How to Make Claude Sound Like You

I want to create a Claude Skill that captures my personal tone of voice, so every piece of AI-assisted writing I produce sounds like me, not generic AI output. The end deliverable is a SKILL.md file I can package and upload into Claude's Skills feature.

Work through this with me in five stages. Do not skip ahead. Wait for my input at each stage before moving on. Do not pad your responses with filler.

STAGE 1: Sample collection
Ask me to paste in 4 to 6 pieces of writing I'm proud of. Anything that sounds like me at my best: LinkedIn posts, emails, articles, Slack messages, blog posts. Wait until I've shared them before moving on.

STAGE 2: Pattern analysis
Once I've shared the samples, analyse them carefully and tell me:
- My typical sentence length and rhythm (short and punchy, long and flowing, or a mix)
- How I open pieces (hook, anecdote, statement, question, contrarian take)
- How I close pieces (call to action, rhetorical question, sharp line, callback)
- Words, phrases, or constructions I use repeatedly, both good and bad
- Formatting habits (bullets, line breaks, capitalisation, punctuation choices)
- Tone qualities (warm, dry, direct, formal, playful, self-aware, etc.)
- What I never seem to do (passive voice, hedging language, corporate filler)

Show me your full analysis before moving on. I'll correct anything that's off.

STAGE 3: Rule extraction
Turn the analysis into a usable rule set. Ask me direct questions to lock in:
- 3 to 6 hard rules I want enforced every single time (e.g. "never use em dashes")
- 5 to 10 banned words and phrases I refuse to publish (clichés, corporate filler, anything that makes me cringe)
- 3 to 5 tone descriptors that capture how I sound at my best
- 3 to 5 structural habits I want preserved (opener style, paragraph length, closing style)

Push me to be specific. Vague instructions ("don't be corporate") don't survive contact with the model. Concrete bans ("never use 'leverage,' 'unlock value,' 'circle back,' or 'I hope this email finds you well'") do.

STAGE 4: Trigger description
Help me write a clear, specific, "pushy" description of when this skill should fire. Skills tend to under-trigger by default, so the description needs to be assertive.

The description must include two parts:
1. A pushy "use when" trigger. Use language like "Make sure to use this skill whenever..." and list the specific scenarios.
2. An explicit "Do NOT use for" non-trigger that prevents the skill firing on tasks it shouldn't touch (e.g. code, internal scratch notes, technical documentation).

Bad: "writing tasks."
Good: "Make sure to use this skill whenever the user asks Claude to draft or edit any content that will be sent or published externally, including LinkedIn posts, emails, Slack messages, web copy, articles, and cover letters. Apply the rules and tone defined below. Do NOT use this skill for code, technical documentation, internal scratch notes, or research summaries the user is reading for themselves."

Draft three versions. I'll pick or combine.

STAGE 5: Generate the SKILL.md
Output the final SKILL.md file in this exact format, ready for me to package:

---
name: [my-name]-voice
description: [pushy use-when + explicit do-not-use-for from stage 4]
---

# [My name]'s voice

## Hard rules (non-negotiable)
[3 to 6 rules from stage 3, each as a bullet, written as absolute instructions starting with "Never" or "Always"]

## Tone
[3 to 5 descriptors from stage 3, each as a bullet]

## Structure
[3 to 5 structural habits from stage 3, each as a bullet]

## Banned phrases
[5 to 10 phrases from stage 3, each as a bullet]

## Examples
Generate 4 examples that show the voice in action. Two "good" (matching the rules) and two "bad" (violating them, would be rewritten). Cover at least two formats from my actual use case (e.g. LinkedIn post + email).

### Good: [format 1]
[3 to 5 sentences in my voice]

### Bad: [format 1]
[3 to 5 sentences in generic AI voice that breaks my rules]

### Good: [format 2]
[3 to 5 sentences in my voice]

### Bad: [format 2]
[3 to 5 sentences in generic AI voice that breaks my rules]

Do not generate the skill until we've worked through every stage. Do not assume my voice without samples. Do not invent rules I haven't given you. If my answers are vague or contradictory, push back and ask me to sharpen them. Be direct with me throughout.