Across 2,066 reels, none of the things we measure cleanly separates a breakout from a dud
Across a 127x spread in reach, our own hook and retention scores move by a fraction and length does not move in one direction at all. 17 September 2026.
What the numbers say
From the bottom fifth to the top, views per follower goes 0.08, 0.22, 0.52, 1.56, 10.17. That is a spread of 127 times. Over the same five bands our hook score goes 8.64, 8.67, 8.71, 8.76, 8.85, a total movement of 0.21 of a point, and our retention score goes 78.2 to 81.9, a movement of 3.7. Median length does not even move in one direction: 45.3 seconds, then 47.9, then 52.0, then 52.0, then 42.9. Speech is flat at 85 to 90 percent throughout.
There is a mechanical reason the score columns barely move, and it is worth stating because it tells you how much weight to put on a score anywhere. Across the same 2,066 reels, 71 percent were given a hook score of exactly 9 out of 10. When seven reels in ten share one value, no average built from that column can separate anything, whatever is happening underneath. The retention score is a little wider but the same problem applies: its lower quarter and its upper quarter are both 82. So this page is not evidence that hooks do not matter. It is evidence that a compressed score cannot tell you which reel will travel, and that is a fact about the instrument as much as about the reels. The honest answer to what makes a video go viral, from the largest set of analysed reels we hold, is that nothing we can measure before publishing predicts which fifth a reel lands in. That is worth more to you than a confident wrong answer. What the same 2,066 reels do give you is a target: the 90th percentile is 10.15 views per follower and the median is 0.52, so going viral in any sense you can act on means clearing roughly ten times your own audience, while a typical reel reaches about half of it.
Sample of 2,066, measured on 2026-09-17. Scrptly production database, queried 2026-09-17. Instagram Reels only, deduplicated by shortcode, one query run.
The measurement
Every reel ranked by views divided by its own creator follower count, which controls for audience size, then cut into fifths. The five bands add to 2,066. Hook and retention are mean scores within the band. Views per follower and length are medians. Share with speech is the proportion whose transcript holds real spoken words rather than a trending audio placeholder.
| Band | Reels | Median views per follower | Mean hook score | Mean retention score | Median seconds | Share with speech |
|---|---|---|---|---|---|---|
| Bottom fifth | 414 | 0.08 | 8.64 | 78.2 | 45.3 | 86% |
| Second fifth | 413 | 0.22 | 8.67 | 79.2 | 47.9 | 89% |
| Middle fifth | 413 | 0.52 | 8.71 | 80.1 | 52 | 90% |
| Fourth fifth | 413 | 1.56 | 8.76 | 80.8 | 52 | 87% |
| Top fifth | 413 | 10.17 | 8.85 | 81.9 | 42.9 | 85% |
One Reel, beat by beat
The most extreme breakout in the corpus relative to its audience: 11,088,685 views from an account with 34,955 followers, which is 317 times its own following. It runs 11.8 seconds. Hook score 9, retention 82, both of them ordinary, which is exactly the point this page is making. The beat labels are our analysis wording, lightly edited to house style, and are not a transcript.
| At | What happens |
|---|---|
| 0:00 | Hook: an on-screen provocative question over the first joyful clip. The only spoken line is, but what if most people are good? |
| 0:01 | Evidence clips one and two: quick cuts of different people in joyful moments |
| 0:03 | Evidence clips three and four: close-ups of genuine smiles, laughter and connection |
| 0:06 | Evidence clips five and six: more variety, taking in playful, romantic and communal scenes |
| 0:09 | Payoff: a final heartwarming clip that seals the emotional argument |
| 0:11 | End: the last image lingers a beat before the cut, with no spoken call to action |
What to do next
- Judge a reel against your own follower count rather than against somebody else. Ten times your following is roughly the 90th percentile here and the median reel reaches about half of it, so that is a target you can read straight off Insights.
- Stop treating a scored hook as a forecast. At the 8 and 9 where almost every reel in this data sits, the score does not tell you which fifth the reel will land in. Our hook score page shows the rare perfect 10 does pull away, and that is the narrow case where the number earns its keep.
- Since the measurable inputs do not separate the bands, the thing worth changing is the idea rather than the execution. Our reel ideas page has fifteen that ran past their own audience, and the smallest account on it had 9,127 followers.
- Treat the one thing that did vary as a warning rather than a tactic. Median length runs 45.3, 47.9, 52.0, 52.0 and 42.9 seconds across the five bands, which is not a trend in either direction. Anyone selling you an optimal reel length is reading a pattern into numbers that do not hold one.