Rules, Not Adjectives
Specific enough that two writers make the same choice.
A voice described as "friendly but professional" tells a writer nothing, which is why organisations with documented voices still sound like whoever wrote each piece. Brand voice has to be specific enough to act on — actual rules, actual examples, and the words you use and avoid.
Adjectives describing a voice do not help someone writing a sentence.
Specific enough that two writers make the same choice.
The same sentence written wrongly and rightly.
Terms you use and terms you do not.
An error message and a case study need different registers.
Applied to existing copy to see whether it changes anything.
Voice guidance fails in a predictable way: it describes a personality using adjectives every company would claim, then leaves writers no better off than before. These checks all test whether the guidance actually changes what somebody writes.
Paired examples do more work than every adjective combined. "Friendly but professional" tells a writer nothing; two versions of the same sentence with a note on why one is right teaches the pattern in seconds.
Four things, and none of them is a list of adjectives.
Discuss Your Voice →Sentence length, contractions, person, formality.
This, not that. Concrete pairs rather than descriptions.
Words to use and words to avoid, listed.
How the voice shifts between a homepage and an error message.
A definition writers can apply without asking.
What is said is covered by brand messaging. What is actually said is brand messaging; interface and error copy where voice matters most sits with UI design. What is actually said is brand messaging; interface and error copy where voice matters most sits with UI design.
Look at what you already write, then make the good parts into rules.
Across every channel, including support replies.
The writing that already sounds right.
Concrete enough to apply.
Before and after, from your own copy.
So it survives being handed to other writers.
Voice work matters in proportion to how many people write on the brand’s behalf and how often. A single founder writing everything already has a voice; the question is whether it is documented before anyone else has to reproduce it.
Where blog posts, emails, social copy and product text are produced weekly by several people, voice is the only thing preventing the output from reading as several unrelated companies. This is where the investment repays fastest, because the inconsistency is already visible.
External writers are competent and have no access to the brand’s instincts. Without paired examples they default to a generic professional register, and every piece needs heavy editing. Documented voice converts editing time into briefing time, which is cheaper and happens once.
The voice that matters most is the one used when something has gone wrong, and it is the one guidance almost never covers. An apology written in marketing voice reads as evasive. Deciding in advance how the brand speaks under pressure is more valuable than deciding how it speaks in a headline.
Where mandatory language and required disclosures constrain how much personality is available. Voice work here is largely about finding the room that exists inside the constraints rather than pretending they are absent, which matters for law firm branding and healthcare in particular.
Interface copy, empty states, error messages and confirmations are read by more people than any marketing page, and they are usually written by whoever built the feature. Extending voice into the product is often the highest-leverage part of the work and the part most often left out of scope.
It describes an intention without providing a single decision a writer could make.
Use contractions. Address the reader as "you". Keep sentences under twenty-five words where possible. Do not begin with a subordinate clause. Never use "leverage" as a verb.
Each of those settles a decision. A writer facing a sentence can apply them, and two writers applying them produce compatible output.
"Friendly but professional" settles nothing. Every writer already believes they are being both, which is precisely why the results diverge.
Because it is read with more attention than anything else and is almost never covered by voice guidelines. It gets written by whoever built the feature, in whatever register they use.
It is also where tone does the most work. A failure message that is breezy reads as flippant; one that is cold reads as blame; one that explains what to do reads as competent.
Extending the voice definition to cover interface and error copy is usually the highest-value part of the exercise, and the part that gets left out.
The standard voice document lists three or four adjectives — approachable, confident, human — and stops. Every one of those words fails the same test: no company would choose the opposite. Nobody writes a voice guide describing themselves as distant, hesitant and robotic, which means the adjectives distinguish nothing and give a writer no basis for choosing between two drafts.
What changes behaviour is demonstration. A sentence written two ways, with a note explaining why one is on-voice, teaches more in ten seconds than a page of description. Writers learn voice by pattern-matching, the same way they learned to write at all, and patterns require examples rather than definitions.
The second reason these documents fail is that they omit decisions. Whether the brand uses contractions, whether it addresses the reader as you, how long sentences typically run, whether it starts sentences with conjunctions — these are the choices that actually produce the texture of a voice, and leaving them unstated means every writer settles them differently.
