When copy drifts off-voice, build a brand voice document writers and AI assistants can apply consistently. — Claude Skill
Define how the brand sounds so every writer stays on-voice.
- Voice attributes paired with what they are NOT (confident, not arrogant) to prevent drift
- Tone shifts mapped across 8 to 15 contexts (onboarding, errors, pricing, cancellation)
- Vocabulary preferences plus grammar rules (contractions, pronouns, sentence length defaults)
- Paired-example library of 15 to 25 bad-vs-good copy samples writers actually use
- Stress-test step that re-runs a real brief against the doc before sign-off
Who this is for
Marketing leads ship a single voice doc that aligns 3 to 10 writers, freelancers, and AI assistants on how the brand sounds; cuts editor rework cycles per draft.
See skills for this roleContent marketers reference the paired-example library every time they draft a new piece; reduces voice-drift comments from editors.
See skills for this roleWhat it does
Marketing lead has zero documented voice; produces a 4-layer voice doc in one working session, distributed to 3 freelance writers next morning.
Catalogues 40 live copy samples across web, email, and product UX; identifies 12 off-voice patterns; rewrites 25 paired examples in one week.
Voice doc plugs directly into system prompts; reduces editor rewrite cycles from 3 to 1 on the next 20 drafts.
New freelancer reads voice attributes, tone-shift table, and paired examples; ships first on-voice draft same day instead of round 4.
How it works
Audit any existing copy across web, product, and email to map what already sounds on-brand versus off
Generate 5 to 8 voice attribute candidates with we-are-X-not-Y framing, then narrow to 3 to 5 keepers
Map 8 to 15 writing contexts (hero, errors, pricing, cancellation, 404) and document the tone shift for each
Lock vocabulary preferences, grammar defaults, and a paired-example library of 15 to 25 bad-vs-good rewrites
Stress-test the doc against a fresh writing brief; if output is off-voice, the doc gets a second pass before distribution
Example
Define voice for Rampstack, a developer-tools startup launching a self-serve SaaS. Audience: backend engineers at 20 to 200-person teams. We have 8 existing landing pages and 30 product copy strings that feel inconsistent. Reference brands: Linear, Stripe, Vercel.
Confident, not arrogant. Direct, not blunt. Practical, not dry. Curious, not unfocused.
Hero: signature voice fully on. Error message: calm, matter-of-fact, no apology theater. Cancellation flow: quiet, respectful, no jokes.
Use contractions. Sentence-case headlines. Active voice default. Banned word list lives in the doc; writers reference it before publishing.
Off: Welcome to the future of seamless deployment. On: Deploy on every push. No staging dance.
Applied doc to a fresh announcement brief; first draft cleared editor review in one pass.
Metrics this improves
Works with
Catalogue existing copy samples and track which fall on-voice versus off during the audit step.
Distribute the voice doc and run voice-question polls with writers and reviewers.
Mirror voice attributes inside design system copy components so UI strings inherit the doc.
Host the voice document so writers, designers, and AI prompts can reference it inline.
Want to use Brand Voice?
Choose how to get started.
Install and run this skill locally on your computer.
Open a terminal on your computer and paste this command:
This downloads the skill with all its files to your computer:
Add -g at the end to make it available in all your projects.
Start Claude Code, then type the command:
Brand Voice
Define how a brand sounds in writing. Document it in a way that anyone (writer, designer, founder, AI assistant) can apply it consistently.
This skill produces a standalone voice document that can either live in the brand style guide or feed into it.
When to use
- Defining brand voice for the first time
- Auditing existing copy for voice consistency
- Training a team, freelancer, or AI assistant on the brand voice
- Refining a voice that exists but is inconsistent across applications
- Adapting voice when a brand evolves (new audience, new positioning)
When NOT to use
- Writing specific copy (use
content-and-copy) - Brand visual identity work (use
brand-identity) - Documenting a complete brand system (use
brand-style-guide, which incorporates voice) - Initial brand exploration (use
brand-ideation)
Required inputs
- The brand and its positioning
- The audience and what they need to hear
- 5 to 10 examples of existing brand writing (if any)
- 3 to 5 reference brands whose voice resonates with where this brand should go
- Any voice attributes already specified in the brief
The framework: 4 layers
Voice has four layers, stacked. Each layer constrains the one below it.
Layer 1: Voice attributes
The constants. The personality traits that define how the brand sounds across every context.
Pick 3 to 5 attributes. Pair each with what it is NOT (the failure mode if overdone).
Common attribute pairings (NOT a menu - generate your own):
- Confident, not arrogant
- Direct, not blunt
- Warm, not saccharine
- Witty, not sarcastic
- Smart, not academic
- Honest, not harsh
- Playful, not silly
- Practical, not boring
- Bold, not loud
- Curious, not unfocused
The "not" half is what saves writers from overshooting. "Confident" alone produces swagger. "Confident, not arrogant" tells writers where the line is.
