home / skills / different-ai / agent-bank / youtube-rl-tracker

youtube-rl-tracker skill

/.opencode/skill/youtube-rl-tracker

This skill helps you optimize YouTube video performance by testing thumbnail styles and titles to identify winning patterns and accelerate growth.

npx playbooks add skill different-ai/agent-bank --skill youtube-rl-tracker

Review the files below or copy the command above to add this skill to your agents.

Files (1)
SKILL.md
6.3 KB
---
name: youtube-rl-tracker
description: Track YouTube video performance for "poor man's reinforcement learning" - learn what thumbnails, titles, and hooks work
license: MIT
compatibility: opencode
metadata:
  service: notion, youtube
  category: content
---

## What I Do

Track YouTube video performance to discover patterns in what works. This is "poor man's reinforcement learning" - manually logging outcomes to improve over time.

## The RL Loop

```
1. PUBLISH  -> Upload video with hypothesis (thumbnail style, title hook, topic)
2. WAIT     -> Let it run for 48-72 hours
3. LOG      -> Record in Notion with views, CTR, retention
4. ANALYZE  -> Compare winners vs losers
5. REPEAT   -> Apply learnings to next video
```

## Key Insight from First Data Point

**Video 1:** "Using AI agents to pay bills and send invoices"

- 8 views in 1 day
- Plain talking head thumbnail
- Generic title

**Video 2:** "Paying My Contractor Through Claude | AI-Powered Finance"

- 133 views in 5 days (16x better!)
- Thumbnail shows: Face + Product UI overlay + Text "I Let AI Pay My Bills"
- Title has: Specific action + Brand name (Claude) + Category tag

### What Made Video 2 Win:

1. **Thumbnail has TEXT overlay** - "I Let AI Pay My Bills" creates curiosity
2. **Shows the PRODUCT** - UI screenshot proves it's real, not just talk
3. **Face + Context** - Person looking at the UI, not just talking
4. **Specific title** - "Paying My Contractor" > "pay bills" (concrete vs abstract)
5. **Brand name in title** - "Claude" attracts AI-interested audience
6. **Category tag** - "AI-Powered Finance" helps discoverability

### Hypothesis to Test:

> Thumbnails with TEXT + PRODUCT UI + FACE outperform plain talking head thumbnails by 10x+

## Database Schema

### Core Fields (Outcomes)

| Property  | Type     | Purpose                            |
| --------- | -------- | ---------------------------------- |
| Title     | title    | Video title                        |
| Views     | number   | Total views                        |
| CTR       | number   | Click-through rate (%)             |
| Retention | number   | Average view duration (%)          |
| Days Live | number   | Days since publish                 |
| Views/Day | formula  | `Views / Days Live`                |
| Worked?   | checkbox | Binary gut-check - was this a win? |

### Input Features (What You Controlled)

| Property        | Type     | Options                                         |
| --------------- | -------- | ----------------------------------------------- |
| Thumbnail Style | select   | Talking Head, Face+UI, Face+Text, UI Only, Meme |
| Has Text        | checkbox | Does thumbnail have text overlay?               |
| Has Product     | checkbox | Does thumbnail show the product/UI?             |
| Title Hook      | select   | How-To, Story, Listicle, Question, Bold Claim   |
| Has Brand       | checkbox | Does title mention a brand (Claude, ChatGPT)?   |
| Topic           | select   | AI Finance, Automation, Product Demo, Tutorial  |
| Duration        | number   | Video length in minutes                         |
| Posted          | date     | When published                                  |

### Reference Fields

| Property  | Type      | Purpose                 |
| --------- | --------- | ----------------------- |
| URL       | url       | Link to video           |
| Thumbnail | files     | Screenshot of thumbnail |
| Notes     | rich_text | Why did it work/fail?   |

## First Entry: The Baseline

```
Video 1 (LOSER):
- Title: "Using AI agents to pay bills and send invoices"
- Views: 8
- Days Live: 1
- Thumbnail Style: Talking Head
- Has Text: No
- Has Product: No
- Title Hook: How-To
- Has Brand: No
- Notes: Plain talking head, generic title, no visual hook

Video 2 (WINNER):
- Title: "Paying My Contractor Through Claude | AI-Powered Finance"
- Views: 133
- Days Live: 5
- Views/Day: 26.6
- Thumbnail Style: Face+UI
- Has Text: Yes ("I Let AI Pay My Bills")
- Has Product: Yes (shows invoice payment UI)
- Title Hook: Story
- Has Brand: Yes (Claude)
- Notes: Text overlay creates curiosity, UI proves it's real, specific action in title
```

