The Second Decision: Where Early Retention Loss Falls in YouTube Shorts

Every time you watch a short-form video, you make two decisions: whether to stop scrolling and whether to keep watching.On YouTube, the first decision shows up as a metric called “Stayed to watch”. The second shows up as a metric called retention, which can be visualized by the retention curve: the share of the audience still watching at each moment. Two short-form videos could have the same retention, holding viewers equally well overall, but still lose them at different moments. Does that matter? Most Shorts are experience goods (Nelson, 1970): you only learn if one is worth watching by actually watching it; I ask whether where viewers quit is associated with how well a video performs.I analyzed 2,597 Shorts from 11 channels and three creators, including my own channel, GabeSweats (2M+ subscribers). I measured where each video’s early dropoff (viewers leaving within 1–30% of runtime) happened, and compared that with its views per day.On “promise” channels, videos open by promising, directly or indirectly, to explain or close some gap in information (say, “who is the smartest person ever?”, which promises... well, you get it). On these channels, a bigger share of dropoff between 10–20% of the video’s runtime was associated with fewer views per day. Within the same channel, adjusting for length, year and average retention, a typical difference (one standard deviation) went with about 39% fewer views per day (partial r = −0.28, N = 1,734; −0.17 with the strictest controls). Losing people at the start of a video was surprisingly associated with more views, while losing them after the 20% mark was mostly unrelated.On “format” channels—where you know what you’re getting, like a channel that posts almost nothing but clips from the same TV show—the same link originally showed up under the simplest comparisons, but under the strictest controls, the link nearly disappeared (−0.05).Much of the promise result comes from just two of the eleven channels, and by nature this is an observational study and does not establish causality.