Every creator has watched one video pull ahead of the rest and called it luck. Sometimes it is. But "luck" teaches you nothing you can use on the next upload.
So we took one video from the 100 MrBeast videos we logged, the one where the crew is stranded on a deserted island [VERIFY: exact title and publish date, video id erLbbextvlY], and asked a narrower question. What does this video do that a typical video in the set doesn't?
The answer we found is a stack. Three different retention tricks fire within about two minutes of each other. Each one is common on its own. Three in a row is not.
Below are three hypotheses for why that might matter, and one small test for each that you can run on your own channel.
Where this comes from
This is our MrBeast library: 100 of MrBeast's long-form uploads, logged by timestamp. Every hook, retention device, edit event and story beat was entered the same way for every video. Nothing below comes from our own channel.
Three limits you should know before reading further:
No one has hand-checked the hook rows yet. Treat every number here as a tendency, not a law.
The library holds no view counts. We can't show from this data that the island video outperformed the others. "Outlier" here means a video that stacks more than the typical one, not a proven record. [VERIFY: public view count against the channel's typical video, before calling it an outlier anywhere.]
One video proves nothing alone. What it can do is point at a pattern worth testing.
Hypothesis 1: the stack matters more than any one device
A retention device is anything that makes leaving feel costly: a question left open, a countdown, a progress bar, a hint of what's coming.
Across the 100 videos:
Device | Videos that use it |
|---|---|
Open loop | 95 of 100 |
Countdown | 61 of 100 |
Progress bar | 31 of 100 |
Foreshadowing | 20 of 100 |
The typical video (the median) uses 2 devices.
The island video's opening act appears to use three types in close succession [VERIFY: timestamps]. A mystery crate is raised and left unopened (an open loop). A night-vision clip teases a coming storm (foreshadowing). A snapped tether lands as a sudden sound cue (a pattern interrupt).
Our hypothesis: a viewer who has seen a device before tunes it out. Three devices layered into the same scene reset attention three times instead of once.
Hypothesis 2: the opening line does two jobs at once
We also logged what each opening hook is built from. Across the 100 videos:
94% of hooks carry a promise.
73% carry stakes.
41% add a curiosity gap.
A hook can carry more than one. The typical opening runs about 10 seconds of speech before the video moves on.
The island video's opening line makes the stranding promise and, in the same breath, teases a surprise waiting at the ruins [VERIFY: quote the line from the log]. That's a promise and a curiosity gap inside the same ten seconds.
Our hypothesis: an opening that does two jobs in that window keeps more viewers past it than an opening that leans on one.
Hypothesis 3: the biggest moment lands late on purpose
Every story beat in the library has a tension score from 1 to 5. The averages by type:
Beat | Average tension |
|---|---|
Setup | 2.2 |
Escalation | 3.4 |
Payoff | 4.5 |
Twist | 4.0 |
Resolution | 2.2 |
In 48 of the 100 videos, the highest-tension beat starts in the last third.
In the island video, the tether-snap twist appears to sit well past the midpoint, after the storm teaser has been hanging unresolved for a while [VERIFY: timestamp of the twist against the video's length]. And the crate? The video keeps it closed for a long time.
Our hypothesis: holding the peak for the last third works because every question opened earlier has to stay open that much longer.
Your three tests
Run them one at a time, not all three at once. If you change three things in one upload, you won't know which one moved the number.
Test 1: the stack. Count the retention-device types in the first two minutes of your last upload: open loop, countdown, progress marker, foreshadowing. If you find one or none, script your next video to place two types inside the same 30 seconds. Then read your retention graph just past that point.
Test 2: the two-job opening. Score your last three opening lines against three boxes: a promise, stakes, a curiosity gap. Most tick one. Write your next opening to tick two inside the first ten seconds. Compare that video's average view duration with your channel's usual.
Test 3: the late peak. Map your last video's beats (setup, escalation, twist, payoff) onto its timeline. If the biggest moment sits in the first half, hold it for the final third of your next upload. Watch whether the retention curve holds longer before it drops.
Write down what you changed and what you expect before you publish. Give each test a full upload cycle before you judge it.
What we still can't tell you
We don't know which of the three matters most, or whether it's the combination that does the work. One video can't answer that. It needs more videos and more channels, and it needs your tests.
Before any of this matters
Every one of these tests happens after the click. None of them helps a video nobody opens.
A video stuck at a few hundred views usually wasn't watched and judged. Its title and thumbnail were shown, judged in about a second, and scrolled past. Nobody got far enough to meet the crate.
That's what my book is about. Make Them Watch shows you how to build a title and thumbnail a stranger can't scroll past: six checks you can score on any package (Seen, For me, Payoff, Open question, Easy, Believed), a routine of twenty titles and three thumbnails for every video, and the first ten seconds written before you film, so the promise is kept. It's built on a 1928 publisher's sales ledger, a hundred years of counted advertising tests and 7,646 of today's YouTube videos.
Part 1 is already in your inbox, free. The whole book is {PRICE: book}, with a 30-day refund and no questions. At checkout you can add 50 Tested Ways to Win the Click for {BUMP}: fifty one-page ideas for what to put in each test.
[Get Make Them Watch →]({LINK: art21-book})
Xander
P.S. Next video: back into the 61 of 100 videos that use a countdown, to see whether a countdown on its own goes with that late peak, or only when it's stacked with something else.