How I Built a YouTube Title-Writing Skill for Claude (With Real Data)

I collected a list of 587 viral YouTube videos from 2025–2026 and analyzed them for patterns. Then I turned those findings into a single [.md] file that makes any AI dramatically better at writing YouTube titles and descriptions — on demand, every time. 🤔 The Problem With Asking AI to "Write a Good YouTube Title" Ask Claude or Gemini to write a YouTube title cold and you get... fine titles. "Top 10 Ways to Save Money in 2025" "How to Build a Morning Routine" They're grammatically correct. They're forgettable. They could have been written in 2018. The AI isn't being lazy — it simply lacks current, niche-specific data . It doesn't know that 71% of viral titles use a specific number, that 48% include the year, or that the highest-performing titles in 2025 follow a very specific money-stakes-survival formula. That's the gap a skill file closes. 🔬 The Methodology: Scraping 587 Viral Videos I used yt-dlp (a free command-line tool) to search YouTube across 100+ query categories — challenges,