An 8-module YouTube research and analytics tool, built for one niche: AI trading and build-in-public content.
While documenting the build of my forex trading bot on YouTube, I was posting completely blind. Some videos got 700+ views, others got 14 — and I had no idea why. Tools like TubeBuddy and VidIQ exist, but they're built for general creators, not for a specific niche like algo trading and AI-built systems.
So I built my own — configured for my exact corner of YouTube: AI trading, forex bots, algo trading, Claude AI coding, build in public, MetaTrader 5.
======================================================
YouTube Research Tool
======================================================
1. Full research (all features)
2. Competitor analysis only
3. Title & keyword analysis only
4. Score a title
5. Check for trending changes
6. Download thumbnails from last research
7. My channel stats
8. My channel deep analytics (OAuth)
======================================================
Core engine — searches 7 configurable niche keywords, pulls the top 20 videos per keyword with full metadata, filters out channels over 500k subs and Shorts, deduplicates across keywords, flags repeat competitors.
Competitor analysis — thumbnail downloader, channel deep dives (last 10 videos + upload frequency per competitor), upload timing heatmap.
Title & keyword intelligence — sentiment classifier (How-To / Results / Curiosity / Fear / Story / Question / General, ranked by performance), keyword gap finder, title scorer (0-10 based on length, power words, sentiment, numbers, brackets), trending tracker (flags 20%+ growth between runs).
My channel stats — subscriber count, total views, full video metadata, best/worst performers, benchmarked against small competitor channels.
Deep analytics (OAuth) — watch time, average view duration, retention per video, Shorts vs long-form breakdown, traffic sources, upload-day impact. Exports to CSV.
Every run produces timestamped CSVs across all modules, plus organised thumbnail archives.
Running this against my own channel (2 months old, 28 videos) immediately surfaced a clear pattern:
- "Did You Know" format titles: ~81 views average
- Story/Outcome format titles: ~340+ views average
- ~4x performance difference
That single insight — drop one title format, double down on another — came directly from this tool's sentiment classifier. It's what it was built to do.
- Python — core language
- Claude Code — entire system built through it, no prior formal programming background
- YouTube Data API v3 — search, metadata, public stats
- OAuth 2.0 + YouTube Analytics API — owner-only deep analytics
Active CLI tool, running weekly research cycles for The Imperfect Algorithm. Build process documented on YouTube.
Currently a personal CLI tool. Next steps: offering it as a done-for-you research service for creators in adjacent niches, eventually a self-serve web dashboard with OAuth channel connection.
Need niche-specific YouTube research, or a custom version for your niche?