Most voice guidance is written for good news. It covers the headline, the about page, the launch announcement. Then something breaks, a customer is angry, a payment fails, and the guidance is silent — so the writer falls back on either corporate deflection or over-familiar cheerfulness, both of which damage trust at the exact moment trust is fragile.
A voice that has been thought through in advance handles this differently. It knows that humour is out, that hedging reads as evasion, that the customer needs to know what happened, what is being done and when to expect resolution, in that order. None of that is a personality decision; it is a decision about what the brand owes someone who is having a bad experience with it.
The most useful section of a voice document is often the shortest: how the brand sounds when apologising. It gets used more than anyone expects, and it is where the gap between brands that feel trustworthy and brands that feel slippery is most visible to customers.
Voice, messaging and visual identity are three answers to different questions, and confusing them causes most of the friction in brand projects. Messaging is what the business claims. Voice is how it sounds while claiming it. Visual identity is what it looks like doing so. A business can get one right and the other two wrong, and the inconsistency is usually more noticeable to customers than any individual weakness.
The most common mismatch is a premium visual identity paired with copy written in the flattest available register. The design promises considered craft, the words deliver a category description, and the gap between them reads as a brand that hired a designer rather than a brand that decided anything.
Voice also constrains design more than it is usually credited for. A voice built on directness does not sit comfortably in a layout that buries the point below three decorative sections. This is why voice, when it exists before a website redesign, tends to improve the design rather than merely fill it.







Brand voice is how an organisation sounds in writing, defined by concrete rules, examples and vocabulary rather than by adjectives — so different writers produce consistent output.
Because they describe intentions rather than decisions. Every writer already believes they are being friendly and professional, which is why results diverge despite the guidance.
That it settles a choice. "Use contractions", "address the reader as you", "avoid this word" — each of those makes two writers produce compatible sentences.
Yes, and it usually does not. Error copy is read with more attention than anything else and gets written by whoever built the feature, in whatever register they happen to use.
Messaging decides what is said. Voice decides how it sounds. The same claim can be written in several registers and the choice should not be accidental.
Voice is constant and tone varies with context. The voice is the brand’s underlying character — it does not change between a launch announcement and a service outage. Tone is the adjustment made for the situation: the same voice is warmer in a welcome email and more restrained in an incident notice. Useful guidance defines the voice once and then shows tone shifting across several genuinely different situations.
It can, if it is written in generic adjectives, because generic inputs produce generic output. Guidance built on specific vocabulary choices, settled grammar decisions and paired examples does the opposite — it makes writing more distinctive, because it commits to choices most competitors never make. The risk is real and it comes from vagueness rather than from documentation.
Whoever edits the most published writing, which is not always the most senior person available. Ownership means keeping the examples current as the business changes and settling the arguments about edge cases. A document with no owner accumulates exceptions until nobody is sure what the rule was, and the same drift affects brand guidelines for exactly the same reason.
It applies more, not less. A language model asked to write for a brand will default to the average of everything it has read unless it is given specific constraints, and the average is precisely the flat professional register most brands are trying to escape. A voice document with paired examples and a vocabulary list is directly usable as that constraint — but the output still needs a human check, because the model has no way to know whether a claim it produced is true.
By rewriting real assets rather than distributing a document. A team shown its own homepage and its own three most recent emails rewritten in voice understands the change immediately, because the comparison is concrete and about work they recognise. A team handed a PDF of adjectives will agree with it and keep writing exactly as before, since nothing in it told them what to do differently.
Still deciding if brand voice is right for you?
Talk to UsVoice guidelines almost always land on a pair of adjectives with a "but" between them. Friendly but professional. Confident but approachable. Expert but accessible.
The formulation is popular because it sounds balanced and because nobody can object to it. It also gives a writer nothing to act on — they were already trying to be both, and their idea of the balance is their own.
So four writers produce four voices, all of which their authors would describe as friendly but professional, and the organisation concludes that voice guidelines do not work.
Send us copy from your website, your product and your support replies. We will tell you how consistent they actually are.
