AI Content Forensics
Reverse-engineer any YouTuber's content strategy with data, not guesswork.
AI Content Forensics is a skill for Claude Code and Cowork that runs an autonomous research pipeline on any YouTube creator's channel. It analyzes their titles, thumbnails, hooks, and video structure — then distills the findings into a viral-ready thread with production-ready carousel visuals.
One command. Full analysis. Copy-paste-ready output.
What It Does
Give it a YouTuber's name. It does the rest:
- Research — Scrapes the creator's long-form YouTube catalog (metadata, transcripts, thumbnails, view counts)
- Analyze — Extracts packaging features across every video, runs 4-layer pattern analysis, identifies what actually drives performance
- Synthesize — Creates 5 operational "constitutions" (title rules, thumbnail rules, hook rules, structure rules, master rules) backed by corpus evidence
- Write — Produces a 9-post Synthesizer-style thread with data-backed insights
- Visualize — Generates 9 carousel images (SVG + HTML) matching each thread post
Installation
As a Claude Code / Cowork plugin (recommended)
claude plugins add ai-content-forensics
Manual installation
Clone this repo and copy the skill folder into your Claude skills directory:
git clone https://github.com/lennoxsaint/ai-content-forensics.git
cp -r ai-content-forensics/skills/ai-content-forensics ~/.claude/skills/
Usage
Once installed, just tell Claude what you want:
"Analyze Ali Abdaal's YouTube packaging"
"Run AI Content Forensics on MrBeast"
"Break down how Alex Hormozi titles his videos"
The skill asks for the target YouTuber (required) and a few optional settings, then runs the full pipeline autonomously.
Optional Configuration
| Setting | Default | What it does |
|---|
output_mode | full | full = research + thread + visuals. research_only = just the analysis. thread_only = analysis + thread, no visuals |
your_channel_handle | none | Your YouTube channel — enables portability scoring (which patterns you can actually steal) |
publisher_handle | none | Your handle for the thread CTA (e.g. @lennox_saint) |
platform_name | Threads | Target platform: Threads, X, LinkedIn, or Bluesky |
time_window_months | 24 | How far back to analyze |
| Visual colors | warm paper tones | Customize visual_background, visual_accent, visual_text |
Example: Research Only
"I just want the research on MKBHD's channel — no thread, no visuals"
→ output_mode = research_only
Example: Full Pipeline with Portability
"Analyze Ali Abdaal's packaging and compare it to my channel @myhandle.
Post the thread from @lennox_saint."
→ target_youtuber = Ali Abdaal
→ your_channel_handle = @myhandle
→ publisher_handle = @lennox_saint
What You Get
After a full run, you'll find a structured research folder:
research/youtube-packaging/ali-abdaal/
├── 00_run_report.md ← Progress log and summary
├── 02_target_creator_profile.md ← Creator classification
├── 04_video_index.csv ← All videos with metadata
├── 10_exhaustive_synthesis.md ← Full findings + 15+ ranked insights
├── constitutions/ ← 5 operational rule sets
│ ├── 00_master_packaging_constitution.md
│ ├── 01_title_constitution.md
│ ├── 02_thumbnail_constitution.md
│ ├── 03_hook_constitution.md
│ └── 04_script_and_structure_constitution.md
├── thread/
│ ├── final_thread.md ← Copy-paste-ready 9-post thread
│ └── insights_audit.md ← Why each insight was chosen
├── visuals/assets/ ← 9 carousel SVGs + HTMLs
└── logs/ ← Full audit trail
Sample Thread Output
Ali Abdaal has mass-produced 400+ videos in 4 years and still pulls
8-figure views on a "boring" niche.
I broke down 87 of his long-form videos to find the patterns.
Here are 7 lessons smaller creators can steal:
—
His top 10 videos average 14.2M views.
They all start with a number in the title.
Not a hack — just pattern. 91% of his top performers
use a specific number in the first 3 words.
Threads takeaway:
Put your strongest number before word 4 in your title.
—
[... 5 more insight posts ...]
—
What you should do now:
I went through 87 videos, 200+ hours of transcripts, and
5 packaging constitutions for this.
The patterns are clear: lead with numbers, open with proof
not promises, and use your title to set a specific expectation.
Follow @lennox_saint for more data-backed creator breakdowns.
Once the thread is ready, the skill opens Threadify so you can schedule and publish it directly to Threads.
Requirements