Retention benchmarks are relative, not absolute
A 35% first-minute drop can be a disaster or an above-average result. Build a length-banded baseline from your catalogue instead of chasing published averages.
MrBeast's leaked production document treats losing 21 million of 60 million viewers in the first 60 seconds as an above-average result. That is a 35 percent drop in a minute: the kind of number a small channel would screenshot and call a failed hook. At that scale, it is the rate they were trying to beat. The number did not change. The comparison did.
Published retention averages fail for the same reason. "50 percent is good" is a sentence about somebody else's catalogue, somebody else's length, somebody else's click-through mix. It is not a target. The useful number is the median of your last dozen videos in the same format and length band. Everything else is a conversation about a different channel.
Why a borrowed percentage cannot be a target
Three things sit underneath any retention percentage, and none of them travel between channels.
Length. Average percentage viewed falls as runtime grows, even when the video is doing its job. A 12-minute tutorial at 42 percent and a 90-second explainer at 68 percent can both be healthy. Comparing them to one published average punishes the longer one for existing.
Who clicked. High click-through with a steep first-minute drop often means the title and thumbnail pulled in people the video was never for. Low click-through with the same drop often means the opening itself failed. The percentage looks identical. The repair does not. Pairing the curve with CTR is the start of audience retention analysis; treating the percentage as a grade skips the diagnosis.
The opening is a different regime. Misclicks, wrong-audience impressions, and feed churn all land in the first seconds. They are not editorial problems, and they inflate every "first 30 seconds" number you will ever read in a blog. Judge that window against your own first-30-second median, not against a flat line.
YouTube Studio's relative retention view (how the video holds against others of similar length) is closer to the comparison that recommendation actually makes than any absolute percentage. Use it. Just do not let it replace a baseline you computed yourself, because Studio's "similar length" bucket still mixes formats. Your 8-minute listicle and your 8-minute sit-down will share a bucket and should not share a target.
Short-form has the same trap with different branding. Practitioner write-ups of YouTube Shorts circulate a "Viewed versus Swiped Away" band of 70 to 90 percent as healthy, with collapse talked about below 60 percent; Paddy Galloway's frequently cited analysis of 3.3 billion Shorts views is the source most of those posts are repeating. TikTok-for-Business has a separate, primary number: 63 percent of its highest-CTR videos hook within three seconds. Those are useful priors. They are not your floor. A cooking Short that typically holds 81 percent viewed will look "fine" against the 70 percent folklore and still be a miss against its own history.
Build the baseline from the catalogue you already have
You need a sheet, not a philosophy. Twelve videos is enough to start; eight is the floor before the median is noise. Below that, you are describing luck.
- Pick one format. Tutorials, vlogs, listicles, and commentary do not share a retention shape. If you publish two formats, you keep two sheets. Mixing them produces a median that matches neither.
- Band by length. Group into the buckets you actually publish into. A working set for most channels: under 60 seconds, 1–4 minutes, 4–8, 8–15, 15 and over. A 9-minute and an 11-minute sit in the same 8–15 band. A 9-minute and a 22-minute do not.
- Record five numbers per row. Click-through rate, percentage remaining at 30 seconds, percentage remaining at 60 seconds, average percentage viewed, and (on YouTube) whether relative retention vs similar length sat above, at, or below typical. You already have these in Studio. The hook rate and completion rate glossary entries map onto the early and late columns if you need names for them.
- Drop the two outliers, then take the median. The viral one and the dead one will pull a mean out of place. A median of the middle ten is the number you will actually use. Write it at the top of the band: "8–15 min tutorials, median 60s hold = 61 percent, median APV = 44 percent."
- Recompute when the catalogue moves. Recalculate when you have four new videos in the band, or monthly, whichever comes first. A baseline that is 40 videos old is a different channel.
