The Anatomy of Attention
The cover gets the click. Retention decides whether the video stops at ten thousand views or keeps going. The graph is an electrocardiogram of the edit.
100%
every graph starts here
15–30%
a normal drop in 30 seconds
50–65%
possible on a tight 10–15 min story
In short. The line always starts at 100% and falls. The shape of that fall tells you where people got bored, skipped, or left. This is the physics of the curve, real averages by length, and the cuts that hold an audience to the last frame.
Why the line always falls
At 0:00, everyone who clicked is still there. From the next second, friction starts. A falling line is not failure. A falling line with the wrong shape is.
0–15 seconds · the check
They test whether the voice, the pace, and the first shot match the cover they just clicked.
The outside world
A notification, a noise in the room, or the next video in the sidebar.
Predictability
When the brain thinks it already knows the ending, the reason to stay drops and so do they.
Four zones
- 1
The hook
The steepest drop, an upside-down hockey stick. Losing 15% to 30% here is normal. You cannot erase it. You can make the line go flat, and stay high, right after 30 seconds.
- 2
The plateau
From 0:30 to about 80% of the runtime, the line should sag, not stair-step. Almost flat means the people who stayed are inside the story.
- 3
The dips
A step down in the middle is a failure at that exact second: a long explanation, a slow transition, a joke that broke the tension, or a 10-second skip.
- 4
The cliff
In the last seconds of a typical video the line falls off a ledge. That is the moment people hear “the video is over” and the host starts saying goodbye.
What “good” actually is
45% is not a grade by itself. On a 5-minute video it is 2 minutes 15 seconds of watch time. On a 20-minute video it is 9 minutes. The algorithm weighs the minutes.
3–5 min
30s. 75–85% still in at 30s
Weak. Under 40%
Typical. 45–55%
Elite. Over 60–70%
8–12 min
30s. 70–80% still in at 30s
Weak. Under 35%
Typical. 40–50%
Elite. Over 55–65%
15–25 min
30s. 65–75% still in at 30s
Weak. Under 30%
Typical. 35–45%
Elite. Over 50–58%
30+ min
30s. 60–70% still in at 30s
Weak. Under 25%
Typical. 28–38%
Elite. Over 42–50%
Averages for long-form, not Shorts. Drawn from VidIQ, TubeBuddy, Retention Rabbit, and notes from producers such as Paddy Galloway.
Kids, challenges, games
5 to 8 seconds
A channel intro or a still conversation can push the 30-second drop past 40%.
Spikes over 100%
They rewind the joke, the scare, or the loud visual. The line goes up. That minute is what they love.
A real episode
A 10 to 15 minute story with a beginning, middle, and end can hold 50% to 65%, because they watch it like a cartoon.
Read it second by second
An average of 48% is the final grade. It does not show which questions you missed. Open the graph and play your own video while the line moves.
If half the room is gone by 0:30
The cover lied
They clicked a burning house or a transformed character. The first 15 seconds are people talking somewhere else.
The hello
“Hey guys, smash like before we start” is an exit button. The best openings are already inside the action in the first 3 seconds.
A sag in the middle
Between 4:10 and 4:45 the line loses 6% in 35 seconds. Play that stretch. Someone stopped the story to explain a rule the picture already showed, or the editor left a long silence with no sound and no new shot. After you see that three times, make a rule: no explanatory beat longer than 10 seconds without something new on screen.
When the line climbs
A spike means people rewound that moment or shared that minute. The face, the sound, or the turn is what this audience actually wants. Build the next video toward that, not toward what looked clever in the script.
Hold them to the last frame
Open a new question before you close the last one
An unfinished question is hard to leave. That is the Zeigarnik effect. If the only promise is paid off at minute 5, half the audience leaves at 5:01. A 10 or 12 minute video needs a stack of loops.
- 0–1
The incident
The problem starts now. One strong picture, matched to the cover.
- 2–4
False win
They almost solve the first step. Then something worse happens. The door opens. What is coming down the stairs?
- 5–8
All is lost
The danger doubles, an ally fails, or a rule changes. The cut gets faster.
- 9–11
The payoff
The confrontation or the reveal that was promised at second zero.
Reset the picture every 5 to 8 seconds
Reframe
A small push in or out on the face that matters.
Drop the music
Silence before a scare or a reveal puts the brain on alert. Nobody closes the tab in that second.
Cut the walk
Long breaths, stumbles, and trips from A to B where nothing happens belong in the bin.
Hard cut on the climax
The goodbye
“So guys, that was it, subscribe, hit the bell.” The moment that tone arrives, they know the story is over and they leave. Forty seconds of farewell on a 10-minute video wrecks the final percentage.
Cut to black
The reveal lands. Eyes hit the lens, or the timer hits zero. Hard cut. No time to predict the ending. The graph stays high through the last millisecond. That is the strongest signal you can send: they drank the whole video.
48 hours after you publish
1
Relative retention
Compare the video with others of the same length. You want “above average” or “high” in the first 2 minutes and in the last minutes.
2
The first big drop
If more than 25% left before 30 seconds, check the winning A/B cover against the first 5 seconds of the edit.
3
Hunt the steps
Find the steepest dip in the middle. Watch it with the editor. Name the cause, and ban it from the next script.
4
The last 15 seconds
A cliff at the end means the ending is still predictable. Shorten it until the cut is the peak of the scene.
Stop guessing
A cover proved by watch time, an opening that matches it in 5 seconds, a middle with no sag, and a hard cut on the climax: that is the stack. Curiosity gets renewed every minute, and the video ends at the top of the feeling. People do not leave in the middle, and they do not leave at the end.
Sources
- YouTube Creator Academy. Key moments for audience retention: intros, spikes, and dips. Studio Analytics, 2024–2026.
- Galloway, Paddy. Audience retention and the anatomy of a video that travels. Benchmarks across entertainment videos, and killing the outro drop. 2024–2025.
- Zeigarnik, Bluma. On finished and unfinished tasks. Psychologische Forschung, 1927. The open loop.
- VidIQ and Retention Rabbit. What is a good audience retention rate, by length. 2025–2026.
- Donaldson, Jimmy, and the MrBeast production notes that circulate in creator case studies, 2023–2026. Every second counts, and the bureaucratic outro goes.