Re-hook cadence: how often to reset attention

A re-hook every 30 seconds is a metronome. Viewers habituate to metronomes. The same reset, on the same interval, becomes part of the furniture, and the retention curve keeps its dip: it just looks slightly more "energetic" on the way down.

The useful cadence is not a round number. It is the cluster of timestamps where your last few videos of this format already lose people. You read those off the graph, plant a reset 10 to 15 seconds earlier, and then check the next curve to see whether the dip got shallower, moved later, or ignored you.

Instinctive resets still happen. They just happen late. You feel a sag at 2:40; the graph says it started at 2:18. The line you added was a eulogy.

What a re-hook is (and what it is not)

A re-hook is a reason to stay that did not exist in the previous 20 seconds. It is not a cut, and it is not volume.

A cut changes the picture. Cut rate is a separate lever; talking-head long-form that sits at a 15–25 second baseline and then bursts is using variance to fight habituation, which is the finding that actually matters in the platform cut-rate numbers. A burst of cuts with the same information is still the same information.

A re-hook changes the question, the scale, or the evidence:

• A new open loop ("the version that actually failed is the next one")
• A visible reveal (the result, the number, the face, the wrecked part)
• A scale shift (wide setup to insert, or the reverse)
• A new speaker, location, or on-screen artifact
• A claim the thumbnail did not already make

The first frame of the video is a hook, not a re-hook. That job is diagnosing the opening and treating the hook as a frame. Re-hooks start after the promise is on the table. Planting one at 0:08 because a blog said "reset often" crowds the opening and usually steepens the first-30-second drop.

Short-form still uses re-hooks. They just have less room, so one well-timed reset beats three. A 45-second clip with a reveal at 0:18, placed because your last six clips bleed from 0:20, is the same method on a shorter clock.

Read the drops, then schedule against them

Do this on the last six to ten uploads of one format. Mixing a tutorial with a vlog will invent a cluster that neither format has.

1. Open each retention graph. Ignore the first 20 seconds for this pass; that regime is packaging and the opening frame.
2. Mark every drop steeper than the surrounding slope. Write the timestamp, not a vibe. "2:14" is usable. "Sometime in the middle" is not.
3. Also mark valleys (dip then partial recovery) and any rewatch spike. Valleys are late re-hooks: the reset arrived, just after people had started leaving. Spikes are already-working moments; do not cover them, clip them.
4. Cluster the timestamps. Put them on one line. If five of eight videos lose a step between 2:05 and 2:35, you have a cadence. If they scatter, you have per-video segment problems, not a cadence problem, and a scheduled reset will not help.
5. Plant the next reset 10–15 seconds before the cluster, not on it. A re-hook on the dip is arriving as people leave. Early enough to be in frame when the habit would have fired; not so early that it collides with the previous block's payoff.
6. Name the reset. "Insert a wow" is not a name. "Show the failed print at 2:10, full-frame, no talking for two seconds" is a name.

A cluster table from a real review looks like this (timestamps rounded):

Upload	Steepest body drop	Valley?	Notes
1	2:18	No	Chapter 1 ended on a summary
2	2:26	Yes, recovered 2:50	Reveal was the recovery
3	4:40	No	Different problem (ad-read)
4	2:09	No	Same chapter seam
5	2:22	Yes, recovered 2:41	Same late reveal
6	2:15	No

The cadence here is ~2:15, not "every two minutes." Uploads 2 and 5 already told you the fix: the reveal works, it is just late. Move it to ~2:00. Upload 3 is a different shape and should not vote.

If you cannot see a cluster after ten videos, stop adding re-hooks on a clock. You do not have a cadence problem. You have either a density problem (slow bleed, no events) or a series of one-off segments. Those get a cut or a rewrite, not a gong every 90 seconds.

Confirm the dip actually flattened

Shipping a reset and feeling better is how you accumulate decorations. You need a before/after that would still be true if you disliked the new beat.

Wait for a comparable video. Same format, same length band, similar CTR. A 3-minute Short cannot confirm a 10-minute change. One video is a hint; three is a result.

Overlay the body, not the opening. Line the graphs up at 0:45 or 1:00, once the first-minute tax is paid. You are asking whether the 2:00–3:00 window got shallower. If the new video's opening is also better, the whole curve sits higher and you will credit the re-hook for work the title did.

Score three outcomes, only three:

Outcome	What the overlay shows	What it means
Flattened	The step at the old cluster is gone or obviously shallower	The reset was the right medicine
Moved	The step is later, similar size	You bought time. Plant the next reset before the new cluster; do not celebrate yet
Ignored	Same step, same time	That beat is not a re-hook problem. Watch the ten seconds before the drop: tangent, recap, or a chapter that ended too cleanly

A moved dip is the one people misread as a win. The video's average percentage viewed may even rise, because people stayed through the old danger zone. Then they leave at 3:10 instead. Log it as "moved to 3:10" and treat 3:10 as the new cluster. That is the cadence updating, which is the point.

Rewatch spikes after a reset are extra confirmation. If the line bumps above the surrounding curve at the new beat, you did not just stop an exit; you made something replayable. That beat is your next Short, not only a retention patch.

When the reset is a visual you do not already have, generate that insert as its own clip and drop it on the timeline. A b-roll shot of the object, the wrecked part, or the number card is a legitimate re-hook if it is new information. It is not a legitimate re-hook if it is coverage of the sentence you were already saying. On the editor, check the join with the free 480p preview pass (it carries a short per-user cooldown; the final export is charged once) rather than guessing from the script.

A hook pack before the body exists is the right tool for minute zero. It will not place a 2:10 reveal for you.

A cadence you can reuse without freezing it

Once a cluster is stable, write it into the outline as a named beat, not as "add energy." Example for that 2:15 channel:

• 1:55 — finish the first example on an unfinished question
• 2:00–2:10 — full-frame failed result, two seconds of silence or near-silence
• 2:10 — new question that Block 3 must answer

Run that until the overlay stops flattening, or until the cluster moves. Then update the outline. Cadence is a living measurement of where this format leaks. It is not a brand asset. Channels that keep a "reset every 90 seconds" rule for a year are servicing the rule.

Two limits, so this does not eat the video:

• Do not re-hook on top of a payoff. If 2:00 is already the moment the first example lands, the reset is the payoff. Adding a second trick beside it splits attention and can create the step you were trying to prevent.
• Do not re-hook in the last 10 percent unless the job is a loop into the next video. Late resets that introduce a new topic read as a new video starting, and people leave rather than restart.

FAQ

Is there a universal interval I should start from?

No. There is a universal method: cluster your own body drops, plant 10–15 seconds earlier, overlay. Round intervals (30 seconds, 60, two minutes) are what you use only when you have fewer than six comparable graphs, and you should treat them as a guess you are trying to retire.

How is a re-hook different from a faster cut?

A faster cut changes picture rate. A re-hook changes the reason to stay. You can have both at once (a five-cut burst that also reveals the result). You can also have a 20-second locked shot that re-hooks because the sentence is a new claim. If you only accelerate the cutting, habituation just happens at a higher tempo.

How many videos before I believe the overlay?

Three in the same format and length band, after the change. One can be distribution luck. If the third still shows "ignored," stop adding ornaments at that timestamp and inspect the ten seconds before the drop for a tangent or a finished-sounding summary.

Should the reset sit on the dip or before it?

Before it. Ten to fifteen seconds earlier is on screen when the habit would have fired, and late enough not to collide with the previous payoff. If the cluster is a 40-second smear, plant at the start of it, not the middle.