THE BRAND VOICE CLONER

Drop in 3 things you've written. Claude builds a voice profile + 5 templates so every email, social post, and sales page sounds like YOU. The skill that makes AI write FOR you, not in some generic "AI voice."

THE PROBLEM

You ask Claude to write an email. It comes out sounding like a polite robot. Generic. Em-dashes everywhere. "Just wanted to..." opener. None of the personality people actually like about your writing.

The fix isn't worse AI. It's a voice profile so specific that everything Claude writes for you is recognizably yours.

THE SKILL

Setup once: drop in 3 samples of your writing — an email you sent, a social post you wrote, a longer piece (article, blog, newsletter).

Claude analyzes:

Sentence rhythm + length
Vocabulary preferences
Greeting + sign-off style
Em-dash habit (or aversion)
Emoji use
Self-references and inside jokes
What you avoid

Builds a voice profile + 5 templates:

Email reply
Social post
Sales / promo copy
Long-form intro
Difficult message

Every future request that says "in my voice" uses this profile.

INSTALL

Standard.

THE FULL SKILL FILE

---
name: brand-voice-cloner
description: Builds a voice profile from 3 writing samples. Outputs voice-profile document + 5 templates (email, social, sales, long-form, difficult). All future content "in my voice" uses this profile.
when_to_use: User says "match my voice," "sound like me," "brand voice," or shares writing samples and wants a voice analysis.
---

# The Brand Voice Cloner

You analyze writing samples and clone the voice. Specific. Pattern-aware. Anti-generic.

## SETUP MODE (run once)

Ask user for 3 samples:
1. **An email they sent** (any context)
2. **A social media post** (Instagram caption, LinkedIn post, tweet)
3. **A longer piece** (newsletter, article, blog post)

Process: extract patterns and build voice profile.

## VOICE PROFILE OUTPUT

🎤 VOICE PROFILE — [USER NAME]

SENTENCE RHYTHM:

Average length: [X words]
Range: [shortest – longest]
Cadence: [staccato / flowing / mixed]

VOCABULARY:

Common words: [list 10]
Technical level: [casual / mixed / technical]
Slang / colloquialisms: [yes/no, examples]

GREETINGS + SIGN-OFFS:

Email greeting: [Hi/Hey/None]
Email sign-off: [thanks/cheers/initials/none]
Social opener: [pattern]
Social closer: [pattern]

PUNCTUATION:

Em dashes: [aggressive / occasional / never]
Exclamation marks: [count per 100 words]
Parentheses: [common / rare]
Periods vs. ellipses

EMOJI:

Frequency: [never / occasional / frequent]
Favorite ones: [list]
Where placed: [end of post / inline / never in serious]

PERSONALITY MARKERS:

Self-references: [I / "we" / third-person]
Humor type: [dry / warm / self-deprecating / none]
Vulnerability: [open / private / mixed]
Recurring phrases: [list 5]

NEVER WRITES:

[Phrase user demonstrably avoids]
[Pattern user demonstrably avoids]
[Trope user demonstrably avoids]


## TEMPLATE OUTPUTS (5)

For each, generate a sample in user's voice:

### 1. EMAIL REPLY (3-sentence acknowledge)
[sample reply to a hypothetical "thanks for sending the doc" message]

### 2. SOCIAL POST (single thought)
[sample 1-paragraph Instagram or LinkedIn post]

### 3. SALES / PROMO COPY (selling something)
[sample 1-paragraph promo for a hypothetical product]

### 4. LONG-FORM INTRO (newsletter / blog opener)
[sample 3-paragraph intro to a hypothetical post]

### 5. DIFFICULT MESSAGE (hard email or boundary)
[sample 4-sentence "I can't do this" message]

## What NOT to do

- Don't impose generic "good writing" rules over the user's actual patterns
- Don't recommend changes to their voice ("you should use more emojis")
- Don't generate templates that don't reflect at least 3 specific markers from the profile
- Don't extract patterns from samples that look like edited corporate copy (push for more authentic samples)

## When samples are too few or too uniform

- **All samples are emails** → "Send me one social post + one longer piece. Voice varies by medium."
- **Samples are all from work** → "Want one personal sample too? Otherwise the voice will read corporate."
- **Samples are too short** → "Each sample should be at least 100 words to extract real patterns."

## Delivery
End with: *"Voice profile saved. Future requests with 'in my voice' will use this. Update samples every 6 months — voice drifts."*

SAFETY CHECK

Same as Day 1.

WHAT'S NEXT

Day 23 of 100. Pair with Day 5 — The 3-Word Email Reply (which uses this voice profile) and Day 22 — Founder Bio Builder.