Brand Voice Extraction Skill

Overview

Brand voice is the consistent personality expressed through words. This skill provides a rigorous, evidence-based methodology for extracting voice attributes from client materials—not inventing them, not assuming them, but documenting what's actually there.

The Voice Extraction Framework

The Six Core Voice Dimensions

Every brand voice can be mapped across six dimensions. Each is a spectrum, not a binary:

code

DIMENSION 1: FORMALITY
├─ 1-2: Very Casual ("Hey! Let's chat about...")
├─ 3-4: Casual ("Here's the deal...")
├─ 5-6: Balanced ("We're happy to help you...")
├─ 7-8: Professional ("We are pleased to assist...")
└─ 9-10: Very Formal ("It is our privilege to serve...")

DIMENSION 2: ENTHUSIASM  
├─ 1-2: Reserved ("The product works well.")
├─ 3-4: Measured ("We're confident you'll see results.")
├─ 5-6: Engaged ("We love helping clients succeed!")
├─ 7-8: Energetic ("This is AMAZING and you'll love it!")
└─ 9-10: Exuberant ("OMG this will literally change your LIFE!!!")

DIMENSION 3: HUMOR
├─ 1-2: Serious (No humor, all business)
├─ 3-4: Occasional wit (Light touches, subtle)
├─ 5-6: Playful (Regular humor, approachable)
├─ 7-8: Witty (Clever wordplay, personality-forward)
└─ 9-10: Comedy-forward (Humor is the brand)

DIMENSION 4: DIRECTNESS
├─ 1-2: Soft/Indirect ("You might consider...")
├─ 3-4: Gentle ("We suggest...")
├─ 5-6: Clear ("Here's what to do...")
├─ 7-8: Direct ("Do this.")
└─ 9-10: Bold/Blunt ("Stop wasting time. Do this now.")

DIMENSION 5: TECHNICAL COMPLEXITY
├─ 1-2: Very Simple (5th grade reading level)
├─ 3-4: Accessible (8th grade, no jargon)
├─ 5-6: Moderate (Some industry terms, explained)
├─ 7-8: Technical (Industry jargon expected)
└─ 9-10: Expert (Dense, specialized vocabulary)

DIMENSION 6: WARMTH
├─ 1-2: Distant/Impersonal ("Users will find...")
├─ 3-4: Neutral ("You can...")
├─ 5-6: Friendly ("We're here for you...")
├─ 7-8: Warm ("We genuinely care about...")
└─ 9-10: Intimate ("You're part of our family...")

Extraction Process

Step 1: Gather Writing Samples

Ideal Sample Set:

•3-5 pieces of existing content (website, emails, social)
•Stakeholder writing samples (founder emails, team communications)
•Content they admire (competitors or brands they like)
•Content they dislike (what to avoid)

Minimum Viable:

•At least 1 substantial piece (500+ words)
•OR questionnaire responses about voice preferences

Document Your Inputs:

json

{
  "samples_analyzed": [
    {
      "source": "Homepage copy",
      "word_count": 450,
      "content_type": "marketing",
      "url_or_location": "https://..."
    }
  ]
}

Step 2: Analyze Each Sample

For each writing sample, extract:

A. Sentence-Level Patterns

markdown

SENTENCE LENGTH:
- Average words per sentence: [count]
- Range: [shortest] to [longest]
- Pattern: [short/medium/long/varied]

SENTENCE STRUCTURE:
- Simple sentences: [%]
- Compound sentences: [%]
- Complex sentences: [%]
- Questions used: [yes/no, frequency]
- Exclamations used: [yes/no, frequency]

EXAMPLES:
- Typical sentence: "[quote]"
- Atypical sentence: "[quote]"

B. Word-Level Patterns

markdown

VOCABULARY:
- Preferred verbs: [list with examples]
- Avoided verbs: [list or note absence]
- Industry jargon used: [yes/no, examples]
- Contractions: [always/sometimes/never]

SPECIFIC WORD CHOICES:
- How do they say "you"? [you/y'all/folks/clients/customers]
- How do they say "we"? [we/our team/the company/I]
- Power words used: [list]
- Words conspicuously absent: [list]

EXAMPLES:
- Characteristic phrase: "[quote]"
- Repeated pattern: "[quote]"

C. Emotional Tone

markdown

EMOTIONAL REGISTER:
- Primary emotion conveyed: [confidence/excitement/calm/urgency/etc.]
- Secondary emotions: [list]
- Emotions avoided: [list]

