← Back

Brand voice examples and what makes them work

Three real brand voice examples, from Mailchimp's clarity rule to GOV.UK's banned-word list, and what makes each one specific enough to actually follow.

The most useful brand voice examples are not three adjectives on a slide. They are a real rule you can hold a sentence up against: a banned word with its replacement, a sentence-length limit, a line about where humor is allowed. Most pages that call themselves a brand voice guide skip straight to adjectives like "friendly" and "bold," which describe a feeling without telling a writer, or an AI tool, what to actually type.

Below are three public brand voice examples worth studying: Mailchimp's style guide, GOV.UK's content guide, and the IBM Carbon design system. Each ties its voice to something a stranger can check, and each gets at something different about doing that well.

Key Takeaways

  • Mailchimp's style guide states one priority over the rest: "it's always more important to be clear than entertaining," which settles arguments an adjective like "witty" never could.
  • GOV.UK's content guide names specific words to avoid, including "utilise" and "facilitate," with a plain-English replacement for each.
  • IBM's Carbon design system caps sentences at roughly 25 words, swaps "utilize" for "use," and lets tone shift between a direct error message and a friendlier onboarding screen while the underlying voice stays the same.
  • The brand voice examples that hold up share one trait: a word, a sentence rule, or a length limit a stranger can check, not an adjective to interpret.
  • TrueKeep keeps a brand's own word list and sentence rules in a design brief, connected to Claude, ChatGPT, and Cursor through an MCP connection, so an AI draft follows the same checkable rules a person would.

What a brand voice example actually shows

A brand voice example is only worth studying if it replaces a feeling with a test. "We sound confident and approachable" gives a new writer nothing to check a draft against. "Never use utilise; write use instead" gives them an answer in seconds, and it gives the same answer to an AI tool asked to draft a support reply.

The three examples below pass that test. None of them describe a personality for very long before they hand over something a stranger can apply. Brand voice guidelines frame this kind of rule the same way, as six working parts rather than a paragraph of adjectives, and that structure is what all three examples below slot into.

Mailchimp: a voice built on one priority, not a mood

Mailchimp describes its own voice as "an experienced and compassionate business partner," plainspoken and genuine, with dry humor that is "weird but not inappropriate, smart but not snobbish." That description alone would be just another mood board. What makes Mailchimp's style guide usable is the rule sitting underneath it: "it's always more important to be clear than entertaining."

That single priority settles the argument an adjective can't. A writer unsure whether a joke belongs in an error message doesn't need to guess at "weird but not inappropriate." They check it against clarity first, and the guide tells them to keep a straight face if they're not sure the joke lands.

GOV.UK: a replacement for every banned word

GOV.UK's content guide takes the opposite approach from Mailchimp's. Instead of one guiding priority, it lists specific words to avoid and names what to write instead:

utilise -> use
leverage -> use, or name the actual effect
empower -> allow, or give permission
facilitate -> run, or name the specific action
streamline -> simplify

The guide also flags metaphors like "drive," "hub," and "ring fencing" as vague once they're pulled out of their original field, and asks writers to name the real action instead. Two of the words in that table are also on TrueKeep's own banned list, for a related reason: a word that reads confident on a slide reads unspecific on the page, and a reader (or a model) can't act on a word it has to guess the meaning of.

IBM Carbon: voice stays, tone moves

IBM's Carbon design system separates two things that most brand voice examples blur together. Carbon's voice is "clear and logical, confident without boasting," expressed through concrete sentence rules: sentence case everywhere, "use" instead of "utilize," and sentences kept to roughly 25 words or fewer.

Tone is the part that moves. An error message stays "economical and direct," while an onboarding screen uses full sentences and friendlier language, and both still follow the same 25-word limit and the same sentence case. Voice is the rule that doesn't change page to page; tone is how that rule gets applied to what the reader needs in the moment.

What the best brand voice examples have in common

Three traits repeat across Mailchimp, GOV.UK, and IBM Carbon.

  • A rule beats an adjective. "Clear over entertaining," a banned-word list, a 25-word sentence cap: each one tells a stranger exactly what to do, where "friendly" or "bold" only tells them how it should feel once they're done.
  • Voice holds, tone moves. All three let formality shift with context (a joke in a newsletter, a direct line in an error message) without changing the underlying rule a writer is following.
  • The example is short enough to use. None of the three needs a chapter. A one-line priority, a word-swap table, and a sentence-length limit each fit on a page someone will actually read before they write.

A voice page that lists three adjectives and stops there, the version almost every brand starts with, fails on all three counts. It describes a mood nobody can check, gives tone and voice the same single rating, and still leaves a new writer, or a connected AI tool, guessing at the first sentence.

Turning a brand voice example into something your team and your AI tools can use

A word-swap table and a sentence-length rule help a new hire write their first draft. They are also exactly what an AI tool needs before it writes anything on a brand's behalf, since a model has no default voice of its own to fall back on.

TrueKeep calls this written-down version of a brand's voice a design brief: the word list, the sentence rules, and the priority a brand follows when clarity and personality pull in different directions. Connect Claude, ChatGPT, or Cursor once through the AI connection guide, and the design brief reads the same way to the model that it does to a new writer. The brand kit skill carries the same brief into Claude directly, for people who draft there most of the day.

The tradeoff is that a design brief only helps once someone has actually finished it. A brief that still says "friendly and professional" hands a connected AI tool the same vague mood it would hand a new hire, and the fix is the same in both cases: replace the mood with a word list and a rule.

FAQ

What is a brand voice example? A brand voice example is a real company's stated voice along with something a stranger can check a sentence against: a banned word and its replacement, a sentence-length rule, or a single priority like "clear over entertaining." Mailchimp, GOV.UK, and IBM's Carbon design system are three public ones.

What makes a brand voice example good instead of generic? A good example hands over a rule, not a mood. "Never utilise, write use" tells a writer what to do. "Sound confident and approachable" tells them how it should feel once they've already guessed right.

Should a brand voice guide list adjectives or rules? Rules, with at most one or two adjectives as the posture underneath them. GOV.UK and IBM Carbon barely use adjectives at all; both lead with a word list or a sentence rule instead.

How is brand voice different from tone in these examples? Voice is the rule that stays fixed, like IBM Carbon's 25-word sentence limit or GOV.UK's banned-word list. Tone is how that rule gets applied in the moment, like a direct error message versus a friendlier onboarding screen.

How do I give an AI tool my own brand voice example? Put the word list and sentence rules somewhere the model reads before it drafts, rather than retyping them into every conversation. TrueKeep keeps that version in a design brief, connected to Claude, ChatGPT, and Cursor through an MCP connection.

Do I need a public example before writing my own brand voice guide? No, but studying one helps. Mailchimp, GOV.UK, and IBM Carbon show three different ways to turn a voice into something checkable: one priority, one banned-word list, and one sentence rule. Any of the three is a better starting point than three adjectives.

The short version

The brand voice examples worth studying share a shape, not a size. Mailchimp settles its arguments with one priority: clarity over entertainment. GOV.UK pairs every banned word with a plain replacement. IBM's Carbon design system keeps voice fixed at a 25-word sentence limit and lets tone move underneath it. None of the three needed more than a page to be useful.

Write your own the same way: one priority or word list a stranger can check, applied consistently, with tone doing the work of shifting by context. Then keep it where your team and their AI tools can both read it. If you want your brand voice rules in a portal connected to Claude, ChatGPT, and Cursor, get started with TrueKeep.