Packaging problem or content problem?
Click-through and retention form a two-by-two that tells you whether to repackage or re-edit. Most creators re-edit videos that only needed a new title.
A video underperforms. The creator watches it back, decides the middle drags, and spends an afternoon recutting it. The video was fine. Nobody clicked it.
This is the most common wasted afternoon in the job, and it happens because underperformance is felt as a single sensation and diagnosed with a single number. Two numbers read together resolve it in about thirty seconds, and they point at completely different departments.
The two numbers
Click-through rate measures the decision made before playback. It is a verdict on the title, the thumbnail, and the fit between them and the surface the video appeared on. It says nothing about the video.
Retention measures the decisions made during playback. It is a verdict on the video and on whether the video honoured what the packaging implied. It says nothing about the click.
Neither is interpretable alone. A low CTR with excellent retention is a good video nobody found. A high CTR with poor retention is a promise the video did not keep. These need opposite responses, and averaging them into "it flopped" destroys the only information you had.
Read both against your own channel median rather than against a published benchmark. CTR is heavily dependent on surface, niche and impression mix; a number that is strong on one channel is weak on another. Your last ten uploads are the only fair comparison set you have.
The quadrant
| Retention low | Retention high | |
|---|---|---|
| CTR high | Overpromise. The packaging wrote a cheque the video didn't cash. Re-edit or realign the claim. | Working. Turn it into a format, not a one-off. |
| CTR low | Wrong idea, or wrong audience. Neither fix helps — change the topic. | Discovery failure. Repackage. Do not touch the video. |
The bottom-right cell is the one creators habitually get wrong. A video with strong retention and weak CTR is, by definition, a video that satisfied everyone who watched it. Recutting it is destroying the only asset in the picture. The correct response is a new title and a new thumbnail on the existing upload.
The top-left cell is the mirror image, and it is the harder one to accept, because a high CTR feels like a win. It is a win at the impression stage and a debt at the playback stage. MrBeast's leaked internal production document assigns minute one exactly one job: confirm the promise the thumbnail made. When CTR is high and retention collapses in the first minute, minute one is not doing that job — either because the payoff is buried, or because the packaging promised something the video never contained.
The bottom-left cell is the honest bad news. Low on both means the idea did not land and the video did not save it. There is no packaging fix and no editing fix for an idea nobody wanted. Log it, look for the pattern across the catalogue, and move on.
Which fix, in which order
Repackage (CTR low, retention high). Change one element at a time so you learn something. On YouTube this is a native experiment; elsewhere it is a manual swap and a patient wait. What to change, in order of usual impact: the visual concept of the thumbnail, then the title's specificity, then the thumbnail's execution. Reported lifts in field write-ups cluster somewhere in the 15 to 40% range on CTR, with occasional outlier case studies well above that. Treat those as directional priors from vendor-analysed data rather than as measurements — reproduce the effect on your own channel before you design around it. The archetypes that keep working are covered in thumbnails that earn the click, and generating three candidates per upload rather than one is a thumbnail generator task, not an afternoon in a design tool.
Re-edit or realign (CTR high, retention low). Two sub-cases, and they are separated by where the drop is.
If the drop is in the first 30 seconds, the packaging and the opening disagree. The cheaper fix is usually to move the payoff earlier so minute one confirms the promise. The weak-opening diagnosis routine joins the drop timestamp to what was actually on screen at that second, which turns "the intro is weak" into a specific edit.
If the drop is in the body, the packaging is fine and a segment is at fault. Name the curve shape first — the retention analysis breakdown covers the standard shapes — and cut or replace the segment rather than tightening globally. On an EDL-based timeline that is an edit to the instruction list rather than a rebuild, and a free 480p preview pass lets you check the join before committing. That pass carries a short per-user cooldown, so it is for verifying a specific change; the final export is charged once regardless of how many clips the timeline holds.
