brand-voice-codifier

Brand Voice Codifier

Consistent voice is what separates a brand from a marketing template. This skill takes 3-5 samples of a brand's existing writing and produces a reusable Voice Profile artifact that other copy-generating skills can load as context.

When to Use This Skill

Load this skill when the user says any of:

• "match my voice" / "sound like me" / "make it sound like our brand"
• "codify my brand voice" / "build me a voice profile" / "style guide"
• Before launching any copy-generating skill (email-sequence-writer, sales-page-long-form, landing-page-builder, weekly-content-planner) for a brand that doesn't yet have a voice profile file

The Doctrine (Read First)

Before doing anything, read assets/guides/voice-analysis-doctrine.md. It codifies:

• The 8 dimensions of a voice profile (vocabulary, sentence structure, POV, tense, tone, signature moves, banned words, capitalization/punctuation habits)
• What "distinctive" means (a voice trait must be repeated across samples AND unusual enough to be recognizable)
• How to weight samples (weight the highest-performing pieces heavier)
• The Kennedy specificity check (voice profile itself must be specific — no "friendly but professional" fluff)

Then read assets/guides/sample-collection-guide.md for what samples to request.

Workflow

Step 1: Collect samples

Ask the operator for 3-5 samples. The mix matters:

• 1 high-performing sales piece (page or email that converted well)
• 1 educational piece (blog, LinkedIn post, YouTube script)
• 1 quick-hit piece (short-form: DM, comment, X post, cold email)
• Optional: 1 unstructured piece (voice memo transcript, personal note)

Ask about performance context for each sample — "which of these performed best?"

Step 2: Analyze each sample independently

For each sample, run through assets/guides/voice-analysis-doctrine.md and note:

• Vocabulary tendencies (favored words, industry jargon usage, formality level)
• Sentence-length distribution (short/med/long ratio)
• POV (first person singular / plural / second person / third person)
• Tense (present/past/future dominant)
• Tone markers (self-deprecating / confident / contrarian / empathetic / etc.)
• Signature moves (repeated structural patterns, unique openings, unique closers)
• Punctuation habits (em-dashes, ellipses, ALL CAPS, single-word sentences)
• Capitalization habits (Title Case in headers, lowercase intentional, etc.)
• Any words that appear frequently that most brands don't use
• Any words that are conspicuously absent

Write findings for each sample in a scratch file.

Step 3: Cross-sample synthesis

Compare findings across all samples. A voice trait qualifies for the profile only if:

• It appears in ≥ 60% of samples (repeated pattern, not one-off)
• It's distinctive enough that a reader would notice its absence

Discard one-off quirks. Keep repeated distinctive patterns.

Step 4: Fill the voice profile

Fill in assets/templates/voice-profile.md. This is the output artifact. Structure:

1. Brand + author + purpose — metadata
2. The elevator description — 2-3 sentences describing the voice
3. Vocabulary — favored words (with examples), banned words (with alternatives), industry-jargon rules
4. Sentence structure — length distribution, favored sentence constructions
5. POV + tense — default settings
6. Signature moves — repeated structural patterns with examples
7. Formatting habits — em-dashes, ALL CAPS, single-word sentences, list preferences
8. Tone rules — what the voice IS and what it is NOT
9. Do / Don't matrix — 10 pairs of "do this / not this" comparisons with examples
10. Sample transformation — one "generic AI copy" example rewritten in the voice

Step 5: Kennedy specificity pass on the profile itself

The voice profile MUST be specific. Bad voice profile:

"Friendly but professional. Uses conversational language."

Good voice profile:

"Uses em-dashes 2-3 times per paragraph as pauses. Opens every LinkedIn post with a single-word sentence. Never uses the word 'passionate.' Prefers 'shipped' over 'launched.' Sentence-length distribution: 40% short (<10 words), 45% medium (10-20), 15% long (>20). Never uses semicolons."

Step 6: Validate with a transformation

Take one piece of generic marketing copy (provided in the template) and rewrite it in the voice using ONLY the profile. If the rewrite reads like the brand — profile is good. If it still reads generic — the profile is missing distinguishing traits, go back to step 2.

Step 7: Save and hand off

Save the completed profile as [brand-slug]-voice-profile.md in a canonical location the operator can reference in future skill invocations.

Downstream skills that read the voice profile:

• email-sequence-writer — every email loads this profile
• sales-page-long-form — the sales letter follows this voice
• landing-page-builder — landing page copy follows this voice
• weekly-content-planner — content plan follows this voice
• cold-email-campaign-builder — cold email templates follow this voice
• review-response-generator — responses to bad reviews follow this voice
• case-study-builder — case study copy follows this voice

Multi-Brand Setup

Users operating multiple brands (like Joey Zoccali — Focal Point Coaching, Fieldmesh, Level Up, Clearwater Notary) should have one voice profile per brand. Voice profiles should be stored in a predictable path so any skill can reference them:

voice-profiles/
├── focal-point-voice-profile.md
├── fieldmesh-voice-profile.md
├── level-up-voice-profile.md
└── clearwater-notary-voice-profile.md

Each profile is independent. Do not merge brands.

Filled Example

See assets/examples/filled-example.md — the completed Focal Point Coaching voice profile derived from 4 samples (sales page, LinkedIn post, cold email, video script).

Related Skills

Every copy-generating skill reads a voice profile as input:

• email-sequence-writer, sales-page-long-form, landing-page-builder, weekly-content-planner, cold-email-campaign-builder, review-response-generator, case-study-builder, lead-magnet-builder, landing-page-copy-refresher

Banned Words (in the voice profile itself)

The voice profile artifact must not contain vague adjectives like:

• "professional yet friendly"
• "authentic"
• "authoritative"
• "engaging"
• "conversational" (without specifics)
• "authentic voice"

Every trait must be observable and reproducible.