What is an audience retention graph?

Reading the axes

The horizontal axis is the video's own timeline and the vertical axis is the share of the starting audience still there. Every retention curve falls overall, because people can leave partway through but nobody new arrives in the middle, so a falling line is not bad news by itself. What matters is how fast it falls and where the rate changes. YouTube's key moments report names the parts worth reading. The intro figure is the percentage of the audience still watching after the first 30 seconds. Top moments are stretches where almost no one dropped off. Spikes are moments that were rewatched or shared, and dips are moments that were skipped or where viewers stopped watching altogether.

How the curve is built

Each play is laid against the timeline, and every second a viewer actually watches adds to that second's count. A viewer who jumps from second 10 to second 40 is counted at the start and after the jump but not in between, which is why skipped sections show up as dips and why a line can rise where people rewind or skip ahead to a particular moment. Autoplayed starts, clicks from search and visits from a link all feed the same line, although they behave differently. That is also why a curve built from a few dozen plays is noisy. One or two people leaving creates a visible notch, and the shape only becomes trustworthy once enough plays have accumulated to smooth individual behaviour out.

When it matters most

It matters most for videos with a job in the middle, such as tutorials and demonstrations, where a viewer who leaves at step two never learns step three. Published research explains why length is the first suspect. Vidyard's analysis of 943,305 business videos found 65 percent of viewers stay to the end of videos under one minute, against 20 percent for videos over 20 minutes. Wistia's report across 13 million videos found educational and tutorial videos hold attention better than most formats as length grows. An earlier study of 6.9 million viewing sessions on the edX course platform found shorter videos were much more engaging. A curve tells you whether your own video follows that general pattern or breaks it at one specific scene that can be fixed.

Common mistakes

Blaming length for a cliff at second four is the first. An early cliff is almost always a mismatch between the promise and the opening, and trimming the ending will not repair it. The second is cutting the scene where the line is lowest when the fall actually starts a scene earlier, since people often leave a moment after the part that lost them. The third is editing on curves from too few plays, where a handful of viewers creates dips that mean nothing. The fourth is treating every spike as praise. People rewind for good and bad reasons, and a spike on one step of a tutorial can simply mean the instruction was unclear the first time it was said. The fifth is comparing curves from videos of very different lengths on the same chart, where a two minute video will always look worse at its end than a thirty second one.

Where it shows up in the product

Retention is measured by wherever the video is hosted; the video builder here draws no curve and reports no watch time. What it offers is scene level structure, so the moment a curve points at is one scene rather than a blur of footage. Every video is a sequence of scenes with their own durations, and length settings are scene budgets rather than stopwatch targets: short builds 3 to 6 scenes, standard 4 to 9 and long 6 to 12. Tutorial steps run 5 to 9 seconds each with 10 to 25 words of narration, and AI tutorials come back on the Calm pace. When a curve says step four is being skipped, the usual fix is a tighter narration line or a clearer screenshot on that single step.

Questions people ask