## Thumbnail Patterns to Test

Based on initial data:

| Pattern          | Example | Hypothesis                                          |
| ---------------- | ------- | --------------------------------------------------- |
| Face + UI + Text | Video 2 | Best performer - proves reality + creates curiosity |
| Face + Bold Text | -       | May work for controversial takes                    |
| UI Only          | -       | Good for tutorials, may lack personality            |
| Talking Head     | Video 1 | Worst - no visual hook                              |
| Before/After     | -       | Good for transformation content                     |

## Title Patterns to Test

| Pattern                 | Example                               | Hypothesis                     |
| ----------------------- | ------------------------------------- | ------------------------------ |
| Specific Action + Brand | "Paying My Contractor Through Claude" | Winner - concrete + searchable |
| Generic Action          | "Using AI to pay bills"               | Loser - too vague              |
| Question                | "Can AI Really Pay Your Bills?"       | Untested - may drive curiosity |
| Number + Outcome        | "I Automated 5 Hours of Finance Work" | Untested - quantified value    |

## Weekly Review Process

1. **Sort by Views/Day** - normalize for time live
2. **Filter by Worked? = true** - what patterns emerge?
3. **Group by Thumbnail Style** - which visuals win?
4. **Group by Title Hook** - which hooks work?
5. **Compare Has Text vs No Text** - does text help?
6. **Compare Has Brand vs No Brand** - do brand names help?

## Integration with YouTube Studio Skill

Use `skill("youtube-studio")` to:

- Upload new videos with test hypotheses
- Update thumbnails based on learnings
- Track which changes improve performance

## Reference Image

The winning thumbnail (133 views) vs losing thumbnail (8 views):

![YouTube Comparison](/Users/benjaminshafii/git/zerofinance/.opencode/skill/youtube-rl-tracker/reference-comparison.png)

**Key visual differences:**

- Winner: Face looking at UI, text overlay "I Let AI Pay My Bills", product screenshot visible
- Loser: Plain talking head against brick wall, no text, no product context

Overview

This skill helps creators run a simple, repeatable experiment to improve YouTube performance using manual 'poor man's reinforcement learning.' It captures video inputs (thumbnails, titles, hooks) and outcome metrics so you can identify what patterns actually move views, CTR, and retention. The goal is faster iteration: publish hypotheses, wait, log results, analyze winners, and repeat.

How this skill works

You publish a video with a clear hypothesis about thumbnail style, title hook, or topic. After 48–72 hours (or a few days), log key outcomes (views, CTR, retention, days live) and controlled inputs (thumbnail style, text overlay, product UI, title type, brand mentions). The skill normalizes by views/day, surfaces winning patterns (e.g., Face+UI+Text), and guides the next hypothesis. Integrations allow uploading and updating assets via a YouTube Studio connector.

When to use it

  • Testing thumbnail designs to boost CTR and early velocity
  • Comparing title formats to improve search and click intent
  • Learning which product visuals or brand mentions drive discovery
  • Normalizing results across videos with different publish dates
  • Running a weekly review to extract repeatable patterns

Best practices

  • Treat each video as one experiment: change only a few variables at a time
  • Record outcomes after 48–72 hours and compute views/day for fair comparison
  • Use a consistent Notion schema (title, views, CTR, retention, thumbnail style)
  • Prioritize visual hooks: test Face+UI+Text before pure talking head
  • Group winners by thumbnail style and title hook during weekly reviews

Example use cases

  • Prove whether adding product UI to thumbnails increases CTR and retention
  • Validate if including a brand name (Claude, ChatGPT) attracts a niche audience
  • Optimize title wording: specific action vs generic claim vs question
  • Run a sequence of videos where only thumbnail text is varied to test curiosity hooks
  • Automate asset updates: upload improved thumbnails for live videos via YouTube Studio

FAQ

How long should I wait before logging results?

Check early performance after 48–72 hours and use views/day to normalize for longer windows.

What thumbnail pattern performed best in initial tests?

Face+Product UI+Text overlay showed a large early lift versus plain talking head thumbnails.