Average view duration is the metric YouTube weights most heavily and the one most creators only look at after publishing, when it is too late to change the video. The Watch Time Optimizer predicts a retention curve from your planned video structure, identifies the three points where viewers are most likely to leave, gives specific script edits for each, and tells you the video length your niche actually sustains.
This guide explains how the prediction works, how to interpret the drop-off zones, and the length question every creator gets wrong.
Predicting retention before filming
Retention curves in a niche follow recognisable shapes: a first-30-second cliff, a mid-video sag where the setup ends and the payoff has not arrived, and an outro collapse. The model maps your described structure — intro, sections, reveals, CTA — against those shapes and against the duration band, and produces a curve with the sag points marked. It is a prediction, not analytics, but it is right about where the risks are far more often than not.
The three drop-off zones
Each zone comes with a cause and an edit. Typical causes: intro longer than 20 seconds (cut to the promise), a section that explains before it shows (reorder), a payoff at the end that could be teased at the midpoint (add a preview), an ad read or sponsor segment in the middle third (move to 60–70%), a CTA before the content is delivered (move after the first payoff). The projected improvement quantifies what the fixes are worth in average view duration.
How long should the video be?
Long enough to cover the topic, short enough that the retention percentage holds above the niche median. The report gives a range for your niche and topic type. Most tutorials sustain 8–12 minutes; most commentary sustains 12–20; most product comparisons sustain 9–14. A 25-minute video at 30% retention delivers less watch time and a worse signal than a 12-minute video at 55%.
Tips
- Describe the structure in order with rough timings. "Intro 1m, unboxing 3m, test 5m, verdict 2m" is enough.
- State where the biggest surprise or payoff is. The model's best edit is usually to move it earlier or tease it.
- Run it again after the edit with actual timings; the second pass catches a sag you introduced while cutting.
Common mistakes
- Padding to hit 8 or 10 minutes — Mid-roll eligibility is not worth a retention collapse. The extra ad slot earns less than the lost reach.
- Sponsor segment in the first quarter — It is the highest-drop-off placement possible.
- Trusting the curve as fact — It is a structural prediction. Check the real curve in Studio at 48 hours and compare.
Worked example
Input: "14-minute budget grinder comparison: intro 90s, unboxing 3m, grind tests 6m, taste test 2m, verdict 1.5m; audience beginners."
What the report returned: Predicted AVD 46%. Zones: 0:00–1:30 intro (cut to 25s), 4:30–7:00 grind tests before any result (show the first taste result at 4:45), 12:30+ verdict (state winner at 12:35, then justify). Recommended length 10–11 min. Projected AVD 56%.
Frequently asked questions
Can it analyse a published video?
Yes — paste the URL and the model uses the actual duration, chapters and description. It still cannot see your private retention graph, so pair it with Studio data.
How accurate is the predicted curve?
It is reliable about where the drop-off risks are; the exact percentages are estimates. Creators typically find two of the three zones match their real curve.
Does it help with Shorts?
Shorts retention works differently (loop-based). Use the Hook & Loop Generator for Shorts.
What AVD percentage is good?
Above 50% for videos under 10 minutes and above 40% for 10–20 minutes is strong in most niches. The report gives your niche's median.