The length tax
Competitors · VidCC Research · 5 Oct, 2026
A 48% average looks weak next to a 66% one. In our data it was simply what a typical 15-minute video achieved. Length changes average retention far more than it changes the opening, so it is the first thing to match before comparing curves.
Introduction
Comparing your retention with another video's is only useful if the comparison is fair. We measured how much video length alone moves the numbers, so creators know what to hold constant.
Key takeaways
- Longer videos had lower average retention: ρ = −0.51 (95% CI −0.63 to −0.36).
- Median average retention was 66% for 1–5 minute videos, 61% for 5–10 minutes, and 48% for 10–20 minutes and for longer videos.
- Retention at 0:30 changed far less, from 74% to 63% (ρ = −0.28, 95% CI −0.44 to −0.10). Long videos open about as well and lose people later.
- Compare a video only with others of similar length. YouTube's own typical retention benchmark does the same.
The data behind this study
This study uses 158 YouTube Studio audience retention graphs that creators chose to post publicly, mostly on X, between 2015 and 2026. We matched each graph to its YouTube video and channel, then traced the curve into 100 points, one for every 1% of the video's runtime, exactly as Studio plots it.
The set covers 146 channels: 123 long-form videos (longer than 60 seconds) and 35 Shorts. This study uses the 123 long-form videos. We report results only in aggregate. No individual video or channel is identified.
Definitions
- Average retention
- The mean of the retention curve across the whole video. It approximates YouTube's average percentage viewed.
- Retention at 0:30
- Absolute retention 30 seconds into the video.
- Retention at end
- Absolute retention at the last point on the graph.
Length moves the average far more than the opening
Across the four length groups, retention at 0:30 fell by 11 points. Average retention fell by 18 and retention at the halfway point by 18. The opening of a 20-minute video holds viewers nearly as well as the opening of a 3-minute one; the difference builds up over the extra minutes.
| Length | Videos | At 0:30 | Halfway | Average | At end |
|---|---|---|---|---|---|
| 1–5 minutes | 45 | 74% | 65% | 66% | 41% |
| 5–10 minutes | 30 | 71% | 61% | 61% | 35% |
| 10–20 minutes | 34 | 67% | 47% | 48% | 32% |
| Over 20 minutes | 14 | 63% | 47% | 48% | 27% |
Average retention vs length: ρ = −0.51 (95% CI −0.63 to −0.36).
What this means for your channel
YouTube Studio's typical retention view compares a video with your 10 latest videos of similar length for exactly this reason. Use the same rule whenever you benchmark against yourself or a competitor.
Match length before you compare
A 15-minute video at 48% average retention is typical in our data, not weak. Measure it against videos of 10 to 20 minutes.
In long videos, watch the slope
Past ten minutes, a steady downward slope costs more than any single dip. Trim weak stretches between segments rather than chasing one fix.
Still win the opening
Length barely changes what happens in the first 30 seconds, so every video, short or long, is judged first on its opening.
Methodology
Sample. 163 publicly posted Studio retention graphs, each matched to its YouTube video. We excluded four whose video had been re-cut after the graph was posted (the graph's length no longer matched the video) and one whose trace could not be verified, leaving 158 videos from 146 channels. 109 come from Japanese-language channels, 45 from English-language channels and 4 from others. Channels range from 6 to 1.75 million subscribers, with a median of 4,090.
Tracing. Each graph was traced into 100 points of absolute retention, as plotted, and every trace was checked visually against an overlay of the original screenshot. 149 of the 158 traces were rated high confidence and 9 medium.
Measures. Retention at 0:30 is read from the curve at 30 seconds for long-form videos. Average retention is the mean of the curve and approximates average percentage viewed. A rewatch rise is an increase of at least 3 points over the lowest value in the preceding 4% of the runtime, outside the first and last 5%. Views and subscribers were recorded when the dataset was assembled in 2026, not when each graph was posted.
Statistics. We report medians and Spearman rank correlations (ρ). 95% confidence intervals come from a bootstrap with 2,000 replicates, resampling videos with replacement and recomputing each statistic.
Caveats. This is an observational study, not an experiment, so it shows association, not cause. Creators chose which graphs to post, the sample leans towards smaller and Japanese-language channels, and tracing from screenshots can be off by about a point. Treat the results as a description of these 158 videos, and a strong hint about many others.
About this research
VidCC Research keeps a private library of 2,151 findings on retention, storytelling, sound, packaging and analytics. Each one is tied to the source it came from and tagged by topic, and the library informs how we build and test VidCC's analysis. This article explains one part of it in our own words.
Where we cite numbers, they come from our analysis of 158 publicly posted YouTube Studio retention graphs. Studies that rely on them include the full methodology.