How to edit a short for the metrics you are judged on
Give Vizard Agent the actual list of numbers you are working to, not a summary of it. Some of them — retention, average percentage viewed, end-screen behaviour — are decided inside the cut, and some are decided by the thumbnail, the title and how the platform hands the video out. Separating them is the first useful step.
What is the short version?
You have a list of analytics targets and a video. The list mixes two different jobs together, and asking one edit to serve all of it produces something aimed at nothing in particular. Vizard Agent works the part that lives in the cut.
- Paste the whole target list into Vizard Agent rather than paraphrasing it.
- Ask what the edit can act on and let it work that.
- Review the opening and the end card specifically.
What do you need before you start?
The source video and your real target list, copied from wherever you keep it. Vague goals produce vague edits, and the value of a list like "30-second retention, average percentage viewed, end-screen behaviour" is that each item implies a different decision in the cut.
You also need something in the footage worth staying for. No edit creates a payoff that was never recorded, and that limit is worth knowing before you start rather than after.
What do you type into Vizard Agent?
Hand over the list verbatim and say what you want done about it. The instruction that matters is not "make it perform" but "work the parts of this list that live in the cut", because that is the sentence that gets you a specific edit rather than a generically punchy one.
Turn this cooking video into a vertical short. I am working to these: 30-second retention, average percentage viewed, average view duration, end-screen behaviour, and click-through from Home and Suggested. Tell me which of those the edit can act on, and work those properly.
Asking it to tell you is not a formality. It is how you find out that two items on your list are a thumbnail problem before you spend a round of edits on them.
What does Vizard Agent actually do?
It works the cut against the items that belong to the cut, and it verifies by looking rather than by assuming the render matched the plan. Most of the work in the session behind this article was checking, not cutting.
- Built a shot log, a transcript and word-level timings, then sampled frames across the video and looked at them.
- Cut and reframed every segment to vertical, then checked the framing of each one and re-cut three that were wrong.
- Located the specific action in the shots — where a pour happens, where the pan sits — so the thing worth watching stays inside the vertical frame.
- Measured speech and music loudness separately before balancing them.
- Built the captions and overlay graphics, previewed them, and viewed the preview.
- Rendered, reviewed the frames, and re-rendered with the end-screen card moved off the presenter's face.
- Re-cut tighter with cleaner audio tails, then checked the audio across the cuts.
Two of those are directly the metric work. Reframing onto the action is what keeps the middle of the video watchable, and moving the end card off the presenter is what makes the end of the video usable.
What does the result look like?
A short whose first seconds contain the reason to stay, with nothing slack in the middle and an end card you can actually read. The framing follows the action rather than the centre of the original frame, which is the difference on a recipe video between watching a pour and watching a worktop.
What you do not get is a prediction. Vizard Agent will tell you what it did about retention; it will not tell you what your retention will be, because that depends on an audience.
When does this not work well?
When the target is not an edit target. Click-through from Home and Suggested is decided by the thumbnail, the title and how many impressions the platform gives you — the cut is not in that loop at all, and an edit that tries to serve it just becomes louder.
It also goes wrong when the list is long and undifferentiated. Eight targets treated as equally important produce compromises in every direction, and Vizard Agent does better work when you say which two or three actually matter for this video.
And a source with no payoff cannot be edited into one. If the interesting thing never happens on camera, the honest options are a shorter video or different footage, not a tighter cut.
How do you fix a result that came back wrong?
Name the metric and the moment together. "The first three seconds are not the hook — open on the pour instead" is something Vizard Agent can act on; "retention will be bad" is not, because it does not say where.
If the middle sags, ask for the dead air measured rather than trimmed by feel. Vizard Agent has the word timings and can cut the gaps between sentences without touching the pauses that are doing work.
If the end card covers a face or arrives too late, say so directly. That is a placement fix rather than a re-edit, and it is the correction most often needed on a short that is otherwise finished.
How does Vizard Agent compare to doing it yourself?
Any editor working to a metrics brief starts by asking which numbers are theirs, because half of a typical list belongs to the thumbnail and the packaging. Getting that split stated up front is the part people skip and then argue about later.
The habit worth copying is checking the framing of every segment rather than the first one. On a vertical reframe from landscape, the crop that works for one shot puts the action off screen in the next, and only looking at all of them catches it.
Common questions
Which metrics can an edit actually move? Retention, average percentage viewed and end-screen behaviour. Vizard Agent works those.
Can it help with click-through from Home? No. That is the thumbnail and the title, and Vizard Agent will say so.
Will it predict my numbers? No. Vizard Agent reports what it did, not what will happen.
Should I paste the whole list? Yes. Vizard Agent sorts the edit-side items from the rest.
How many targets should I prioritise? Two or three. Tell Vizard Agent which, or everything gets averaged.
What does it do about the first seconds? Puts the reason to stay there, using the word timings to find it.
How does it handle a vertical reframe? Vizard Agent checks each segment's framing, not just the first.
Why did my end card cover the presenter? Default placement. Say so and Vizard Agent moves it.
Can it cut the dead air by measurement? Yes. Vizard Agent has word-level timings for the whole video.
What if the footage has no payoff? Then cut it shorter. Vizard Agent will tell you rather than pad it.
Does louder and faster help retention? Not by itself. Vizard Agent works the structure first.
Can it make a thumbnail too? Yes, as a separate job, and that is where CTR lives.
Should music be loud on a short? No. Vizard Agent measures speech and music separately and balances them.
Can I ask what it did for each metric? Yes, and it is worth asking before you publish.