Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

YouTube Marketing › YouTube Analytics

Reading Retention Curves

How to read YouTube audience retention curves: the intro drop, dips and spikes, relative retention, Shorts retention, and the editing changes they call for.
Guideintermediate10 min readUpdated

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 duration 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 (CTR) 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.

ShapeWhat viewers didUsual causeWhat to change
Steep early dropLeft in the openingPackaging mismatch, slow intro, cold audienceCut the preamble; open on the promise; check the traffic source filter
Gradual declineLeft steadily throughoutNormal attrition; pacing without peaksAdd structure and payoffs at intervals; use chapters
Sharp dip that recoversSkipped a segment and kept watchingA sponsor read, a recap, a tangent, a repeated pointShorten or move the segment; signpost it; make it earn its place
Sharp dip that does not recoverLeft at a specific momentThe promise was fulfilled, or a reason to leave was givenDeliver the promise later or restructure; never announce the end mid-video
SpikeRewatched or shared a momentA demonstration, a result, a joke, a detail too fast to catchSlow the moment down; chapter it; build the next video around it
PlateauStayed through a stretchThe section held attentionLearn its pacing and reuse it
Late fall before the endLeft as the ending beganOutro, credits, long sign-off, a visible end screenKeep the ending short; place the end screen over content, not over a blank
Rise above the startReplayed the whole videoLoops, tutorials followed step by step, musicFine; 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 KPI, 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 (Reach) 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.

Illustrative example — not a real result

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.

Frequently asked questions

What is the difference between absolute and relative audience retention?

Absolute retention shows the share of views still playing at each moment of the video, which is the view for finding the exact second where people leave or rewatch. Relative retention rates each moment against YouTube videos of similar length, which is the view for judging whether the video holds attention as well as viewers expect from a video of that length. Use relative to judge and absolute to locate.

Why does my retention curve go above where it started?

Because viewers rewatch. YouTube documents that rises in the absolute retention graph indicate parts of the video being watched more than once, and on Shorts the loop means whole replays are counted, which can push average percentage viewed beyond the video's full length. A rise is a signal of a moment worth building on, not a reporting error.

Does audience retention affect how YouTube recommends a video?

YouTube describes its recommendation systems as learning from whether viewers watch and are satisfied, and watch time is among the signals it names in its help material. It does not publish a threshold or a formula, so treat retention as something that matters without a number attached, and judge each video against the channel's own curves rather than against a target.

Terms used on this page

Where this page is used

Part of these learning paths

Also related

Sources

  1. YouTube Help — YouTube Help

Related