Claude Code Hooks, Not Skills, for Stripping AI Writing Tells

My notes

Summary

A short TikTok walks through building a “humanizer” system for Claude Code that makes AI-written marketing copy read less like AI. Instead of a one-off prompt, the creator turned two source references, the Wikipedia “Signs of AI Writing” article and a research-paper-backed dataset called “Story Scope” that explains why LLMs write the way they do, into Claude Code hooks and core instructions, applied to funnels, emails, and marketing material.

Key Insight

• The method is grounded in two concrete, checkable sources rather than vague “sound more human” prompting: the Wikipedia “Signs of AI Writing” article (a maintained, crowd-sourced list of AI writing tells such as em dashes, “delve”, inflated symbolism, and promotional tone) and “Story Scope,” a dataset tied to a research paper on why LLMs default to those patterns.
• The creator deliberately chose hooks plus core instructions over a skill. Hooks and instructions apply automatically at write-time rather than needing to be invoked, which matters for something like anti-AI-writing style: it should be always-on, not opt-in per task.
• It is applied specifically to funnels, emails, and marketing material, meaning external-facing copy where AI tells are most damaging to credibility and conversion, not internal docs.
• The video itself is a lead-gen device, with the repo promised only to commenters, and no actual repo link or instruction text is shown, so the technique has to be reconstructed rather than copied.