A row looks like this. The numbers are a filled example of the shape of the sheet, not a benchmark to copy.
| Video | Length | CTR | 30s hold | 60s hold | APV | Relative |
|---|---|---|---|---|---|---|
| A | 9:40 | 5.8% | 68% | 61% | 46% | Above |
| B | 11:05 | 4.1% | 71% | 63% | 44% | Typical |
| C | 8:20 | 7.4% | 54% | 47% | 38% | Below |
| Band median (n=10, outliers out) | 9–11 min | 5.6% | 67% | 60% | 44% | — |
Video C is the interesting one. Absolute-benchmark thinking calls 47 percent at one minute a crisis. Against this band, it is a 13-point miss at 60 seconds and a high CTR, which is the overpromise pattern: packaging pulled, opening did not confirm. The repair is not "add more re-hooks in minute six." The repair is the opening, or the thumbnail. Diagnosing a weak opening is the ten-minute join of timestamp to what was actually on screen.
What to do with a miss (and with a hit)
A miss is only a miss against the band it belongs to. Then you still have to name where.
- Below median at 30 seconds, recovered by 60. Opening frame or first line. The body is not the problem; stop cutting it.
- Below at 60, CTR above your median. Packaging mismatch. Test a title or thumbnail before you recut. YouTube's native Test & Compare picks a winner on watch time, not on CTR, which is the comparison you want if the current thumbnail is attracting the wrong people.
- On-median through minute one, then a slow slide. Density. The first-minute number will not tell you this. The retention curve will.
- On-median everywhere, APV down because you added two minutes. You stretched it. Publish at true length next time; do not chase the old APV by stuffing.
A hit is information, not a celebration. If a video beats your 60-second median by a wide margin, write down what minute one actually did: confirmed the thumbnail in the first sentence, opened on the payoff frame, skipped the intro sting. That sentence becomes the next brief. The baseline's job is to make that sentence possible. Without it, every win looks like luck and every loss looks like a reason to copy a large creator's percentage.
Keep the sheet next to whatever you already use for a metrics review. It does not replace Studio. It stops Studio's raw percentages from setting a mood.
One more filter: do not put Shorts and long-form on the same sheet. Viewed-versus-swiped and average percentage viewed are not convertible. A Short that 55 percent of feed viewers actually watch can be a funeral on that surface and would look heroic as a 10-minute APV. Different products, different medians.
The comparison that does not travel
The 21 million figure is useful precisely because it is useless as a target. You do not have 60 million people in the first minute. You will never "beat 35 percent" in any sense that transfers. What transfers is the posture: early drop-off is a rate, scored against peers (or against yourself), not a defect to eliminate. Channels that try to eliminate it start writing openings that stall, because they are protecting a percentage instead of confirming a promise.
If you want a single sentence to put at the top of the sheet: a first-minute hold is good when it beats this band's median, and it is a problem when it does not. Replace "this band's median" with a published average and the sentence stops being true.
FAQ
How many videos do I need before the median means anything?
Eight in the same format and length band is the floor; twelve is where it starts to settle. Below eight you cannot tell a miss from a distribution-luck week. If you publish two formats, you are collecting toward two floors, not one combined pile of 12.
Should I include a breakout video in the baseline?
No. Drop the highest and lowest in the band before you take the median. A breakout moves the mean and then every subsequent upload looks like a decline. YouTube's own "compared to typical performance" view will also jump after a breakout; your sheet should not follow it up.
Can I use Studio's relative retention instead of building this?
Use both. Relative retention is the right direction of comparison (similar length, not a fantasy flat line). It will still mix a tutorial with a vlog of the same runtime. Your sheet is what keeps format in the comparison. If you only have time for one, keep the sheet; it is the one you can segment.
Do Shorts get the same columns?
Keep the same habit, swap the columns. For Shorts, the load-bearing pair is Viewed versus Swiped Away and average watch time at a fixed 24-hour checkpoint, not 30-second and 60-second hold. Do not convert a Shorts viewed-rate into a long-form APV. They do not measure the same decision.