AVD is not the same as watch time
AVD is mean minutes per view. Watch time is the total. Raising the mean by cutting the file can still destroy hours.
AVD is mean minutes per view. Watch time is the total. Raising the mean by cutting the file can still destroy hours.
Read retention against caption changes: isolate one variable, compare first-3s hold, then recaption. This is the measurement loop.
Platform auto-captions arrive too late, in a font you do not own. Burn Versely captions in before you post: they do the hook, the silent watch, and the brand lock — and they only work if you pick one family and stop.
The in-app clips are one line, ten styles. Watch them muted. That is the test, not the picker thumbnail.
Evenly spaced small drops mean your transitions are working as exits. The bridge pattern that carries a question across the seam, and how to measure the fix.
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.
A slow bleed is the one retention shape you can diagnose from the script alone. How to score claims per minute, find your band, and edit toward it.
Place re-hooks on your own drop-off timestamps, not on a round interval. Graph-driven placement, then an overlay that shows whether the dip actually flattened.
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.
YouTube Test & Compare crowns the variant with the highest watch time, not CTR. Why a winner can lose the click race and why Studio CTR is the wrong comparison.
Click-through and retention trade off. Combine them into watch time per impression, which YouTube's thumbnail test accumulates when it picks a winner.
Audience retention analysis for AI creators: read the four curve shapes, diagnose intro cliffs, dips, and spikes, and fix each one with targeted regeneration.