Short-form video retention: how to decide what to edit

Retention describes how long viewers watch and how that changes during a video. The examples and steps in this guide use YouTube Shorts metrics. To decide what to edit, locate the drop and replay what happens there. A low average view duration alone cannot identify the scene that needs work.

A product video might attract interest at the start and lose it during a pause further in. Changing the opening line would leave that pause untouched. This example shows how to separate the two.

A backpack video loses viewers during the demonstration

Suppose a shop publishes a 30-second video answering whether a laptop fits in a backpack. The product and figures below are fictional; they illustrate how to read the data.

During the first six seconds, we see a laptop slide into the compartment and hear: “Here’s how a 14-inch laptop fits inside this backpack.” Four seconds of logo and music follow. The demonstration then continues with the zipper, the space left inside and a shot of someone wearing the backpack.

The team checks these points on a simplified retention curve, expressed relative to the playback base used in that report:

Swipe the table to see all columns.

Between seconds 6 and 10, the value falls by 31 percentage points. Between seconds 10 and 20, it falls by four points. The logo pause deserves the first review: it interrupts a question the video is still answering. The curve locates that coincidence; it cannot establish why each viewer left.

Our first edit would remove the pause and connect the laptop shot directly to the zipper shot. We would keep the view of the remaining space because it completes the answer. The logo could appear in a brief closing shot if the team needs it. We would keep the opening unchanged in this test to assess one specific edit.

Before cutting, we would also check for missing audio, text covering the product or a confusing transition in that section. The same drop can have different causes. In its audience retention documentation, YouTube distinguishes viewers leaving from skipping, and notes that a spike can reflect interest or a section viewers need to replay to understand.

Three YouTube Shorts metrics that answer different questions

For Shorts, separate how often viewers stayed to watch, how much those who stayed watched and what happened during the video. YouTube distinguishes them in its content performance definitions.

Imagine another fictional report: the video appeared 10,000 times in the Shorts feed and its “Stayed to watch” metric reads 40%. Separately, average percentage viewed is 80% and average view duration is 24 seconds for a 30-second Short.

The duration calculation is 24 ÷ 30 × 100 = 80%. For Shorts, that average is calculated from engaged views and their corresponding watch time. It does not mean that 80% of the 10,000 feed appearances led to a completed video. Nor does it mean that every play lasted exactly 24 seconds. The average can combine very different viewing patterns.

If initial viewing is weak compared with similar Shorts on the account, we would inspect the first shot and whether the question is immediately clear. If that initial metric is close to the account’s usual performance but a concentrated drop appears later, we would inspect that section. Both readings inform the decision.

For Reels, check the dashboard definitions before transferring the calculation. When Meta introduced watch-time metrics, it explained that total watch time includes replays. That 2023 announcement provides background; for a current report, keep the metric name and base your account actually displays. Do not automatically divide by accounts reached or compare the result with a Shorts average.

What to inspect, depending on where the curve drops

If the loss is concentrated at the start, watch the first shot without context: is it clear what the video will show? A screen with the shop’s name might occupy the time when viewers should see the open backpack. Test an opening that shows the action and makes a promise the rest of the video delivers.

If it drops during an explanation, check whether the image adds information. In our example, closing the zipper helps demonstrate that the laptop fits; repeating the product name over a static shot could delay that evidence. Cut the repetition before speeding up the whole voiceover.

If there is a spike, watch and listen at that moment. It might be a useful detail worth developing, but it could also be a measurement that disappears too quickly or a hard-to-read label. Holding the shot longer might help more than adding another cut.

If the drop comes at the end, check whether the answer is already complete. If the goal was to demonstrate the laptop fitting and that demonstration is clear, viewers leaving before the logo does not invalidate the whole video. To assess catalogue visits, you also need data for the link or that action.

These are editing hypotheses. A curve’s shape cannot prove a lost sale, a platform penalty or the feelings of someone who left.

How to review the next version

Save the original video and record the section you changed. Compare pieces with similar topics, lengths and distribution, using the same window after publication. Separate promoted content when the report allows it. Two consecutive organic posts may reach different audiences; a later improvement alone does not isolate the effect of the cut.

In YouTube Studio, open Content, select the video and open Analytics to inspect retention. Data can take one or two days to process. Automatic key-moment labels have their own conditions, including a video at least 60 seconds long with 100 views. Their absence on a 30-second Short does not establish that everything is fine or that the report is broken. Use the data your account actually provides.

If a curve is not available yet, keep average view duration and initial viewing metrics with their definitions. You can review the edit, but you cannot attribute the loss to a second the report does not show.

Before editing, write down what you will keep, what you will remove and what observation you want to check. For the backpack: “Keep the laptop going in. Remove the logo pause. Check whether the loss between the first demonstration and the zipper shot is still concentrated there.” You can put that change into a script sheet pairing voice and visuals or document it using the criteria for content testing.