What the retention report shows
YouTube draws audience retention from the moment a view starts. Absolute audience retention plots, for each point in the video, the share of views that were still playing at that point. The line starts at the top and generally falls, because people leave and nobody can join a view after it has begun; where the line rises, YouTube's documentation explains that viewers are rewatching that part, and a segment can sit above the line that precedes it for the same reason.
The report sits in the Content tab under key moments for audience retention, and in Advanced mode it can be filtered and compared. The filters that matter: traffic source, because viewers who arrived from search behave differently from viewers who arrived from suggested; new against returning viewers, because a loyal audience forgives a slow start that a cold one does not; and device, which is worth checking when the edit relies on on-screen text. The guide to reading YouTube audience reports explains what the returning and new viewer split means for the channel as a whole before it is used as a filter here. Comparison lets you overlay a video against another, or against the channel's own typical curve, which is how a single video's shape becomes meaningful.
Two headline numbers summarise the curve. Average view durationTotal watch time of a video divided by its number of views. is the mean time watched per view; average percentage viewed is the same figure as a share of the video's length. Both are useful for ranking videos against each other and useless for diagnosis, because two videos with the same average can have entirely different shapes: one that loses half its audience at the start and keeps the rest to the end, and one that bleeds evenly. The curve is the diagnosis; the averages are the index. The YouTube Analytics hub covers where the report sits among the others and how it fits into a weekly review.
The intro: the first drop
Every curve drops at the start, and YouTube marks the intro as a key moment for that reason: it reports how many viewers are still watching after the opening stretch its help pages define. The size of that drop is the first thing to read and the easiest to act on.
Three different things cause an early drop, and they call for different fixes. The first is a broken promise: the thumbnail and title sold something the opening does not show, and the viewer leaves on discovering the mismatch. The curve falls steeply in the first seconds and the video's click-through rate (CTRShare of impressions that result in a click.) is often healthy, which is the tell; the fix is in the packaging or in opening with the promised thing, and the thumbnails and click-through guide covers the packaging side. The second is a slow start: a branded intro, a greeting, a summary of what the video will cover, a request to subscribe before anything has been earned. The curve falls steadily rather than steeply, then settles once the content begins; the fix is to cut everything before the first moment of value. The third is the wrong audience: the video was shown to people who were never going to want it, which happens when suggested traffic widens quickly. Filter the curve by traffic source and compare; if subscribers hold and suggested viewers leave, the content is fine and the distribution is the issue.
Hooks are a craft of their own. The principles are the ones paid video lives by — open on the payoff or the problem, make the first frame the most interesting one, delay nothing — and the hooks and user-generated content guide sets them out for ads, where the stakes per second are highest. On YouTube the hook has a second job: it must also be honest about the minutes that follow, because a hook that wins the opening and loses the middle shows up as a plateau that never arrives.
Dips, spikes and plateaus
After the intro the curve becomes a record of the edit. YouTube labels some of these shapes as key moments; the fuller vocabulary is below.
| Shape | What viewers did | Usual cause | What to change |
|---|---|---|---|
| Steep early drop | Left in the opening | Packaging mismatch, slow intro, cold audience | Cut the preamble; open on the promise; check the traffic source filter |
| Gradual decline | Left steadily throughout | Normal attrition; pacing without peaks | Add structure and payoffs at intervals; use chapters |
| Sharp dip that recovers | Skipped a segment and kept watching | A sponsor read, a recap, a tangent, a repeated point | Shorten or move the segment; signpost it; make it earn its place |
| Sharp dip that does not recover | Left at a specific moment | The promise was fulfilled, or a reason to leave was given | Deliver the promise later or restructure; never announce the end mid-video |
| Spike | Rewatched or shared a moment | A demonstration, a result, a joke, a detail too fast to catch | Slow the moment down; chapter it; build the next video around it |
| Plateau | Stayed through a stretch | The section held attention | Learn its pacing and reuse it |
| Late fall before the end | Left as the ending began | Outro, credits, long sign-off, a visible end screen | Keep the ending short; place the end screen over content, not over a blank |
| Rise above the start | Replayed the whole video | Loops, tutorials followed step by step, music | Fine; note it when comparing averages |
Read dips with the timeline open. The question is always what happens on screen at that second, and the answer is usually obvious once seen: the cut to a sponsor, the moment the host says "before we get into it", the chapter title that tells viewers the rest is optional. Chapters make dips sharper, because they let people skip precisely; that is not a reason to remove chapters but a reason to make every chapter worth staying for.
Relative retention and comparing videos
Absolute retention cannot be compared across videos of different lengths — a long video will always lose a greater share of its audience by the end than a short one — so YouTube provides relative audience retention, which rates each moment of the video against YouTube videos of similar length, on a scale that runs from below average to above average. It answers a different question: not how many stayed, but how well this video held people compared with what viewers expect from a video this long.
