Voice is who a brand is; tone is how that identity adjusts to a specific moment — a brand voice guide that only lists adjectives ("friendly," "bold," "innovative") gives no one anything usable, because every brand claims the same three words. A usable guide defines what those words actually mean in sentences, with concrete examples of doing it right and doing it wrong.

Defining voice traits

Pick 3-4 voice traits maximum — more than that stops being a decision and starts being a wish list nobody can apply consistently. Fewer, sharply defined traits beat a long vague list every time.
For each trait, write a one-sentence definition specific enough to rule things out, not just in. "Direct" should specify what gets cut (hedging, corporate qualifiers) as much as what gets kept.
Pair every trait with a do/don't example pair, using real sentence-length examples, not single words: show the trait applied to one actual line of copy, and show the same line written the wrong way. Abstract adjectives don't transfer; example pairs do.
Anchor each trait against what the brand is not — "confident, not arrogant," "warm, not cutesy" — the contrast is often more useful than the trait alone, because it's usually the overcorrection that trips people up.

Formality calibration

Place the brand on an explicit formality scale (e.g. 1-5, from legal-document formal to texting-a-friend casual) rather than leaving "professional but approachable" to interpretation — that phrase means something different to every writer who reads it.
Specify concrete markers at the chosen formality level: contractions allowed or not, sentence fragments allowed or not, exclamation points allowed and how often, first person singular vs. plural ("I" vs. "we").
Note where formality is allowed to shift by channel (a support ticket reply can be warmer/more casual than a legal terms page) and where it must not (never sarcastic in anything involving money, safety, or a customer complaint).

Consistency across channels

Apply the same core voice traits across every channel — website, email, support replies, social, error messages — a brand that sounds like a different company in each surface undermines the reader's trust in all of them.
Adjust length and format per channel without abandoning the voice: a tweet-length version of a trait is still recognizably the same voice, just compressed, not a different personality entirely.
Give every channel owner (support, marketing, product) the same guide rather than separate ones per team — voice drift usually starts when teams write their own local interpretation of "friendly."

Voice vs. tone

Voice stays constant; tone shifts with context. The same brand voice sounds different congratulating a user on a milestone versus responding to a service outage — the traits don't change, but which ones lead does.
Define tone shifts explicitly for foreseeable high-stakes contexts: apologies, outages, price increases, security incidents. Specify what's dialed down (humor, exclamation points) and what stays (directness, taking responsibility) in each case.
Never let "tone can shift" become an excuse to abandon voice traits under pressure — a brand that's normally playful still shouldn't joke during an outage, but it can still be direct and human about it rather than switching to generic corporate-speak.

Banned phrases

Maintain an explicit list of phrases the brand never uses — corporate clichés ("synergy," "best-in-class," "at the end of the day"), overused marketing intensifiers ("revolutionary," "game-changing," "seamless"), and anything that contradicts a defined trait (e.g. "Oops!" banned for a brand whose trait is "serious, not cutesy").
Update the banned list when a phrase starts appearing across the industry generically enough to feel hollow, even if it was fine when the guide was written — voice guides need periodic maintenance, not just initial drafting.
Apply the banned list as a hard rule, not a preference — a writer under deadline pressure defaults to clichés unless told explicitly they're off-limits.

Applying the guide to new copy

Before publishing, check new copy against each defined trait individually — does this sentence sound "direct"? Does it sound "warm"? A piece that passes on average but fails badly on one trait still reads off-brand.
When two traits pull in different directions on a specific line (e.g. "playful" vs. "precise" in a technical explanation), default to whichever trait the guide has explicitly ranked higher for that content type.
Test unfamiliar copy by reading it aloud — voice inconsistencies that look fine on the page are often audible immediately when spoken.

Tone of Voice Guide