Data & Research
Rush Hour: Predicting TikTok Virality Before It Happens
A 950-video study showing you can identify a viral TikTok within two hours of launch — before the algorithm finishes deciding.
Problem
Follower count used to be the whole story for reach. On algorithmically-curated feeds like TikTok's For You page, it isn't — and nobody had shown how early you could tell a post was going to break out, or what replaces "follower count" as the real signal.
Approach
Tracked 950 newly-posted TikTok videos from 86 fashion and apparel brand accounts at 10-minute intervals over 7 days. Clustered posts into reach archetypes by the shape of their view-accumulation curve, then trained a classifier to spot eventual top-10%-reach posts using only their early trajectory.
Results
Follower count explains just 22% of 7-day reach variance (R² = 0.216; Spearman ρ = 0.49). Posts sort cleanly into three reach archetypes — High (16% of posts, median 1.36M views), Middle (36%, 39K), Low (48%, 5.5K) — with up to 250x separation between tiers. A classifier using only the first hour of view data hits 0.72 AUC, rising to 0.79 at two hours and 0.90 within 24 hours, comfortably beating follower-count-only prediction (0.63).
Stack: Python · Pandas · Statistical Modeling · Clustering