Free. Unsubscribe anytime.
Built so you can paste, answer 5 questions, save the file, and have Claude writing in your voice in 15 minutes.
The paste-ready 5-question prompt. Claude asks one question at a time, pushes for specifics, then generates your complete brand voice file as a code block you can copy.
18 patterns that scream "this was written by AI", grouped into 5 categories: vocabulary, sentence shapes, structural tells, rhetorical filler, and tone. Drop the worst ones into your voice file as a starter ban list.
The 3 paths explained: user global (everywhere), project-level (per repo), skill file (per workflow). Plus the /memory command to confirm it's actually loading.
The interview prompt and the AI-isms list work in any LLM. The setup section (chapter 3) is Claude Code specific, because CLAUDE.md is a Claude Code feature. For Claude.ai, you'd save the brand voice file as a Project instructions block. For ChatGPT, paste it into Custom Instructions.
10 is the floor. More is better. Mix formats: short captions, long-form posts, casual emails, formal copy. The model pattern-matches against the corpus you give it, so a wider corpus produces more flexible output.
Mostly yes, sometimes no. CLAUDE.md content is delivered as a strong nudge, not a hard rule. Specificity helps a lot: "never use the word 'delve'" holds tighter than "write naturally". When Claude slips, add the slip to your file as a new ban. Your file should grow over time.
Yes. CLAUDE.md is loaded at the start of each session. Mid-chat edits aren't picked up. After you tweak the file, start a fresh session for the changes to take effect.
Use a project-level CLAUDE.md at the root of that project. It stacks on top of your user-global file, so both load together. Useful if you write in a different tone for client work vs your personal brand.
Drop your email and I'll send the 5-page PDF straight to your inbox.
Send it to my inbox →