01
What does the average conceal?
That most clips have two populations rather than one. A large group leaves in the first seconds and a smaller group watches through, sometimes more than once. The average sits between two behaviours that almost nobody actually exhibits, which is why the retention graph is more informative than the single figure.
02
Two populations inside one average
An average that sits between two behaviours nobody actually exhibits is why the graph is more informative than the single figure.
| Group | Typical behaviour | Effect on the average |
|---|---|---|
| Left in the first seconds | Most of the audience | Pulls it down hard |
| Watched most of it | A minority | Pulls it up |
| Watched twice | A small minority | Can push it above the clip length |
| The average itself | Between two behaviours | Describes almost nobody |
| The retention graph | Shows both groups | This is what to read instead |
03
Reading the retention graph second by second
The graph usually starts with a steep drop in the first few seconds and then flattens. The height of the flat part matters more than the depth of the first drop, because it tells you what share of viewers who gave the clip a chance stayed with it. A clip that loses half its audience early but keeps nearly all the rest is often healthier than one that declines slowly all the way through.
Look for bumps and cliffs after the opening. A sudden cliff in the middle usually marks a specific moment: a tangent, a pause, a cutaway to something dull. A bump, where the line rises, means people rewound to see something again, which is the strongest signal of interest a short clip can get.
04
Comparing your own clips fairly
Compare clips of similar length. A fifteen-second clip and a fifty-second clip produce different averages even when viewers liked them equally, because the longer one has more room to lose people. Grouping clips into length bands, such as under twenty seconds, twenty to forty, and over forty, makes the numbers comparable.
Compare clips from the same source type as well. Interview clips, tutorial clips and reaction clips attract different viewing habits, so a figure that is weak for one kind can be strong for another. Your own history within each category is the only benchmark that means anything for your channel.
05
Turning the numbers into edits
Once you know where viewers leave, go back to the source and watch that exact moment. Most of the time the cause is obvious in hindsight: the speaker repeats themselves, the point arrives too slowly, or the clip keeps running after the payoff. Those are editing notes you can apply to the next clip from the same show.
Keep a short list of the causes you find. After ten clips you will see which ones recur, and fixing the most common cause across your next batch will move the average more than any single change to one clip ever could.
06
Read the shape, not the average
- Read the graph alongside the average. The shape explains what the number blends.
- Compare against your own history only. Published benchmarks mix formats that are not comparable.
- Watch the trend, not the individual figure. Single clips vary enormously.
Post a few and compare their graphs. Nothing to pay for the first.
Questions people ask
- Can it exceed the clip length?
- On platforms that count loops, yes, and it is a strong signal when it happens.
- Why is my average low on a good clip?
- Because early leavers pull it down regardless of how well it held the people who stayed.
- Is there a target figure?
- Not a portable one. Format, length and how the audience arrived all move it too much for a benchmark to mean anything.
- Does length change what is achievable?
- Yes. Longer clips have lower proportions almost mechanically, which is why comparing across lengths misleads.
- What is a better single number?
- The proportion who reached the end, since it does not blend two behaviours.