Relative retention is the right view for judging a video's craft; absolute is the right view for finding the moments. Use relative to spot where a video underperforms its peers even though the absolute curve looks ordinary, and to confirm that a long video's low final retention is normal rather than a fault. Then switch to absolute to find the second at which to act.
Comparison within the channel is the most useful comparison of all. Overlay a new video against the last few in the same format, at the same age, with the same traffic source filter. The channel's own curves are the benchmark, because they share an audience, a topic and a style; a published figure from elsewhere shares none of those. Over time a channel learns its normal shape — where its intro settles, where its dips cluster, what its endings look like — and departures from that shape are the signal. Treat retention as a diagnostic that feeds the channel's real KPIA metric tied directly to an objective., usually watch time or returning viewers, rather than as a target in itself.
Retention for Shorts
Shorts are watched in a feed that plays each video without a click and moves to the next with a swipe, so the retention question changes. The thumbnail does no work in the feed, the first second does all of it, and leaving is a swipe rather than a decision to stop. YouTube's Shorts analytics reports viewed against swiped away — the share of people shown the Short in the feed who stayed to watch rather than swiping past — alongside the retention curve, and that split is the Shorts equivalent of click-through.
The curve itself behaves differently. Shorts loop, so a viewer who watches twice is counted twice, and average percentage viewed can exceed the full length of the video; a curve that rises near the end is replays beginning. The intro drop is compressed into the first moment, dips are rare because there is nowhere to skip to except the next video, and the ending is the loop point, which is why Shorts that end where they began hold better than Shorts with a sign-off. Reach (ReachUnique people who saw an ad.) on Shorts comes almost entirely from the feed, which means the retention report and the viewed-against-swiped split are nearly the whole diagnosis; traffic source filtering adds little. The Shorts versus long-form comparison sets out how the two formats differ in distribution and purpose, and why a Short's retention is read for the loop and a long-form video's for the structure. Spikes in a long-form curve are also the first place to look for Shorts, and the tutorial on cutting Shorts from long-form footage turns those moments into cuts that stand alone in the feed.
From curve to script and edit
The point of reading the curve is the next video. A workable process:
- Annotate the curve. For each key moment YouTube marks and each shape you notice, write the timestamp and what was on screen. Watch the video with the curve beside it.
- Classify the cause. Packaging, intro, structure, a specific segment, the ending, or the audience. The traffic source filter settles the last one.
- Decide the change. One change per cause, written as an instruction a scriptwriter or editor can act on: open on the finished result; cut the recap; move the sponsor read after the first payoff; end within a few seconds of the last point.
- Carry it forward. Put the instructions into the script template or the edit checklist, not into a note nobody reads. The same dip appearing in every video is a template problem.
- Check it landed. On the next video, compare the curve at the same age and traffic source mix and confirm the shape changed where it should have.
Retention changes are slow and cumulative, and the audience that gives a channel its curve also changes as the channel grows; the comparison that matters is always against the channel's own recent videos. Where the retention report says the video is fine but views are low, the problem is upstream — packaging or distribution — and the traffic sources guide is the next report to open.
A software tutorial channel fixes a recurring dip
A channel that publishes software tutorials notices the same shape on its last five videos: a steep drop to about 55% of viewers inside the first 40 seconds, a plateau, then a sharp dip of roughly 15 points between 2:30 and 3:10 that only partly recovers. Watching with the curve open shows the cause. Each video opens with a 25-second animated intro and a spoken outline of what the video will cover, and the dip coincides with a recap of the previous video. The team cuts the intro to a single sentence over the finished result, deletes the recap and replaces it with a chapter link in the description. On the next two videos, filtered to suggested traffic and compared at seven days, retention after the opening settles at 68% and 71%, the recap dip is gone, and average view duration rises from 3:50 to 5:05 on videos of similar length.
In paid video, the YouTube consideration setup uses the same curve to recut view-priced explainers and to fix the long-form videos those ads send viewers to.
Before reading retention, a views campaign has to be sized; the impression calculator for video view campaigns estimates impressions and paid views from budget, cost per view and view rate.
When the opening of the curve falls more steeply than usual, the diagnosis for audience retention that drops early reads the opening by traffic source and separates edit problems from routing problems.
What the curve teaches about openings feeds the next edit, and the YouTube video publishing checklist asks whether a video opens with the payoff or the question before it is published.
Common mistakes
- Reading the averages instead of the curve. Two videos with the same average view duration can need opposite fixes.
- Fixing the thumbnail when the intro is the problem. A steep early drop with healthy click-through is a content problem.
- Comparing against published benchmarks. The channel's own recent videos are the only like-for-like comparison.
- Removing chapters because they create dips. Chapters make skipping visible; the fix is a segment worth staying for.
- Ignoring the traffic source filter. A curve that is fine for subscribers and poor for suggested is a distribution issue, not an edit issue.
- Treating a Short's curve like a long-form curve. Loops and swipes change what the shape means.
- Logging the insight and changing nothing. The curve only improves when the script template and the edit checklist change.
