Scientists Can Now Predict Which Posts Will Go Viral on X

Researchers in Japan analyzed roughly 58,000 real conversations that spread across X (formerly Twitter), covering politics, entertainment, natural disasters and more. Instead of assuming posts spread in a neat tree — one person tells three, each tells three — they measured how lopsided the sharing actually was. Some posts get quoted, replied to, and re-shared in waves; others just get forwarded straight ahead. That lopsidedness is what they call "non-normality," a math way of saying "the flow leans one direction more than the other."

The main finding: a post's lopsidedness strongly predicts how sharply it peaks, not how large it eventually grows. Highly one-directional posts tend to explode fast, hit a concentrated burst of attention, then fade — think of a firework rather than a campfire that burns slowly for hours. Overall size was not strongly tied to this measure, which surprised the team. In other words, this tells you how fast something spikes, not how far it ultimately travels.

They also tested forecasting. If you've only seen the first 30% of a spreading post's network, predicting its shape is shaky — roughly 51% to 72% accurate within a 20% error margin. But once 50% to 60% of the network is visible, accuracy jumps above 80% across every type of spread they identified. The pattern becomes clear quickly once a post is already halfway through its run.

So what? This offers platforms, health agencies, and newsrooms a way to gauge how fast something is about to spike — useful for flagging fast-moving misinformation or timing an announcement. The honest limitation: this is a research paper, not a public tool, and it needs a lot of spread data before it becomes reliable. Early detection, the most useful moment, is still the weakest point.