Systematise (both high). The failure here is treating a hit as luck. Write down what the packaging promised, what the structure was, and where the payoffs landed, then run the same shape again with different content. Paddy Galloway's framing for this is to milk a format while it still delivers rather than abandoning it out of boredom.
Why the wrong fix is the default
Because editing feels like work and packaging feels like decoration.
The most useful number in Galloway's public work is a time-allocation one: top creators spend roughly 30% of their effort on ideation and packaging, while smaller creators spend around 5%, with the remaining 95% going into filming and editing. That ratio explains the misdiagnosis mechanically. If 95% of your effort sits in the edit, the edit is where you go looking when something breaks — not because the evidence points there, but because that is the room you are standing in.
The correction is not "spend less time editing". It is to make the diagnosis before the fix, every time, in the order given above.
Running the repackage test on YouTube
YouTube's native title and thumbnail test has specific mechanics, and several of them surprise people. From YouTube's own documentation:
- You can run up to three title and thumbnail variants.
- The winner is chosen by highest watch time, not click-through rate.
- Tests complete within two weeks, and can resolve in days at high impression volume.
- Shorts are not eligible, nor are Made for Kids, private, mature or age-restricted videos. Applying an age restriction to a video halts a test already running. Scheduled livestreams and Premieres are ineligible until they convert.
- If there is no clear winner — meaning no strong statistical difference — the first-uploaded variant becomes the default.
The watch-time criterion is the part with real consequences. A variant can win the test while showing a lower CTR than a competitor, because it brought in an audience that stayed. That is the intended behaviour, and it means you cannot line up Test & Compare results against your CTR dashboard and expect the story to match. If you are optimising for clicks and the tool is optimising for watch time, you will spend months confused about your own data.
On sample size, a rough working figure that circulates is around 2,000 impressions per variant to detect a couple of points of CTR difference. The arithmetic is reasonable but the number is not from a published study, so treat it as a floor for taking a result seriously rather than as a stopping rule. The stopping rule on YouTube is simpler: let the full window run. Peeking is the dominant failure mode in every creative test, and the A/B protocol worth copying exists mostly to stop it.
Off YouTube, where there is no native test
TikTok, Reels and Shorts have no equivalent tool, so a packaging test becomes a posting-schedule problem. The workable version:
- Change one packaging element — cover frame or opening text card — and nothing else.
- Post variants at the same slot on consecutive days, never simultaneously. Simultaneous posts cannibalise the same audience pool.
- Pick the metric before you look. For a cover or card test it is swipe-away rate or average watch time at a fixed 24-hour checkpoint.
- Run three or more pairs before drawing a conclusion. One pair is dominated by distribution variance.
The quadrant still applies; only the instrument changes. Swipe-away rate substitutes for CTR, and retention past the first eight seconds substitutes for average view duration. Keeping those two columns side by side in a standing analytics review is what makes the diagnosis a thirty-second read rather than an investigation.
FAQ
Can I repackage a video that has already been live for months?
Yes, and it is usually the highest-return hour available on an established channel. An older video with strong retention has already proven the content works; it has just been carrying a title that stopped earning impressions. Change the packaging and let the recommendation system re-sample it.
What counts as "high" CTR?
Only your own median counts. CTR varies enormously by surface, subject and how much of your impression volume comes from subscribers versus browse. A video sitting a point above your channel's usual figure is high for you, and that is the only comparison the quadrant needs.
If a video is high CTR and low retention, should I delete or unlist it?
Rarely worth it. Repair the claim instead: retitle to describe what the video actually delivers and, if the drop is early, move the payoff forward. Deleting removes the impression history that would have told you whether the repair worked.
Does the quadrant work for ads as well as organic?
The structure does, though the labels change. Thumb-stop rate takes the place of CTR and hold rate takes the place of retention, and the same rule applies: a creative that stops the scroll and loses the viewer needs a different fix from one that nobody stopped for at all.