Hold rate is not completion rate

Hold rate is not completion rate. Hold rate is the share of people still watching after the hook window — the number that grades the body, once the opening has already done its job. Completion rate is the share who reached the last frame. One is a slope. The other is a point. A clip can complete because it was six seconds, not because the body held. A clip can hold for forty seconds and still miss a 60-second complete. Fixing the wrong one is how a team recasts a fine open and leaves a sagging middle.

Split the curve at the hook

Hook rate asks who is still there a few seconds in. Hold rate asks who is still there once that audition is over. Completion is what's left standing at the end. The retention curve is the picture that contains all three. Read it in two spans.

A cliff at second two is a hook problem. The first frame, the first burned line, the first sound. Swipe-away is that event on short-form. After 24 August 2026 a YouTube Short swiped at second one can still write a public view. Engaged views will not follow. Recut the open. Do not touch the middle yet.

A slow leak from second five to the end is a hold problem: pacing, a claim that did not pay off, a talking head that never changes the picture, a pattern the feed has already seen. A pattern interrupt — a new shot, a new claim, a new number — is the usual fix, not a louder voiceover. Completing the file by cutting it in half will raise completion rate and can still leave the leak in whatever middle remains.

Completion rate is a scoreboard number. Useful for comparing two cuts of the same idea at a glance. It says nothing about where the ones who left, left. On short-form it tends to correlate with further distribution more than likes do, which is why unearned runtime is expensive. A good video that runs long loses completions it would have kept at a tighter length. That is not automatically a hold win if you cut the proof beat to chase the point.

Why the two numbers lie about each other

A six-second joke with a 90% complete and a dead rewatch rate did not hold. It ended. Compare inside one length band. A 12-second clip and a 45-second clip are not on the same completion scale. Hold rate, read as the slope after the hook, is the fairer compare across nearby lengths. Still do not compare a Reel to a ten-minute watch.

A long-form explainer with a weak complete and a healthy average view duration may have held the people who came for the lesson and lost the people who came for the title. Completion is punishing the last chapter. Hold in the first third might be fine. The fix is the last act, or a tighter promise in the title, not a new hook pack.

YouTube public views after the first-frame cutover make completion look worse if you put public views in the denominator. Studio's completion and AVD sit on the engaged side. Homemade "completes divided by public views" will sag because you included the swipes. Use the platform's engaged pile.

Rewatch can push a retention curve past 100%. That is loops, not a 100% complete of unique people. Completion rate as "who reached the last frame at least once" is still a first-pass statistic. Loops are extra. Do not report a 140% complete as if 40% more humans finished.

Likes will not diagnose a hold leak. Likes write after a stay. The leak is people who did not stay long enough to tap. Read the slope.

What to change for each miss

Hook miss. First-second picture, burned claim, mute test. Hook pack against one body. Do not rewrite the body.

Hold miss. Cut idle. Interrupt the pattern. Pay off the claim you opened. If the middle is a list, put the best item earlier. If the middle is a talking head, cut away. If the middle is a generate that jumped time, regenerate the shot — a model jump cut is not an editorial jump cut. Cadence of posts will not fix a hold leak inside one file.

Completion miss with a flat hold. The file is too long for the job, or the last seconds are a logo hold, a fake end screen, a second ask. End. On long-form, leave a layout for the end screen instead of talking into it. On looping short-form, join last frame to first.

Completion win with a hold leak. You shortened until everyone who survived the hook also survived the file. Distribution may still be weak if the body delivered nothing. Completion measures finishing, not value. A hollow six-second complete is a skip with extra steps on the next post.

Captions affect all three. Unreadable type in the open is a hook miss. Type covering the pack in the middle is a hold miss. Type fighting an end screen is a last-second miss. Burned-in captions on mute-first; sidecar on the watch page. Recheck after every length change. A 40-second cut with 40-second cues will not time to a 22-second recut.

Pre-commit which number the slot is for

Reach slots: hook rate, then hold. Completion is a tie-break inside one length.

Evergreen how-to: hold and saves. A complete that comes from a short file with no steps is a failed lesson.

Ads: three-second hold and the conversion event. Completion on a 15-second ad is a view-through statistic, not a synonym for hold.

Write the metric on the brief before either variant goes up. Then generate. The AI video generator will give you another take. It will not tell you which span of the curve moved. Put the curve next to the render.

Hold rate is not completion rate. Slope versus last frame. Diagnose, then score.

Read the middle of the curve before you recut the ends, in Versely.

FAQ

If completion is high, is hold automatically fine?

No. A short file completes. Look at the slope after the hook. A leak that never had time to show will appear when you add a proof beat.

Should I fix hold by rewriting the hook?

Not if hook rate is already fine. The people you are losing already stayed for the opening. The body lost them. Interrupt the middle.

Can I compare completion across a Reel and a long-form?

No. Different lengths, different players, different denominators after YouTube's first-frame split. Compare inside a length band, or read hold as a slope, or go back to the curve.