Creator

Retention Analysis reads the hold

Score a published video on whether the script can keep a viewer watching — opening, pacing, curiosity, payoff, and authority. Six credits. Compare mode included.

What Retention Analysis does

Retention Analysis is a Content Spy scored audit. You pick a published video from your library. The model reads title, description, tags, transcript, duration, and metadata — then scores how well the video is built to hold attention after the click. It is not YouTube Analytics. It does not plot a retention curve, invent drop-off timestamps, or tell you the percentage of viewers still watching at 0:47.

Use this when the video is already live and you want a structural read on watch-time design. If you are still writing the script, use Retention Strength in Content Lab instead. Same five pillars. Different stage.

How to run it

  1. 1Open Content Spy and select Retention Analysis — or start Compare mode and add this tool to the run.
  2. 2Choose a published video from your library. There is no extra form. Duration decides the format: 180 seconds or less is scored as Short-form. Longer than 180 seconds is scored as Long-form.
  3. 3Confirm the run. It costs 6 credits — the same in a single analysis or in a comparison.
  4. 4Read the Retention Strength Score, the five pillars, and the lists. The score is a hold-potential verdict, not a replay of your analytics graph.
Short-form here is not YouTube Shorts. A 2:40 video is still Short-form in this tool. A 3:01 video is Long-form. That 180-second line is the only format switch.

What the model actually reads

No thumbnail vision. No analytics CSV. The audit is built from the video record and a long transcript window.

  • Title, description, and tags — used as context for the promise the video made before play.
  • Transcript — first 8,000 characters. That is a much longer window than CTR Audit, because hold lives in the body of the script, not just the first hook line.
  • Duration — converted to minutes and seconds, then classified as Short-form (≤ 180s) or Long-form (> 180s).
  • Language, category, and channel name when they exist on the record.

What is not in the payload

  • Thumbnail image — this tool does not run vision. Packaging is CTR Audit’s job.
  • YouTube retention graphs, average view duration, or audience retention percentages.
  • A draft form. Unlike Retention Prediction, you cannot type a script here. The video has to be published and in the library.
If the transcript is missing, the model is scoring a video with almost no spoken evidence. The audit will still run, but the body-of-script pillars will be weaker than they should be. Transcribe first.

The five equal pillars

Five scores, 0–100 each, equal weight. The overall Retention Strength Score is the rounded average of the five. No pillar is worth more than another.

Short-form and long-form share the same five names. The bar inside each pillar changes. A 12-second cold open that works on a 90-second video is a problem on an 18-minute one.

Opening

20%
  • Does the first stretch earn the next ten seconds? Short-form wants an immediate visual or verbal reason to stay. Long-form can take a slightly longer setup — but it still has to pay rent in the first stretch, not after a two-minute intro bumper.
  • A delayed open, a logo sting, or “hey guys welcome back” with no stake will cut this score even if the rest of the video is strong.

Pacing

20%
  • Are beats arriving at the right density for the format? Short-form punishes dead air and repeated setups. Long-form punishes a slog — and also punishes a cut every two seconds with no rest.
  • This is rhythm in the script, not a stopwatch of your editor’s timeline. The model infers it from how the transcript moves.

Curiosity

20%
  • Is there an open loop that makes the next section feel necessary? A question, a promised result, a contradiction, a “wait until you see.”
  • Curiosity that never tightens — mystery with no destination — scores lower than a loop that clearly has a landing.

Payoff

20%
  • Does the video deliver what the open (and the title) promised? A strong open with a soft landing is a hold problem, not just a packaging problem.
  • Short-form payoff can be a single punch. Long-form payoff usually needs a sequence: proof, then the turn, then the close.

Authority

20%
  • Does the speaker sound like they have earned the right to this topic — through proof, specificity, or lived detail?
  • Generic expertise language (“studies show”, “a lot of people don’t know”) without a concrete receipt is a common cut here.

What you get back

Same report shape as the other Content Spy scored audits. The numbers mean hold, not click.

  • Retention Strength Score — 0–100, rounded average of the five pillars.
  • Score Breakdown — Opening, Pacing, Curiosity, Payoff, Authority, each 0–100.
  • Strengths, Weaknesses, Risks, and Improvements — four lists. Improvements are the ones you can act on in the next cut or the next script.
  • Compare mode — same five axes against another video in the same run.
A 73 here does not mean 73% of viewers stayed to the end. It means the script’s hold design averaged 73 across the five pillars. Your analytics graph can disagree. Treat the disagreement as a question, not a bug.

Case study

An 8-minute “I rebuilt my entire workflow” video. Long-form (over 180 seconds). Strong curiosity and a clean open. Pacing sags in the middle when the rebuild turns into a settings tour.

Example output — fictional scores, real report shape

Retention Strength Score

0–100

73avg
  • Opening78/100
  • Pacing64/100
  • Curiosity81/100
  • Payoff72/100
  • Authority70/100
Opening78
Pacing64
Curiosity81
Payoff72
Authority70

Strengths

  • The open names the old workflow and the cost of keeping it — a stake in the first twenty seconds.
  • Curiosity stays alive: each section teases the next tool in the stack instead of dumping the whole rebuild at once.

Weaknesses

  • The middle third is a settings walkthrough with no new question. Pacing is the lowest pillar.
  • The payoff is the new stack on screen, but the video never returns to the cost named in the open.

Risks

  • Viewers who came for the rebuild may leave during the settings tour and never see the finished stack.
  • Authority is competent but generic — “this is what I use now” without a before/after receipt.

Improvements

  • Cut or chapter the settings tour. Keep one surprising default. Move the rest to a pinned comment.
  • Close by returning to the cost from the open: time saved, or the thing you no longer do.
  • Add one concrete receipt — a number, a before clip, a failed attempt — so authority is earned, not claimed.

FAQ

Is this my YouTube retention graph?

No. YouTube Analytics shows what happened. Retention Analysis scores whether the script is built to hold. It does not invent drop-off points or average view duration.

How is this different from Retention Strength in Content Lab?

Same five pillars, different stage. Retention Strength scores a draft you type in the Lab. Retention Analysis scores a published video from your library. Lab is pre-publish. Spy is post-publish.

How is this different from Retention Prediction?

Prediction is the unpublished form: you type a hook, title, and script before the video exists. Analysis is the published-video audit. Use Prediction to pressure-test a draft. Use Analysis on something already live.

What counts as Short-form?

Duration of 180 seconds or less. That is this tool’s line, not YouTube Shorts (60 seconds). A 2:40 video is Short-form here. A 3:01 video is Long-form.

Does it look at the thumbnail?

No. No vision pass. Packaging is CTR Audit. This tool is the hold after the click.

Why is the transcript window so long?

Hold lives in the body. CTR Audit only needs the first hook. Retention Analysis reads up to 8,000 characters of transcript so pacing, curiosity, and payoff have evidence to score.

What if there is no transcript?

The run still completes, but Opening, Pacing, Curiosity, and Payoff are scoring almost without spoken evidence. Transcribe the video first if you want those pillars to mean something.

Can I compare two videos?

Yes. Compare mode is on. Same cost: 6 credits per video in the run.


Next steps

6 credits · Content Spy