Headline Lab
Generates 10 headline variants against different frameworks (curiosity, benefit, controversy, listicle, how-to) and ranks them by predicted CTR and LLM-citability.
Entry verified April 21, 2026
The short answer
Most content dies on the headline. This skill forces you to see 10 angles before shipping, then ranks them for predicted CTR and LLM-citability — because LLMs are the new SERP. Output is 10 variants with a one-line rationale each, scored.
When to use it
Every post headline. Every newsletter subject line. Every YouTube title — anywhere a single line of copy gates whether the piece gets read.
Setup
- 1
Save the file as ~/.claude/commands/headline-lab.md.
- 2
Feed: target keyword, audience, and the piece's angle.
- 3
Skill returns 10 variants, each with a one-line rationale, plus a predicted CTR + citability score.
- 4
Predictions are directional, not data-backed — for stakes-worthy pieces, A/B test rather than ship the top-ranked variant blind.
Example
You: /headline-lab 'claude code skills' — audience solo devs.
Claude: 1. The 47 Claude Code skills that actually matter...
2. We ranked 200+ Claude Code skills by licence. Only 47 are safe to copy.
3. Your Claude Code setup is missing these 10 skills.
[...10 total, scored]Source & attribution
- Author
- Bryan Collins
- Licence
- MIT
- Type
- Original
Original pattern published under MIT — attribution preserved by convention, not licence requirement.
Caveats
CTR predictions are directional, not data-backed. Don't ship without A/B testing for stakes-worthy pieces.
56 skills across 10 categories, all licence-verified.