name: researching-youtube-niche description: Research a YouTube niche to identify what's working — title patterns, description structure, video length, thumbnail style, hook formats, tags. Use when asked to "research a YouTube niche", "analyse competitor YouTube channels", "what's working on YouTube for [topic]", or "YouTube content strategy for [niche]". Calls youtube-researcher MCP tools. Do NOT use for transcript extraction or SEO optimisation of existing videos. allowed-tools: Read, Write, Bash, Agent
YouTube Niche Research
When to Use
- User wants to understand what performs well in a specific YouTube niche
- Competitor analysis for a YouTube topic or creator category
- Content strategy research before starting a channel or series
When NOT to Use
- Extracting a specific video's transcript
- Optimising existing video titles/descriptions/tags
Prerequisites
youtube-researcherMCP server running- Valid
YOUTUBE_API_KEYset in server env
Output Location
All research output saves to a user-specified folder, or defaults to:
~/Documents/youtube-research/
Naming convention: YYYY-MM-DD Research Title.md
Nothing gets echoed to the main conversation except the file path and a brief summary (5-10 lines max).
Workflow
Step 1: Get the Topic
The user must provide the topic/keywords before any research runs. Do not assume or infer a topic. If the user triggers this skill without a topic, ask:
What topic or keywords should I research?
No other questions. No parameter negotiation. Defaults handle everything else.
Step 2: Run Niche Analysis
youtube_analyse_niche(
query: [user's topic],
maxResults: 10,
order: "viewCount"
)
This returns the top 10 videos by view count with full metadata and aggregate statistics. Check quota after: youtube_quota_status.
Step 3: Download Thumbnails
Download thumbnails for all 10 videos:
youtube_get_thumbnails(
videoIds: [all 10 video IDs],
outputDir: ~/Downloads/youtube-research/[date]-[niche-slug]/
)
Step 4: Delegate Everything to a Single Agent
Spawn ONE agent that:
- Reads the raw niche analysis data (from the tool result file)
- Reads all 10 thumbnail images from the download directory
- Analyses everything
- Writes the full report to disk
Agent prompt template:
Read the raw YouTube niche analysis data from: [tool-result-path]
Read all thumbnail images from: [thumbnail-dir]
Analyse and write a research report to: [output-path]
Report structure:
# YouTube Niche Research: [Topic]
**Date:** YYYY-MM-DD
**Query:** [query]
**Videos analysed:** [count]
---
## What Works — Executive Summary
[5-7 bullet points: the most actionable findings for a creator entering this niche]
## Top 10 Videos by Engagement
| # | Title | Channel | Views | Engagement % | Duration | Published |
[Table of all 10 videos sorted by engagement rate]
## Title Strategy
- Winning patterns with specific examples from the data
- Title length range, power words, structures that correlate with high engagement
- 3 recommended title templates
## Thumbnail Strategy
- Visual patterns across all 10 thumbnails (faces, text, colours, logos, composition)
- What the top 3 by engagement do differently
- Specific recommendations for a new creator
## Description & Tags
- Above-the-fold patterns (first 2 lines)
- Tag frequency analysis — what tags the top performers use
- Recommended tags for this niche
## Duration & Format
- Duration distribution and where highest engagement concentrates
- Optimal length recommendation with rationale
## Channel Landscape
- Who dominates, subscriber distribution, where mid-tier creators break through
- Gap analysis: underserved angles or content types
## Actionable Recommendations
[7-10 specific, numbered recommendations for someone creating content in this niche. Focus on what to do differently from the established creators.]
---
**Thumbnails saved to:** [thumbnail-dir]
Return to me: Only the file path and a 1-line summary. Do NOT return the full content.
Step 5: Confirm to User
Report the file path and a 3-line summary:
Research complete for "[topic]".
Report: [file path]
Thumbnails: [thumbnail dir]
[1-2 sentence highlight of the most actionable finding]
No review checkpoint needed — the user reads the file directly.
Context Discipline
- The main conversation window never sees raw niche data, thumbnail analysis, or report content
- The niche analysis tool returns large JSON — always save to disk, never parse in main context
- All heavy work happens inside the delegated agent
- Main context total for this skill: ~10 tool calls max
Installation
Copy this file to your Claude Code skills directory:
mkdir -p .claude/skills/researching-youtube-niche
cp skill/SKILL.md .claude/skills/researching-youtube-niche/SKILL.md
Customise the output path and report structure to suit your workflow.
Troubleshooting
| Issue | Cause | Fix |
|---|---|---|
| "Missing required environment variable: YOUTUBE_API_KEY" | API key not set | Add key to MCP server config |
| "YouTube search failed: quotaExceeded" | Daily quota exceeded | Wait until midnight Pacific, or request quota increase |
| Few results returned | Niche too narrow | Broaden query terms |
| Thumbnail 404s | Older videos lack maxres | Fallback to hqdefault is automatic |