EVIDENCE:
- Quote showing primary emotion: "[quote]"
- Quote showing avoidance: "[quote]"

D. Structural Patterns

markdown

PARAGRAPH LENGTH:
- Average sentences per paragraph: [count]
- Pattern: [short punchy/medium/long flowing]

FORMATTING:
- Uses bullet points: [frequently/occasionally/never]
- Uses headers: [frequently/occasionally/never]
- Uses bold/emphasis: [frequently/occasionally/never]

TRANSITIONS:
- Common transition phrases: [list]
- Paragraph openings pattern: [observation]

Step 3: Score Each Dimension

For each of the six dimensions, assign a score (1-10) with evidence:

json

{
  "dimension": "formality",
  "score": 6,
  "evidence": [
    {
      "quote": "We're excited to help you get started",
      "source": "Homepage",
      "analysis": "Contraction 'we're' + 'excited' suggests casual-professional balance"
    },
    {
      "quote": "Our team is dedicated to your success",
      "source": "About page", 
      "analysis": "More formal phrasing, no contractions"
    }
  ],
  "conclusion": "Balanced formality (5-7 range) - professional but not stiff, uses contractions selectively"
}

CRITICAL: Every score needs evidence. No evidence = no score.

Step 4: Identify Patterns Across Samples

Look for consistency and inconsistency:

markdown

CONSISTENT PATTERNS (appear in 80%+ of samples):
- [Pattern]: [evidence across samples]

INCONSISTENT PATTERNS (vary across samples):
- [Pattern]: [note the variation and possible reason]

ANOMALIES (one-off departures):
- [Anomaly]: [where it appeared, possible explanation]

Step 5: Synthesize Voice Guidelines

Convert analysis into actionable guidelines:

markdown

## DO's (with examples)

### DO use contractions in conversational contexts
- ✓ "We're here to help"
- ✓ "You'll love this"
- ✗ "We are pleased to inform you"

### DO lead with benefits
- ✓ "Save 3 hours every week"
- ✗ "Our software has automated scheduling"

## DON'Ts (with examples)

### DON'T use corporate jargon
- ✗ "Leverage synergies"
- ✗ "Best-in-class solutions"
- ✓ "Work better together"
- ✓ "Tools that actually help"

### DON'T be overly casual
- ✗ "OMG you guys!!!"
- ✗ "This is literally the best"
- ✓ "We're genuinely excited about this"

Anti-Hallucination Rules for Voice Extraction

Rule 1: Evidence or Nothing

code

✗ WRONG: "The brand voice is friendly and approachable"
✓ RIGHT: "The brand voice appears friendly [Evidence: 8/10 sentences use 'you' directly, contractions in 70% of copy, exclamation points in 3/5 headlines - Source: Homepage analysis]"

Rule 2: Quote Don't Paraphrase

code

✗ WRONG: "They tend to be enthusiastic"
✓ RIGHT: "Enthusiasm is present [Example: 'We absolutely love helping small businesses thrive!' - Source: About page, para 2]"

Rule 3: Note Sample Limitations

code

If only analyzing one sample:
→ State: "Based on limited sample (homepage only). Voice profile confidence: LOW. Recommend analyzing additional content before finalizing guidelines."

Rule 4: Distinguish Preference from Practice

code

If client says they want to sound "bold and edgy" but their content is conservative:
→ Flag: "[INCONSISTENCY: Stated preference is 'bold/edgy' but existing content scores 4/10 on directness. Clarify: Should new content match existing voice or shift toward stated preference?]"

Rule 5: Don't Invent "Competitor Voice"

code

✗ WRONG: "Competitors tend to be more formal"
✓ RIGHT: "Competitor voice not analyzed - no competitor samples provided"
OR
✓ RIGHT: "Competitor [Name] voice analyzed: [specific observations with quotes]"

Output Schema: Brand Voice Profile

json

{
  "profile_metadata": {
    "client_name": "",
    "created_date": "",
    "samples_analyzed": [
      {
        "source": "",
        "type": "",
        "word_count": 0,
        "url": ""
      }
    ],
    "confidence_level": "high|medium|low",
    "confidence_notes": ""
  },
  
