Retention: the number that decides how far a video travels
The retention graph is the only honest feedback short video gives you. Most creators glance at it, see it going down, and conclude that videos go down. Here is how to actually read one.
Improve
Views tell you what happened. The retention graph tells you why, and it is the only piece of feedback either platform gives you that is specific enough to act on. It is also the one most people ignore, partly because every graph goes down and it is not obvious which downs are bad.
All of them go down. You start with everyone who saw the first frame and you can only lose people from there. What matters is the shape of the descent, and there are about four shapes.
Where to find it
- TikTok: open the video, tap the three dots, Analytics. The retention curve sits under “Viewers who watched the full video” alongside average watch time.
- Instagram: open the Reel, View Insights, then the watch time section. Instagram’s is less detailed than TikTok’s, and the number worth extracting is average watch time against the video’s length.
Shape one: the cliff at second two
Fifty or sixty per cent of viewers gone before the video has really begun, then a relatively gentle slope for everyone who stayed.
This is a hook problem and nothing else. The people who stayed are enjoying the video; there simply are not enough of them. The gentle slope after the cliff is the evidence: a bad video does not hold anybody.
The fix: delete the opening. Not rewrite, delete. Find the first genuinely interesting frame and start there. Then add text to that frame, because a meaningful share of the people who left never heard a word. There is more in how to write a hook.
Shape two: the cliff at second six
Retention holds well through the opening, then falls off sharply somewhere between four and eight seconds.
This is the most misdiagnosed shape in short video, because people see a cliff and assume the hook is broken. The hook is fine. It worked, which is why they were still there at second four. What failed is that the video asked for attention and had not yet paid anything back.
Almost always the cause is setup: context, a definition, a caveat, a “before I get into this”. Whatever is between second two and second six is the problem.
“So this is something I have been thinking about for a while, and I want to preface it by saying it will not apply to everyone, but for most people...”
“Most people are wrong about this. Here is the version that actually works, and then I will tell you who it does not apply to.”
Shape three: the sag in the middle
A good open, a good ending, and a visible dip somewhere in between. This is the easiest one to fix because the graph tells you exactly where to look: scrub to the timestamp where the line steepens and watch those ten seconds.
You will find one of these:
- A stretch with no cut and no new information. The camera is static, the subject has not changed, and nothing has appeared on screen for eight or nine seconds.
- A tangent. You went sideways to explain something that was not necessary.
- A repeated point. You made the same argument twice in different words, which the viewer noticed before you did.
- A transition that reads as an ending. A pause, a music change or a wide shot that signals “this is over” when it is not.
The last one is worth dwelling on. Viewers leave at anything that feels conclusive. A summary in the middle of a video is an invitation to stop watching.
Shape four: the flat tail, and the bump
A line that goes down normally and then flattens near the end is good news: the people who got that far are staying to the end, which is exactly what completion rate measures.
A line that goes up at any point is the best thing you can see on one of these graphs. It means rewatching, and rewatching is worth more than almost any other signal. Find the timestamp and understand what happened there, because it is the most valuable second of footage you have produced.
The three-to-five second rule
Underneath every one of the fixes above is one rule: something new should arrive every three to five seconds. New does not mean a cut. It means a change the viewer registers.
Length, one more time
Shorter videos have better completion rates almost mechanically, which has led to the advice that everything should be fifteen seconds. That is one way to game the ratio, and it caps what you can ever say.
The better framing: your video should be as long as the idea and not one second longer. If your retention graph is healthy at 60 seconds, make 60-second videos. If it collapses at 20, your ideas are 15-second ideas and you have been padding them. The graph tells you which, and it is the only thing that does.
A ten-minute monthly review
- Open your last ten videos and write down the ratio: average watch time over length.
- Sort them. Look at the top three and the bottom three, and ignore the middle.
- For each of the bottom three, find the steepest part of the curve and watch those seconds. Name the failure in one sentence.
- For each of the top three, do the same at the flattest part, and name what is holding.
- Write one rule for next month from what you found. One. A list of nine changes is a list nobody follows.
This is the loop that actually improves an account, and it is worth more than any general advice because it is about your footage. Viral Pop will run the same read on a video you import before it is published, which is useful precisely because the retention graph only exists after it is too late to edit.
Read next
How to write a hook that stops the scroll
A hook is not a strong opening line. It is an unresolved one. The difference is the whole craft, and it is the reason most advice about hooks does not survive contact with a real video.
How to write a short-form video script
A 45-second script is about 110 words, which is less than this paragraph and a half. That constraint is the whole discipline, and it is why scripts for short video are written backwards from how people expect.
Why your Reels are not getting views
Eleven real causes, in the order they are worth checking, from the boring technical ones almost nobody looks at to the honest one at the bottom. Shadowbanning is not on the list, and there is a reason.