Comments tell you what a few people said. Analytics tell you what everyone did. And of all the numbers an app gives you, the retention graph is the most useful, because it shows the exact second people decided to leave.
It doesn’t tell you why. But it tells you where to look, and that’s most of the battle.
What the graph shows
A retention graph (YouTube calls it audience retention; other apps use similar names) starts at 100%, everyone who started the video, and slopes down as people leave. Each point answers one question: what share of viewers were still watching at this second?
Every graph slopes down. That’s normal. What matters is the shape: where it falls fast, where it holds and where it bumps.
The common shapes, and what they usually mean
| Shape | What it might mean | What to try |
|---|---|---|
| Steep drop in the first 1 to 3 seconds | The opening didn’t grab, didn’t match the cover, or reached the wrong people | Rework the hook; check where viewers came from |
| Gentle, steady slope | Normal | Tighten for a flatter line |
| Sudden dip in the middle | A slow bit, a repeated beat, a confusing step, or people skipping to the answer | Watch that moment, and what comes before it |
| Flat and high | Excellent | Study it and repeat what worked |
| A bump up | People rewatched or shared that part, or got confused and replayed it | Watch it; delight and confusion both cause bumps |
| Big drop near the end | People left once they got the payoff, or the ending dragged | End sooner after the payoff |
| Above 100% at the start | People looped back to the beginning | Your loop worked |
Treat each row as a first guess, not a diagnosis. Then go and check.
Open the video at every bend
The graph is a map, not a verdict. For every sharp drop or bump:
- Note the second it happens.
- Open the video and watch a few seconds before and after that moment.
- Ask what a viewer was waiting for right then, and whether they got it.
You’ll often find something obvious once you look: a five-second pause while you set something up, a step you explained twice, a shot where nothing changes, text that covers the thing people wanted to see.
Look upstream
People show you where they gave up, not where the problem started. A dip at second 7 might be caused at second 1: a hook that promised something the video wasn’t delivering, or a setup so confusing that people hung on for a few seconds and then quit.
So when you find a drop, check the earlier moments too. Did the opening make a promise? Was the video visibly moving toward it?
Compare like with like
A retention graph only means something next to the right comparison.
- Same format. Comparing a skit to a tutorial tells you very little. Compare a tutorial with your last five tutorials.
- Similar length. Percentages flatter short videos and totals flatter long ones. A 10-second clip that 95% finished didn’t necessarily beat a 90-second video people watched 45% of. Group videos into length bands (under 30 seconds, 30 to 90, longer) and compare within a band.
- Same moments. Check how many people were still there at 3 seconds, at the halfway point and just before the payoff. A longer video that holds more people to its payoff can be the stronger one, even with a lower overall percentage.
- Same door. People from search often skip straight to the answer. People from the recommendation feed came to be entertained. The same video can show two very different graphs for those two groups, so check traffic sources.
Look for patterns, not single results
One video’s numbers are noisy. The day, what else was trending and who happened to see it first can all swing a result. So don’t rewrite your whole approach because of one graph.
Instead, line up your last 10 to 20 videos side by side: cover, hook, length, format and the share still watching at 3 seconds and at the halfway point. What do the best ones have in common? The worst? Patterns you’d never see one video at a time jump out when they’re in a row.
Track the typical video, too, not your best one. Line them up from worst to best and take the middle result (the median). It ignores one lucky hit and one disaster, so it tells you whether your videos are actually getting better.
When there’s no graph
Sometimes a report shows a dash, a zero or “not enough data.” That usually means too few people saw the video for the app to show it, often to protect viewers’ privacy. It isn’t an error, and it isn’t a verdict on the video. Try a longer date range or wait for more views. And if an app doesn’t offer a number at all, write “missing” in your notes rather than guessing it.
Every app hides retention in a different place, and names it differently. The Analytics Decoder shows where to find it on each one and what the numbers really measure.
Turn it into one change
The point of reading a graph is to decide what to do next. Pick the one biggest drop across your recent videos and make one change aimed at it in your next three videos: a faster first second, a cut middle section, an earlier payoff. Then compare those three with the three before.
If you’re not sure which part of your process is causing the drops, the Weakest Step Finder asks a few questions about your recent videos and points to the step to fix first, with a seven-day plan.
Try it
Open your three most recent videos of the same format. For each, write down the share of viewers still watching at 3 seconds, at the halfway point and just before the payoff. Find the biggest drop, open the video at that second and watch the five seconds before it. Write one sentence about what you’d change.
