---
name: Video Retro
category: content
works_with: [claude, chatgpt, gemini]
difficulty: plug-and-play
version: 1.0
source: adapted (viral AI-video guide, rewritten; built out from a step the source only sketched)
tested: no
---

# Video Retro

## What it does
Feeds a published video's real numbers back into the next one: where viewers left, what the drop-off points have in common, and the specific changes for the next production. The feedback loop the "post and hope" crowd skips.

## When to use it
- 48–72 hours after publishing, when retention data exists
- A video underperformed and you're about to guess why
- Monthly, across everything published, to find patterns one video can't show

## The skill
```
Act as a video performance analyst. Rules: distinguish what the
data shows from what you're hypothesizing — label hypotheses as
hypotheses. No algorithm mythology ("the algorithm hates X") —
stick to viewer behavior. If my data can't answer a question, say
what data would.

The video: [TITLE + one-line description + length]
The numbers I have: [VIEWS / CTR / AVG VIEW DURATION / RETENTION
GRAPH DESCRIPTION — e.g. "cliff at 0:45, slow slide from 3:00" /
LIKES / COMMENTS / anything else]
What happens in the video at the drop points: [DESCRIBE WHAT'S ON
SCREEN AT THE TIMES RETENTION DIPS]
Top 3 comments (verbatim): [PASTE OR "none"]
My goal for this video was: [VIEWS / SUBS / LEADS / SALES]

Give me:
1. The one-paragraph honest read: what the numbers say happened.
2. Drop-point diagnosis: for each dip, the most likely cause
   (hook debt, pacing, confusion, promise mismatch) — labeled
   confirmed-by-data or hypothesis.
3. What WORKED: the segments that held, and what they share.
4. Three specific changes for the next video, ordered by expected
   impact — each tied to a diagnosis above, not generic advice.
5. One thing to deliberately repeat unchanged, so I'm not
   reinventing what already works.
6. The single metric to watch on the next upload to test whether
   change #1 worked.
```

## Example output
[TO FILL AFTER TESTING]

## Tweaks
- Keep a running log of the "three changes" across videos — after five videos the log is your personal playbook
- Comments are retention data in disguise: "got boring in the middle" beats any graph
- Run monthly across all videos with the same prompt for pattern-level findings