Brand Voice

Voice is the brand's personality in words, and it stays constant. Tone is how that voice adjusts to the moment: a person reading an error is in a different state than a person reading a launch announcement. Mailchimp's content style guide made this split standard, and this skill uses it. The output is a voice guide that a new writer, a product designer, or a support agent can use to write something on-voice the first time. The standard: every attribute comes with rewrites that show it, and two writers given the same brief produce copy that sounds like the same company.

When to use

• A new brand, or a brand platform that exists without a voice layer
• Copy across product, marketing, support, and email sounds like different companies
• Hiring writers or scaling content, and needing a shared reference
• Deciding how formal, funny, or technical the brand can be

Not for: writing or reviewing specific interface strings, error messages, or empty states (use ux-writing), or the brand's positioning and attributes (use brand-strategy).

Inputs

• Brand attributes from the brand platform [if missing, infer 3–4 from the product and state the guess; strategy and voice can be drafted together for small teams]
• Audience: who reads this, their expertise, their state when they read [from the platform]
• Existing copy: 20–40 samples across site, product UI, emails, support macros, legal [collect them; this is the audit]
• Customer language: verbatim phrases from interviews, reviews, tickets [from user-research, or mine public reviews]
• Constraints: regulated claims, localization languages, reading level [assume plain English at grade 8 or lower for product copy]

Process

1. Audit existing copy. Collect 20–40 samples and sort them into on-voice, off-voice, and "no voice" (generic). Output: a table with each sample, context, and what is wrong or right.
2. Place the brand on NN/g's four tone dimensions (Kate Moran, 2016): funny ↔ serious, formal ↔ casual, respectful ↔ irreverent, enthusiastic ↔ matter-of-fact. Output: a position on each, with the reason from the attributes.
3. Write 3–4 voice attributes. Each is a "this, not that" pair with a paragraph, do/don't examples, and one before/after. Model: the attribute block in templates/voice-guide.md. Copy that shape. Output: attribute section.
4. Build the tone map. For each context, the reader's likely state, the tone adjustment, and an example. Model: the tone map table in the template. Output: tone map for at least onboarding, empty states, errors, success, marketing, transactional email, support, and legal.
5. Write word lists. "We say / we don't say" with a reason per row, plus product terms with definitions and capitalization. Output: word list and glossary.
6. Set mechanics. Casing, punctuation, numbers, dates, contractions, emoji policy, inclusive language. Output: a short mechanics section; defer anything detailed to the product's UX writing style guide.
7. Rewrite real samples. Take 8–10 off-voice samples from the audit and rewrite them. Output: before/after pairs with a line on what changed.
8. Test with writers. Give the guide and the same three briefs to two people who were not involved. Compare. Output: revisions where their results diverged.

Standards

Attributes

• 3–4 attributes. Each one must change a sentence. If removing an attribute would not change any rewrite, cut it.
• Pair each with the failure it slides into when overdone:

Tone map

Voice stays fixed; tone moves along the dimensions from step 2, within a set range.

Humor never appears in errors, payments, security, or anything involving loss.

Word lists

• 15–30 rows of "we say / we don't say", each with a reason. Pull the "we say" column from customer language wherever possible.
• A glossary of product terms: one term per concept, with capitalization. Two words for one thing ("workspace" and "team") is the most common cause of confusion.

Mechanics (defaults; adjust and write down)

• Sentence case for headings and buttons.
• Contractions yes. Exclamation marks: at most one per screen, never in errors.
• Numbers as numerals ("3 files"). Dates unambiguous ("Mar 4, 2026", not "3/4/26").
• "You" for the reader; "we" for the company, and only when the company acted.
• Emoji: decide once. Default: none in product UI.
• Grade 8 or lower reading level for product copy (Hemingway or Flesch-Kincaid check).

How voice feeds ux-writing

The voice guide sets personality and tone ranges; ux-writing applies them to specific strings with patterns (error = what happened + why + what to do). Put the tone map and word list where UX writers work: the design system docs and the string review checklist. Voice is not a license to add personality to every string: in UI, clarity wins every conflict, and voice shows mostly in word choice and in the few moments with room for it (onboarding, empty states, success).

Output

A voice guide in templates/voice-guide.md: summary, tone dimensions, attributes with do/don't and rewrites, tone map, word list and glossary, mechanics, and a rewrite gallery. Also a one-page cheat sheet (attributes, the tone map row for errors, top 10 words) for people who will not read the full guide.

Verify

The guide is done when two writers who were not involved have used it on the same three briefs (one error, one onboarding screen, one marketing paragraph) and their drafts read as the same company, and you have read every sample in it aloud. Then confirm:

• Each attribute has a "not that", at least two do/don't pairs, and one rewrite
• The tone map covers onboarding, empty, error, success, marketing, email, support, and legal
• Errors and legal contexts contain no humor
• Every word-list row has a reason, and the glossary has one term per concept
• Rewrites come from real audit samples, not invented examples
• Product copy samples score at or below grade 8

Delegate to the copy-reviewer subagent to check sample strings and the rewrite gallery against the guide, and to the brand-reviewer subagent to check the voice against the brand platform's attributes. Both report only failures with location and end in a verdict ("ready" or "N blocking issues").

Anti-patterns

• Attributes like "human", "friendly", "clear" with no "not that" and no example
• Voice guides that only show marketing copy, so nobody knows how an error should sound
• Jokes in error messages, payment failures, or outage notices
• Forced personality in every button and tooltip ("Let's gooo!")
• Word lists that ban words without saying what to use instead
• Two terms for one concept across product and docs
• A guide that describes the voice in abstractions without a single before/after

Related skills

• Feeds from: brand-strategy (attributes and audience), user-research (customer language)
• Leads to: ux-writing (string-level application), landing-page-design, design-system-docs (content guidelines), art-direction (headlines in campaigns)

References

From REFERENCE-BANK.md:

• Nielsen Norman Group: "The Four Dimensions of Tone of Voice" for step 2, and tone research on user perception of casual versus formal copy.
• Shopify Polaris content guidelines: voice and tone, word list, and grammar rules organized for product teams; a model for structure.
• Atlassian Design content design section: voice principles with product examples.
• GitHub Primer: content guidelines tied to components.
• Apple HIG writing section: restraint and platform conventions.
• Brand New: how a verbal identity accompanies a visual rebrand.

From FIBO.md: headings and UI copy in sentence case, and each component docs page's "Usage" rule stated in one sentence, as a model for how voice rules should read: one rule, one example.