What's Actually Working on X Right Now

I've been paying attention to what actually makes me stop scrolling on X lately. This isn't an attempt to reverse-engineer the algorithm, nobody writing confidently about that online actually has access to it. It's closer to a running list: here's what I'm noticing, here's what X has actually confirmed, here's what I think is interesting about it.

The one thing X actually confirmed

In July 2026, TechCrunch reported that X changed its algorithm to surface more posts from "mutuals," people you follow who follow you back, specifically inside reply sections. Nikita Bier, X's head of product, said the missing signal had made replies "feel more like a battleground with people you don't recognize." That's a real, dated, on-record change. Everything past this point is my read, not something X has announced.

Writing that still sounds like a person

Around the same time, a dictionary named "AI slop" its word of the year, which tells you how mainstream the complaint about generic content has gotten. I wrote a full edition on why that happens and what actually causes it, worth reading on its own: Why So Much AI Writing Sounds the Same. Short version here: it's easier than ever to spot writing that sounds like an average of everyone instead of a specific person, and people seem less patient with it.

Conversations doing some of the work

Same idea applies to replies, and I already made the longer case for it: Your Replies Are Doing More Work Than Your Posts. Some of the most interesting account discovery on X still happens sideways, inside someone else's thread, not through a person's own timeline. The July 2026 change above doesn't prove that. It doesn't contradict it either.

What feels stale lately

This part is just me watching a feed too closely, no data behind it. Generic motivational one-liners with no real situation attached seem to get scrolled past faster than they used to. Engagement-bait questions with nothing actually at stake are easy to clock and easy to ignore now. Neither is new. The tolerance for it just feels lower, maybe because there's more of it than ever competing for the same four seconds.

None of this proves a strategy. It still points somewhere.

I'm not going to pretend one product change and a cultural mood around AI writing add up to a confirmed X strategy, because they don't. They're two separate things. What they both happen to point at, though, is worth taking seriously either way: being recognizable, sounding like an actual specific person, keeps showing up as the thing that works. Not necessarily because an algorithm rewards it, at least not one we can prove. Because that's what makes someone stop scrolling in the first place.

We built Aminta around that same idea. Not the point of this piece, just worth saying once.