# Brand Voice Extraction

> Extract a brand voice guide from existing copy samples. Use whenever the user shares past copy and asks to define a voice, document a voice, create voice guidelines, or wants to make their existing voice teachable and repeatable.

- Skill: `dragoon0x/brand-voice-extraction` (Agent Skill)
- Install (CLI): `npx skillmds@latest add dragoon0x/brand-voice-extraction`
- Raw SKILL.md: https://api.skillmd.com/api/skills/dragoon0x/brand-voice-extraction/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: dragoon0x (https://skillmd.com/u/dragoon0x)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/dragoon0x/brand-voice-extraction

---


# brand-voice-extraction

most teams have a voice. fewer teams have documented it. this skill turns 5-20 samples of existing copy into a voice guide that someone else can write in.

## inputs

- 5-20 pieces of existing copy that the team agrees represents the voice well
- 2-5 pieces that the team agrees are off-voice (very useful, often skipped)
- optional: notes on the audience

## the extraction pass

read everything before making claims. then go through these dimensions:

### sentence rhythm

- average sentence length (count words, get a sense of the spread)
- shortest and longest sentences
- variation pattern (does the voice mix short and long, or does it stay in a band)

### vocabulary range

- formal / casual / playful / clinical / academic
- specific words that appear repeatedly (these become signature vocabulary)
- words that should not appear (often visible by their absence)

### point of view

- first person plural (we, our)
- first person singular (i)
- second person (you, your)
- third person (the company, customers)

note when the voice switches between these and why.

### figurative range

- metaphors used, what domains they pull from
- analogies and how often
- humor: present? what kind? (dry / absurd / observational / self-deprecating)

### structural patterns

- how openings work (questions / claims / scene-setting)
- how arguments are built (story-first / claim-first / data-first)
- how endings work (invitation / claim / quiet finish)

### what's deliberately not there

- adverbs that add no information
- corporate filler ("delve", "leverage", "robust", "elevate", "seamless")
- exclamation points
- "we listened to your feedback"
- buzzwords specific to the industry

## the voice guide structure

```
# [brand] voice guide

## the voice in one paragraph
how the brand sounds to a reader who's never heard of it before.

## the principles
3-5 short statements. each one a thing the voice does or doesn't do.
example: "we explain, we don't decorate" or "we never sound impressed with ourselves".

## sentence rhythm
average length: [N] words.
range: [shortest, longest].
pattern: [how length varies].

## vocabulary
register: [casual / professional / etc].
signature words: [list of 10-20].
banned words: [list, with the reason].

## point of view
default: [first person plural / second person / etc].
when it shifts: [rules].

## structural patterns
openings often: [pattern].
arguments often: [pattern].
endings often: [pattern].

## what's not there
[list of things the voice deliberately avoids].

## examples

### yes
[3-5 short excerpts of existing copy that nails the voice]

### no
[3-5 short excerpts that miss the voice, with notes on why]

## the test
a piece of writing matches the voice if a reader couldn't tell whether it came from the original samples or this new piece.
```

## what not to do

- describe the voice with adjectives only ("friendly, professional, approachable" tells you nothing)
- invent rules that aren't supported by the samples
- write the voice you wish you had instead of the one you have
- skip the "no" examples (they often teach more than the yes examples)