How to review your own edit before you publish
The edit is the last moment a retention problem is cheap. After publishing, the same problem costs you the video and teaches the algorithm something you did not want it to learn. Yet most creators review their own cut by watching it back once and deciding whether they like it, which is the one method almost guaranteed not to work.
Updated September 2026 · By the Retti team
Why watching it back does not work
You cannot un-know your own video. When you play back your cut you supply the intent, the context and the payoff from memory, and the version you experience is not the version a stranger gets. You already know why the tangent at 4:20 is there. You already know the graphic is about to explain the point. A first-time viewer has neither, and they are the only person whose experience is being measured.
This is the same problem that produces weak hooks, and it has the same shape: the work was reviewed by the one person on earth guaranteed to understand it.
The things that cost you viewers are mostly not in the words
Most self-review — and most AI review — happens on the transcript, because text is easy to work with. But a transcript is a record of what was said, and a large share of the reasons people leave are not things anyone said.
A graphic that lands after the point it explains
You make a claim; the chart supporting it appears nine seconds later. In the edit this feels fine, because you know the chart is coming. To a viewer, the claim arrived unsupported and the chart arrived unprompted. Nothing in the transcript shows this, because the transcript has no idea the chart exists.
Silence that reads as dead rather than tense
A held pause before a reveal and a gap where nothing is happening look identical on paper — both are the absence of words. On screen they are opposites. One is the most engaging second in the video; the other is where people leave.
Music that announces the ending early
A track fading out, a shift to a slower bed, a return to the theme you opened with — viewers read all of these as "this is wrapping up" and act accordingly. If that happens two minutes before your actual ending, you have told a large share of the audience it is safe to leave while there is still video left.
Visual monotony
The same framing held for ninety seconds costs attention even when the content is strong. Nothing is wrong at any single moment, which is exactly why it survives self-review — and why it shows up as a long sag rather than a cliff.
A pass you can run yourself
- The mute pass. Watch it with no audio. Note every stretch where the picture does not change, and every graphic that arrives after the moment it belongs to.
- The audio-only pass. Turn away from the screen. Note the silences that feel like nothing is happening, and the exact moment the music first suggests the video is ending.
- The double-speed pass. Play the whole thing at 2x. Pacing problems are much easier to feel when compressed, and sections that sag stop being tolerable.
- The cold-open test. Play the first fifteen seconds and stop. If someone hearing only that could not say what the video is about and what they are waiting to find out, fix that before anything else — it is the part every viewer sees.
Three passes on a ten-minute video costs about fifteen minutes and will find most of what a first viewer would. Our guide to why viewers stop watching covers what to do with each kind of problem once you have found it.
Checking it against the curve afterwards
Once the video is live, the retention graph is the unbiased review you could not give yourself. Line up the three steepest deviations against what was on screen at those seconds, and the cause is usually obvious in hindsight. The value of doing the pre-publish pass first is that you find out which of your own judgements were right — that is what makes the next review better rather than just the next video.
Our guides to retention curve shapes and reading a retention graph cover how to read the shape, and how long an intro should be deals with the specific case where the opening runs long.
Where Retti fits
Edit Review does this pass for you, and it does it by actually watching the cut and listening to it — the cuts, the graphics and captions as they land, the b-roll, the music changes, the silences and the level shifts, not a transcript of the words. The rhythm is measured from the file rather than estimated, so a pause that is tense reads differently from a pause that is dead, and a graphic that arrives late is flagged at the second it arrives.
What comes back is timestamped notes flagged by severity, on an unlisted cut, before anyone has seen it. Upstream of that, Script Lab catches the structural causes before filming; downstream, Retention Lab reads the real curve once it is live. First analysis is free.
Frequently asked questions
How do I get feedback on my video before publishing it?
Upload it to YouTube as unlisted — the private-draft approach most channels already use — and review it there, or share that link with someone whose judgement you trust. Reviewing it yourself works far better if you do it in separate passes: once on mute, once audio-only, once at double speed. Each pass removes the context your memory keeps supplying, which is what makes ordinary playback such a poor review.
Why can I never spot problems in my own edit?
Because you cannot un-know your own video. You supply the intent, the context and the payoff from memory, so the version you experience is not the version a first-time viewer gets. You know the graphic is coming and that the tangent pays off later; they know neither. Separating the passes — mute, audio-only, double speed — strips out some of that knowledge.
Can AI actually watch my video, or does it just read the transcript?
Both exist, and the difference matters. Transcript-only analysis can only work on what was said, so it cannot see a graphic arriving late, tell a tense silence from a dead one, or notice that the music signalled the ending early. Retti’s Edit Review watches the picture and listens to the audio, with the cuts and levels measured from the file rather than inferred from text.
What should I look for when reviewing my own cut?
Four things that rarely show up in the script: graphics that arrive after the point they explain, silences that read as dead rather than deliberate, music that signals the ending before the video actually ends, and stretches where the framing does not change. All four cost viewers, and none of them are visible in a transcript.
Is it worth reviewing an edit if I already checked the script?
Yes, because they fail differently. A script review catches structure — a buried promise, a section that plateaus, a payoff that arrives too late. An edit review catches everything that only exists once the video is assembled: timing, pacing, the relationship between what is said and what is on screen. A strong script can still be undone in the cut.
How long before publishing should I review the edit?
Early enough that you are still willing to change it. The practical test is whether a finding would actually get fixed: if you are reviewing an hour before a scheduled upload, you will rationalise problems rather than cut them. A day’s gap is usually enough to make the review honest and the fixes cheap.