anyone who has felt like their friends vanished into the noise, this is a quiet win. Keep posting for the people who already care, engage with your mutuals, and the system is a bit more likely to help you keep finding each other.