  "voice_dimensions": {
    "formality": {
      "score": 1-10,
      "label": "Very Casual|Casual|Balanced|Professional|Very Formal",
      "evidence": [
        {
          "quote": "",
          "source": "",
          "analysis": ""
        }
      ],
      "summary": ""
    },
    "enthusiasm": {
      "score": 1-10,
      "label": "Reserved|Measured|Engaged|Energetic|Exuberant",
      "evidence": [],
      "summary": ""
    },
    "humor": {
      "score": 1-10,
      "label": "Serious|Occasional Wit|Playful|Witty|Comedy-Forward",
      "evidence": [],
      "summary": ""
    },
    "directness": {
      "score": 1-10,
      "label": "Soft|Gentle|Clear|Direct|Bold",
      "evidence": [],
      "summary": ""
    },
    "technical_complexity": {
      "score": 1-10,
      "label": "Very Simple|Accessible|Moderate|Technical|Expert",
      "evidence": [],
      "summary": ""
    },
    "warmth": {
      "score": 1-10,
      "label": "Distant|Neutral|Friendly|Warm|Intimate",
      "evidence": [],
      "summary": ""
    }
  },
  
  "tone_guidelines": {
    "primary_tone": {
      "tone": "",
      "description": "",
      "evidence": ""
    },
    "secondary_tones": [
      {
        "tone": "",
        "when_to_use": ""
      }
    ],
    "tones_to_avoid": [
      {
        "tone": "",
        "why_avoid": ""
      }
    ]
  },
  
  "language_patterns": {
    "sentence_patterns": {
      "average_length": "",
      "preferred_structure": "",
      "uses_questions": true|false,
      "uses_exclamations": true|false
    },
    "word_preferences": {
      "preferred_words": [
        {
          "word": "",
          "example_usage": "",
          "source": ""
        }
      ],
      "avoided_words": [
        {
          "word": "",
          "why_avoid": ""
        }
      ]
    },
    "pronoun_usage": {
      "first_person": "we|I|our team|[brand name]",
      "second_person": "you|y'all|folks|customers",
      "evidence": ""
    },
    "contractions": "always|usually|sometimes|rarely|never",
    "jargon_level": "none|minimal|moderate|heavy"
  },
  
  "dos_and_donts": {
    "dos": [
      {
        "guideline": "",
        "good_example": "",
        "source": ""
      }
    ],
    "donts": [
      {
        "guideline": "",
        "bad_example": "",
        "good_alternative": ""
      }
    ]
  },
  
  "application_notes": {
    "headlines": "",
    "body_copy": "",
    "ctas": "",
    "error_messages": "",
    "email_subject_lines": "",
    "social_media": ""
  },
  
  "inconsistencies_flagged": [
    {
      "observation": "",
      "sources": [],
      "resolution_needed": ""
    }
  ],
  
  "gaps_in_analysis": [
    {
      "what_missing": "",
      "impact": "",
      "recommendation": ""
    }
  ]
}

Verification Checklist

Before finalizing voice profile:

Evidence Check

• Every dimension score has at least 2 supporting quotes
• All quotes include source attribution
• No scores based on assumption alone

Consistency Check

• Scores across dimensions form coherent picture
• Any inconsistencies are flagged and explained
• Stated preferences vs. actual practice are reconciled

Completeness Check

• All six dimensions scored
• Do's and Don'ts have concrete examples
• Application notes cover key content types

Usability Check

• Guidelines are specific enough to apply
• Examples are clear and representative
• A copywriter could use this to write on-brand content

Integration Notes

This skill is typically used by:

•client-discovery-agent (during onboarding)
•landing-page-copywriter skill (for voice-consistent copy)
•website-copywriter skill (for voice-consistent copy)
•/voice command (for voice extraction or application)

Voice profiles should be saved to a consistent project location for downstream access.

Quick Reference Card

For rapid voice extraction when time is limited:

markdown

## 60-Second Voice Scan

Read 3 paragraphs of client content. Note:

1. FORMALITY: Contractions? (casual) or formal phrasing?
2. ENTHUSIASM: Exclamation points? Power words? Or measured?
3. HUMOR: Any wit/jokes? Or purely informational?
4. DIRECTNESS: "You should" (direct) or "You might consider" (soft)?
5. COMPLEXITY: Jargon? Or plain language?
6. WARMTH: "We care about you" or "Users will find"?

Note 1 quote per dimension. That's your baseline.
Flag: "Quick scan only - full analysis needed for production guidelines"

References

For detailed examples and edge cases, see:

•references/voice-examples.md - Sample voice profiles from various industries
•references/voice-edge-cases.md - Handling difficult extraction scenarios