X has discovered a bold new strategy for making social media feel social again. It’s going to show your posts more often to people you actually know.
According to X product head Nikita Bier, the platform is boosting the visibility of posts among mutuals, meaning accounts that follow each other. He said this relationship data had been missing from the algorithm, leaving familiar accounts less visible when reply sections filled up.
What is X changing
The adjustment should make someone’s posts more visible to mutual followers. When those people join the conversation, their responses should also become easier for the original poster to find among replies from unfamiliar accounts.
Bier believes this could help groups form around shared interests. That sounds like basic social networking, but modern feeds care heavily about predicted behavior rather than the accounts people deliberately follow. An angry response from a stranger still registers as activity, even when it makes the service feel like a battleground.
Giving existing connections more weight could redirect some of that attention. Whether the difference will be noticeable depends on how much influence the signal actually receives.
Why did the AI need help finding friends
The strange bit is that X’s published algorithm code already contains tools for understanding mutual relationships. Its May 2026 release added mutual-follow graphs to user context and corresponding scores to the data attached to posts being considered for recommendation.
The announcement suggests that this information wasn’t available to the relevant ranking process or wasn’t influencing results as intended. X hasn’t explained where the disconnect occurred, so we can’t tell whether this was an omission, a weighting problem, or something else.
X’s Grok-based model predicts whether a post might earn replies or clicks. Those predictions can identify content that keeps people active, but activity doesn’t reveal whether a conversation is welcome. An algorithm can learn that arguments generate reactions without understanding why everyone involved is miserable.
Will X actually feel friendlier
X hasn’t disclosed the strength of the boost, whether it extends beyond these interactions, or how success will be measured. The company has described the desired result without demonstrating it.
Recognizable names could make conversations feel less random, but a ranking adjustment won’t remove outrage bait or hostile users. Teaching an algorithm who your friends are is relatively easy. Teaching it that every argument isn’t valuable engagement will be harder.
