Good abandonment: when drop-off isn't failure
A viewer who leaves after getting the answer is not the same as one who leaves bored, but average percentage viewed scores them the same. What to use instead.
Two videos, both averaging 43% viewed. The first is a five-minute tutorial where the fix arrives at 2:10 and most of the audience leaves within thirty seconds of seeing it work. The second is a five-minute story where people drain steadily from 0:40 onward and nobody in particular is satisfied.
Average percentage viewed cannot tell those apart. They are not remotely the same video, and if you treat the first one as a retention problem you will "fix" it by making it worse.
What "good abandonment" means, and what it doesn't
The phrase comes out of information retrieval, where it describes a search session that ends without a click because the searcher already got what they needed from the results page. Abandonment, but not failure.
Creators have borrowed it for video, and the borrowing is sound as a diagnostic — a viewer leaving because their question was answered behaved differently from one leaving because they got bored, and the exit alone does not distinguish them.
Be careful about how far you take it. I could not find a platform policy document that names good abandonment as a ranking concept, so treat it as a way to read your own data rather than a documented change to how anything is scored. What is publicly stated is narrower and more useful: platforms rank on a bundle of signals, not one. Adam Mosseri has named Instagram's core trio as watch time, sends per reach and likes per reach, with sends weighted more than likes for reaching people who don't already follow you. That is three signals, and only one of them is duration.
The practical version: average percentage viewed is a summary statistic, and summary statistics throw away the shape. Shape is where the diagnosis lives.
The shape test
Four exits, four meanings. You can tell them apart in about fifteen seconds on the retention curve.
| What the curve does | What it means | Is it a problem? |
|---|---|---|
| Holds flat, then a sharp drop right after a payoff | The answer landed and people left with it | No — this is the healthy version |
| Gentle constant decline from early on | Low information density, nothing pulling forward | Yes — slow bleed |
| Steep loss in the first 15–30s, then flat | Hook failed, or packaging promised something else | Yes, but it's a packaging problem, not a body problem |
| Drop around the 5–6 minute mark on a padded video | A six-minute idea stretched for runtime | Yes, and the fix is publishing shorter |
The first row is the one people misread. A cliff immediately after a payoff is a completion signal wearing a drop-off costume. The viewer got the thing, confirmed it worked, and left. Every additional second you add before that payoff moves the cliff later and raises your APV without helping a single person.
The clearest public statement of how little early loss means in isolation is in the leaked MrBeast production document, which treats losing roughly 35% of the audience in the first minute — 21 million of 60 million — as an above-average result. That is a first-minute loss most creators would call a catastrophe, being described as a win, because it is being judged against what comparable videos actually do rather than against a hoped-for flat line.
Formats where APV is the wrong target
Not every format should be optimised for duration. Sort your catalogue by what the viewer came to do:
| Format | Viewer's job | Optimise for | Not |
|---|---|---|---|
| Answer-a-question clip | Get one fact, fast | Time-to-answer, saves, returns | APV |
| Tutorial with a defined outcome | Complete a task | Completion of the answer segment, saves | APV |
| Reference / spec walkthrough | Look one thing up | Repeat viewers, chapter entry points | APV |
| Narrative, documentary, vlog | Be taken somewhere | Curve shape and APV together | — |
| Series episode | Finish and get the next one | Completion, then part-2 pickup | APV alone |
| Short-form entertainment | Be entertained | Completion rate and rewatch | APV |
The pattern: the more precisely a viewer can define what they came for, the less APV tells you. A search-shaped video, where one question gets one clean answer, is designed to be abandoned. That is the format working. The one-question, one-clip format exists because the fastest possible answer is the product.
Where APV still earns its place is anything where the whole is the product. If a viewer who leaves at 60% missed the point of the piece, then leaving at 60% is a real loss and duration is a real metric.
What to measure instead
Replacing APV on answer formats means picking a metric before you look at the data, which is the discipline most retention analysis skips.
Completion of the answer segment. Mark the timestamp where the payoff is delivered. Measure the percentage of viewers still present at that moment, not at the end. This is the single highest-value substitution and it costs you one number in a spreadsheet.
Time-to-answer, tracked as a design constraint. If you know the payoff lands at 2:10 in one video and 0:50 in another, and the 0:50 version holds more people through the payoff, you have learned something real about your intros. Track it deliberately rather than discovering it.
Saves and sends. A viewer who leaves with the answer and saves the video is telling you the answer was worth keeping. On short-form, sends are the signal that carries a clip past your follower graph, which is the opposite outcome from a bored exit even though both show as an exit.
Return behaviour. Reference content gets watched twice, weeks apart. That does not show up in a single video's curve at all.
Comparison against your own catalogue. Published retention benchmarks are averages over wildly different content. The useful comparison is the five most similar videos you have already published — same format, same length band. Set that up once in your analytics review and it stops being a per-video decision.
One caveat on short-form: platform completion benchmarks do vary by clip length, with reported floors dropping as clips get longer, and those figures come from tool vendors rather than the platforms. Directionally they are consistent across sources. Do not build a target around one.
The damage APV-chasing does
The reason this matters practically is that optimising APV on an answer format has a predictable set of failure modes, and all of them make the video worse:
- Burying the payoff. Moving the answer later raises the percentage viewed and lowers the number of people who get the answer. You are trading the product for the metric.
- Padding to a runtime. Stretching a six-minute idea to hit ten produces a recognisable drop at the 5–6 minute mark, which then reads as a retention problem and invites more padding. The fix is publishing at true length.
- Manufactured suspense. "I'll show you in a second" repeated four times converts a satisfied exit into an irritated one. Same drop-off, worse brand.
- Killing the format's advantage. The thing that makes an answer clip spread is that it answers fast. Slowing it down removes the reason it gets sent.
If you want the honest test: take your best-performing tutorial, cut the payoff forty seconds earlier, publish, and compare the number of viewers present at the payoff rather than the average. On answer formats, that number usually goes up while APV goes down. Decide in advance which one you actually care about.
Re-cutting to move a payoff earlier is a trim, not a rebuild — cutting a section out on a re-renderable timeline, checked on a 480p preview pass that costs no credits and carries a short per-user cooldown, before a single charged export.
FAQ
Is good abandonment an official YouTube ranking concept?
Not that I can verify. The term is borrowed from information retrieval and creators apply it to video because it describes a real distinction — leaving satisfied versus leaving bored. Platforms do state publicly that they rank on multiple signals rather than duration alone. Treat good abandonment as a diagnostic frame for reading your own curves, not as a documented mechanic.
Should I stop looking at average percentage viewed entirely?
No. It is a fine metric for narrative, documentary and series content, where the whole video is the product and an early exit is a genuine loss. Stop using it as the primary target for answer clips, tutorials with a defined outcome, and reference content. Sort your formats first, then assign metrics.
How do I measure completion of the answer segment?
Mark the timestamp where the payoff is delivered, read the retention graph at that moment, and log the percentage still present. One column in the spreadsheet you already keep. Compare it across videos in the same format rather than against any published benchmark.
Does a sharp drop right after the payoff hurt distribution?
A drop after a payoff and a drop after a tangent look the same on the chart, and you should assume ranking systems see more than the chart — engagement, saves, sends and return behaviour all move differently in the two cases. What you can control is not manufacturing the second kind. Extending runtime to hide the first kind reliably makes the video worse.
The full curve-shape taxonomy, including the shapes that genuinely are failures, is in audience retention analysis for AI creators.