Voice Analyzer

The output becomes infrastructure -- a reference document used every time you work with AI to maintain your authentic voice instead of producing generic content.

1. Analyze patterns across all samples
2. Identify your distinctive voice markers
3. Generate a VOICE.md style guide
4. Create a forbidden phrases list specific to your anti-patterns
5. Provide testing prompts to validate the guide

Ideal samples: Newsletter issues, blog posts, emails you're proud of, social media threads that felt "like you"

Avoid: Heavily edited corporate content, collaborative pieces, anything that felt forced

Avoid samples that:

• Were heavily edited by others
• Feel generic or corporate even to you
• Were written under heavy constraints
• Don't represent how you want to sound going forward

More samples = more accurate pattern detection, but diminishing returns after 10.

Voice memo transcription: Record yourself explaining something you know well. Transcribe it. Your speaking patterns often reveal writing voice better than forced writing.

Email archaeology: Search sent mail for messages where you were being genuinely helpful, not performing professionalism.

Chat/Slack messages: Longer explanatory messages often show natural voice patterns.

Old social media: Before you started "performing" for algorithms.

# Voice Profile: [Name]
Generated: [Date]
Based on: [X] writing samples ([total word count] words)

## Voice Summary
[2-3 sentence description of overall voice character]

## Core Voice Characteristics

### Sentence Patterns
- Average length: [X] words
- Variation: [Low/Medium/High]
- Notable patterns: [specific observations]

### Vocabulary Fingerprint
- Formality: [Casual/Conversational/Professional/Formal]
- Characteristic phrases: [list]
- Words to use freely: [list]

### Rhythm and Flow
- Paragraph style: [description]
- Transition patterns: [description]
- Pacing notes: [description]

### Tone Markers
- Primary tone: [description]
- Humor style: [description]
- Reader relationship: [description]

### Structural Habits
- List vs. prose preference: [description]
- Formatting patterns: [description]

### Opinion Expression
- Directness level: [1-10]
- Qualification style: [description]
- Authority stance: [description]

## The Forbidden List

### Never Use (These kill your voice)
- [phrase 1]
- [phrase 2]
- [etc.]

### Use Sparingly (Context-dependent)
- [phrase 1] - only when [context]
- [etc.]

### Watch for Clusters (OK alone, problematic together)
- [pattern description]

## Voice Maintenance

### Monthly Check
- Read 3 recent pieces aloud
- Do they still sound like you?
- Update this guide if voice has evolved

### Quarterly Refresh
- Gather new strong samples
- Re-run analysis
- Compare to this guide
- Update patterns that have changed

## Testing Prompts

Use these to validate the guide works:

**Test 1 - Short form:**
"Using my voice profile, write a 3-sentence response to [common scenario in your field]"

**Test 2 - Long form:**
"Using my voice profile, write the opening 2 paragraphs for a piece about [topic you know well]"

**Test 3 - Edge case:**
"Using my voice profile, write about [topic outside your usual content]"

Compare outputs to your natural writing. If they feel off, update the guide.

Key findings:

• Sentences vary wildly (5-40 words)
• Heavy use of "Here's the thing" and "The reality is"
• Self-deprecating humor about own mistakes
• One-sentence paragraphs for emphasis
• Direct address to reader ("you")
• Specific numbers and timeframes ("three hours later, no hyperbole")

Forbidden list generated:

• "It's worth noting that"
• "In today's digital landscape"
• "Leverage" as a verb
• Paired adjectives ("comprehensive and thorough")
• Any sentence starting with "Furthermore" or "Moreover"

Key findings:

• Longer sentences with technical precision
• Code-switches between casual and technical
• Humor through understatement
• Heavy use of concrete examples
• "Actually" and "technically" as qualifiers
• Lists for technical content, prose for opinions

Forbidden list generated:

• Vague claims without specifics
• Marketing superlatives ("revolutionary," "game-changing")
• Excessive hedging ("it might be argued")
• Emoji in technical sections
• "Simply" before complex explanations

Using with other skills:

• slop-detector will reference VOICE.md for personalized pattern detection
• voice-editor uses VOICE.md to guide rewrites
• Include path to VOICE.md in project's CLAUDE.md for automatic loading

Maintenance schedule:

• Monthly: Quick read-aloud check (15 min)
• Quarterly: Full refresh with new samples (1 hour)
• After major voice shifts: Complete re-analysis

Quality check: The generated guide should allow someone unfamiliar with the writer to produce content that readers would recognize as authentic.

Voice doesn't live in first drafts. It lives in editing choices. The guide gives AI direction; your editing gives the work your actual voice.