brand-voice-extractor（品牌语音提取器）

Brand Voice Extractor

Analyze a company's published content to extract their brand voice and writing style. Reads their top content pieces and produces actionable guidelines for matching their voice in future content, outreach, or campaigns.

Quick Start

Extract brand voice for [company]. Use their blog at [url].

Or with content already cataloged:

Extract brand voice for [client]. Use the content inventory at clients/[client]/research/content-inventory.json.

Inputs

Process

Phase 1: Select Content to Analyze

If content URLs are provided directly, use those. Otherwise:

1. Read the content inventory from site-content-catalog output
2. Select a diverse sample of 10-20 pages, prioritizing:

- Blog posts (primary voice indicator) - Landing pages (marketing voice) - Case studies (storytelling voice) - Mix of recent and older content (to detect voice evolution) - Mix of topics (to see consistency across subjects)

Selection heuristic:

• 8-10 blog posts (mix of how-to, opinion, product updates)
• 2-3 landing pages (homepage, product page, solutions page)
• 2-3 case studies or customer stories (if available)
• 1-2 comparison/vs pages (if available)

Phase 2: Fetch and Extract Text

For each selected URL:

1. WebFetch the page
2. Extract the main content body (strip nav, footer, sidebar)
3. Store: title, URL, raw text, word count

Phase 3: Analyze Voice Dimensions

Analyze across these dimensions:

A) Tone

• Formality spectrum: Casual ↔ Professional ↔ Academic
• Emotional register: Excited ↔ Measured ↔ Dry
• Authority stance: Peer/friend ↔ Expert/teacher ↔ Institution
• Humor usage: Frequent ↔ Occasional ↔ None
• Directness: Direct/bold ↔ Hedged/diplomatic

B) Vocabulary & Language

• Reading level: Approximate grade level (simple vs. complex)
• Jargon usage: Heavy industry jargon ↔ Plain language
• Technical depth: Assumes expertise ↔ Explains everything
• Power words: Common persuasion/action words they favor
• Banned patterns: Words or phrases they conspicuously avoid
• Unique vocabulary: Distinctive terms or phrases they use repeatedly

C) Sentence Structure

• Average sentence length: Short/punchy ↔ Long/complex
• Paragraph length: 1-2 sentences ↔ 3-4 ↔ 5+
• Opening patterns: How they start articles (question, stat, story, bold claim)
• Transition style: How they connect ideas
• Use of fragments: Do they use incomplete sentences for emphasis?

D) Formatting Patterns

• Headers: Frequency, style (question-based, how-to, numbered)
• Lists: Bullets vs. numbered, frequency
• Bold/italic: How they use emphasis
• Images/media: Frequency, types (screenshots, illustrations, photos)
• CTAs: Placement, style, frequency, language used
• Pull quotes/callouts: Do they use them?

E) Content Structure

• Typical article length: Short (<800), Medium (800-1500), Long (1500+)
• Introduction style: Hook type, length
• Conclusion style: Summary, CTA, open question
• Use of data/stats: Frequent ↔ Rare
• Use of examples: Frequent ↔ Rare
• Storytelling: Narrative-driven ↔ Information-driven

F) Persona & Audience

• Who they write for: Inferred target reader (role, seniority, industry)
• Assumed knowledge level: Beginner ↔ Intermediate ↔ Expert
• Point of view: First person singular (I) ↔ First person plural (we) ↔ Second person (you) ↔ Third person
• Reader relationship: Peer ↔ Teacher ↔ Service provider

Phase 4: Generate Brand Voice Profile

Produce a Markdown document with this structure:

# Brand Voice Profile: [Company Name]
**Analyzed:** [Date] | **Content pieces analyzed:** [N]
**Sources:** [list of URLs analyzed]

---

## Voice Summary (2-3 sentences)

[Company] writes in a [tone] voice that [description]. Their content targets
[audience] and assumes [knowledge level]. The overall feel is [adjectives].

---

## Tone Profile

| Dimension | Position | Evidence |
|-----------|----------|----------|
| Formality | [e.g., Professional-casual] | [Example quote] |
| Emotional Register | [e.g., Measured, occasionally excited] | [Example] |
| Authority | [e.g., Expert/teacher] | [Example] |
| Humor | [e.g., Rare, dry when used] | [Example] |
| Directness | [e.g., Very direct, bold claims] | [Example] |

---

## Language & Vocabulary

### Reading Level
[Grade level estimate and what that means]

### Signature Phrases
- "[phrase 1]" — used frequently to [purpose]
- "[phrase 2]" — recurring pattern in [context]

### Jargon & Technical Depth
[How much industry jargon they use, how they handle technical concepts]

### Words They Love
[List of frequently used power words, adjectives, verbs]

### Words They Avoid
[Notable absences or patterns they steer away from]

---

## Structure & Formatting

### Typical Article Structure
[Outline of how their articles are typically organized]

### Sentence & Paragraph Style
- Average sentence length: [X words]
- Typical paragraph: [X sentences]
- Notable patterns: [fragments, rhetorical questions, etc.]

### Formatting Habits
- Headers: [style]
- Lists: [frequency and style]
- Emphasis: [bold/italic patterns]
- CTAs: [where, how often, what language]

---

## Audience & Persona

### Target Reader
[Role, seniority, industry, pain points they address]

### Knowledge Assumptions
[What they assume the reader already knows]

### Point of View
[I/we/you usage and what it signals]

---

## Writing Guidelines (Actionable)

Use these guidelines when writing content, outreach, or campaigns for [Company]:

### Do
- [Guideline 1 with example]
- [Guideline 2 with example]
- [Guideline 3 with example]

### Don't
- [Anti-pattern 1]
- [Anti-pattern 2]
- [Anti-pattern 3]

### Voice Samples

**Their style:**
> [2-3 representative quotes from their content that exemplify the voice]

**How to match it:**
> [2-3 example sentences written in their voice about a neutral topic]

Tips

• 15 pages is the sweet spot. Fewer than 10 won't capture enough variation. More than 25 adds cost without much signal.
• Blog posts are the best voice signal. Landing pages are more formulaic. Blog posts show the authentic voice.
• Look for consistency AND inconsistency. If their tone shifts dramatically between content types, note it — they may have multiple voice modes.
• Check for ghost-written content. If some posts feel dramatically different, they may use external writers. Flag this in the analysis.
• This skill has no code script. It's an agent-executed skill — the AI agent reads the content via WebFetch and performs the analysis directly. The structured output template above guides the analysis.

Dependencies

• Web fetch capability (for reading content pages)
• Optional: site-content-catalog output (for selecting which content to analyze)
• No API keys or paid tools required