Layer 2: Tone shifts
Voice is constant. Tone adapts to context.
Map the major contexts the brand writes in. For each, document how voice expresses differently.
Common contexts:
| Context | Tone shift |
|---|---|
| Onboarding | Warmer, more enthusiastic, slightly slower pace |
| Hero / marketing | Confident, signature voice fully on |
| Product copy / UX | Direct, helpful, brief |
| Error messages | Calm, matter-of-fact, no apology theater |
| Success states | Brief celebration, redirect to next action |
| Empty states | Helpful, slightly playful, suggest action |
| 404 / not found | Self-aware, light, points the way home |
| Account deletion / cancellation | Quiet, respectful, no jokes |
| Pricing | Direct, transparent, confidence-inspiring |
| Legal / TOS | Plain language version sits next to the legal version |
| Support / help center | Patient, thorough, no condescension |
| Crisis communication | Calm, factual, accountable |
| Product announcements | Excited but not breathless |
| Email subject lines | Specific, never click-bait |
Voice stays consistent across all of these. Tone is what shifts.
Layer 3: Vocabulary and grammar
The granular dial settings.
Vocabulary preferences:
- Words and phrases the brand uses
- Words and phrases the brand avoids
- Words the brand has redefined (if any) - e.g., a SaaS product calling its features "huddles" instead of "meetings"
- Industry jargon: keep it (signals expertise) or strip it (signals approachability)
- Sentence-opening preferences (some brands lean on imperatives, some on questions, some on statements)
Grammar and style:
- Contractions (use them = casual, avoid them = formal)
- Sentence length default (short = punchy, medium = considered, long = literary)
- Punctuation marks favored or avoided (em dash is famously polarizing)
- Pronouns ("we" / "you" / "I")
- Capitalization style (title case, sentence case, all-lowercase deliberate)
- Number formatting (spell out under ten, or always digits)
- Oxford comma (use or skip)
- Active vs passive voice (most brands prefer active)
Layer 4: Examples and patterns
Voice is taught through examples, not rules. Build a library.
For each major content type, show:
- Bad example (off-voice)
- Good example (on-voice)
- Brief note on what changed
Common content types to cover:
- Headline
- Subheadline
- Hero CTA
- Feature description
- Testimonial intro
- Email subject line
- Email opening
- Push notification
- Error message
- Success message
- About page paragraph
- Social post
- Sales page paragraph
Aim for 15 to 25 paired examples. This is the most-used part of the voice doc in practice.
Workflow
- Audit existing copy if it exists. Identify what is on-brand, what is off, what patterns recur.
- Layer 1: Voice attributes. Generate 5 to 8 candidates with "we are X, not Y" framing. Pick 3 to 5.
- Layer 2: Tone shifts. List 8 to 15 contexts the brand writes in. Note the tone shift for each.
- Layer 3: Vocabulary and grammar. Define preferences. Skip default rules unless they actually distinguish the brand.
- Layer 4: Examples. Build the paired-example library. 15 to 25 minimum.
- Stress-test. Pick a fresh writing brief and apply the voice doc. Does it produce on-voice copy? If not, the doc is incomplete.
- Document. Use the template in
references/voice-document-template.md. - Distribute. Voice docs only work if they get used. Make the doc easy to reference inline (link to it from CMS templates, brief templates, AI assistant prompts).
Failure patterns
- Generic attributes ("friendly, professional, approachable"). Every brand says this. Means nothing. Pick attributes that genuinely distinguish.
- No "we are NOT" pairings. Without the rejection, attributes drift toward extremes.
- Voice doc with no examples. Rules without examples cannot be applied.
- Examples that are obviously bad and obviously good. Real voice work shows nuanced shifts, not cartoonish before/after.
- Skipping tone shifts. Treating voice as one-size-fits-all leads to a brand that sounds wrong in error states or legal contexts.
- Documenting aspirational voice. If the brand does not actually sound this way today and has no plan to shift, the doc is fiction.
- Voice without distribution. A perfect doc that no one references is worth nothing.
Output format
Default output is a markdown document at voice.md in the brand folder. Sections:
- Voice attributes (with we-are-not pairings)
- Tone shifts by context
- Vocabulary preferences
- Grammar and style rules
- Paired examples library (the most-used section)
- Anti-patterns (specific phrases or constructions to avoid)
- References (the brands and writers we are inspired by)
This doc can stand alone or feed into brand-style-guide.
Reference files
references/voice-document-template.md- Fillable template.references/voice-frameworks.md- Detailed walkthrough of the Nielsen Norman 4 dimensions, Jung archetypes, and the "we are X not Y" approach.