How to Read a YouTube Retention Graph: Six Shapes and What They Mean

A retention graph is a record of decisions: every dip marks a moment where viewers chose to leave. The shape tells you what kind of moment it was — and each shape has a different fix.

Updated June 2026 · By the Retti team

Shape beats number

Most creators read their retention graph as a single number and stop. The number tells you how much got watched; the shape tells you why. A curve that drifts gently down to 40% and a curve that cliffs to 40% inside two minutes describe two different videos with two different problems — and two different fixes.

The six shapes below cover nearly every curve in Retti's corpus of real, creator-submitted retention graphs, almost all from videos longer than eight minutes. None of the directional guidance in this guide applies to Shorts or to videos under eight minutes, because the corpus does not include them. If a term is unfamiliar, the glossary defines it; what is audience retention covers the metric from zero.

The opening cliff

What it looks like: a steep drop that starts at 0:00 and flattens somewhere inside the first minute. Every video has one; the only question is size. The median video in Retti's corpus has shed a big share of its audience by 30 seconds and keeps slipping through the first minute — and a large share of videos lose a heavy chunk in the first half-minute alone.

What causes it: the audience sorting itself. Misclicks leave first, then viewers whose expectation does not match the opening frames. A slow start or a promise mismatch deepens the cliff, but no video escapes it entirely.

The fix: shrink it, because you cannot remove it. If you sit below the bottom-quartile line at 0:30, re-cut the opening so the title's promise is on screen within three seconds and a first payoff lands before 0:30.

The steady slope

What it looks like: after the cliff, a smooth downhill drift with no sudden steps.

What causes it: ordinary decay. Phones buzz, attention runs out, and many viewers leave satisfied with the answer they came for. Some slope is the cost of existing.

The fix: judge the gradient before you treat it. Across the middle half of a typical video, the median curve loses only a modest amount — decay through the body is normally slow and steady once the opening cliff is past. If you are losing much more than that without a single ledge to blame, the problem is pacing rather than one bad moment: points repeated after they have landed, segments that outstay their welcome. Tighten every section that restates something already made clear; the system-level version of this work is in how to increase AVD.

The ledge

What it looks like: the slope suddenly steepens, sheds several points fast, then settles back to its old angle — a step cut into the curve.

What causes it: one specific moment that pushed people out. The usual suspects are a tangent, a point repeated once too often, a lost thread the viewer could not follow, an unannounced sponsor segment, or a topic shift the click never signed up for.

The fix: scrub to where the drop begins and watch the 20 seconds before it. Viewers decide to leave slightly before they actually go, so the cause sits upstream of the visible fall. Then cut the segment, shorten it, or signpost it — a line like 'two minutes on settings, then the results' carries people through a stretch that would otherwise eject them.

The rewatch spike

What it looks like: the curve rises above its surrounding level, meaning that moment was watched more times than the footage before it. Spikes are rare — only a small fraction of curves show a clear rise after the opening.

What causes it: viewers scrubbing back. Something was worth seeing twice: a step in a tutorial, a price or a number, a reveal, a detail that rewards a second look.

The fix: nothing is broken — something worked. Identify exactly what earned the replay and engineer more of it: denser key moments, visuals that carry information, reveals worth pausing on. A spike also marks your best clip material for promotion, because the audience has already voted on it.

The end-screen dive

What it looks like: a sharp plunge in the final seconds of the video.

What causes it: the video is over and viewers can tell. For most videos the drop across the final tenth of the runtime is small and normal. A dive confined to the last seconds is an audience leaving a finished video, not a failed one.

The fix: mostly, leave it alone. What you can control is when the dive starts. Wrap-up phrases — 'so that's pretty much everything' — start the exodus wherever they appear, even minutes early. Keep the conclusion short, skip the begging, and point viewers at a specific next video while they are still watching.

The plateau

What it looks like: a flat stretch, sometimes minutes long, where almost nobody leaves.

What causes it: a promise actively being kept. An open loop in the middle of resolving, an escalating sequence, a story beat the viewer has to see land, or information dense enough that leaving costs something.

The fix: a plateau is your retention engine caught on camera. Find where it starts, name what changed at that moment — format, pacing, stakes, density — and build your next videos so more of the runtime looks like that stretch. This is the rare case where the graph tells you what to do more of, not less.

The five-step diagnostic

1. Read 0:30 first. Below the bottom-quartile line, the opening is the priority — nothing later in the video matters until the cliff shrinks.
2. Mark every ledge. Timestamp each sudden step-down through the middle of the curve.
3. Watch the 20 seconds before each one. Name the segment that pushed people out. Patterns across several videos matter more than any single ledge.
4. Find your stickiest stretch. Locate the flattest minute and any rewatch spike, and write down what you did there.
5. Compare your quarter marks to the medians. Check your retention at the quarter mark, midpoint, three-quarter mark and end against the benchmark table for your video length. Above the median at all four and your bottleneck is probably packaging, not retention. Below them, the ledges and the slope tell you where to start cutting.

The full benchmark tables, including the percentile bands, are in Retti's retention benchmark study. You can run the workflow by hand in YouTube Studio — or paste the video into Retention Lab, which finds each shape and maps it to the transcript moment that caused it. When the fix means re-cutting, Editing Lab reviews the new edit before you publish.

Frequently asked questions