The Tone of Voice Document That Actually Changes Copy

Page nine of the brand book says the voice is confident, human and bold. Everybody nods. Six months later the homepage still opens with "We are a leading provider of end-to-end solutions", the error messages still apologise for any inconvenience caused, and the onboarding email still begins with how excited everyone is. Nothing changed, because nothing in that document could have changed anything.

Adjectives are not instructions. A tone of voice document earns its place only if a writer can hold a real sentence against it and know whether the sentence stays or goes. Confident does not do that. Human does not do that. Every writer who has ever been handed those words has quietly written whatever they were going to write anyway, and nobody could prove them wrong.

The version that works is unglamorous and much shorter than the one it replaces. It is mostly a set of before-and-after rewrites, a list of constructions that are banned outright, and a record of decisions the team has already argued about once and does not intend to argue about again. It reads like an editing guide rather than a philosophy, and that is the point.

Why the adjective list fails every time

Three adjectives fail for a mechanical reason: they cannot be violated. Give ten writers the word "bold" and you get ten defensible interpretations, one of which is an exclamation mark in a payment failure message. There is no version of that sentence a reviewer can reject on the evidence, so the review turns into taste, and taste loses to whoever is more senior or more tired.

The second failure is scope. Adjectives describe the voice of the marketing site and go silent everywhere the product actually speaks: field labels, validation errors, empty states, receipts, invoices, the sentence a subscription cancellation confirms. Those surfaces carry more words than the homepage and get read at higher emotional stakes. A voice document that never mentions them has abandoned most of the writing to nobody.

And a third failure is diagnostic. A team that cannot make its copy sound like itself often does not have a writing problem at all; it has an unresolved positioning problem wearing a writing costume. Before commissioning a voice, it is worth being honest about whether the problem is a brand problem or a marketing problem, because no amount of sentence-level craft rescues a company that has not decided what it is claiming.

What a tone of voice document must contain

Four parts, in this order of usefulness. If you only have time for the first, write the first and skip the rest; it will still change more copy than a forty-page brand bible.

The four parts

• Rewrites. Real sentences from your own surfaces, shown before and after, with the rule that produced the change written underneath. This is the part writers actually read.
• Bans. Constructions that do not get used here, stated flatly, with no explanation required at review time. A ban is enforceable. A preference is a discussion.
• Decisions. The recurring choices: we or the company name, second person or third, British or American spelling, sentence case or title case, how numerals are written, what the product is called in the second reference.
• Surfaces. How the voice bends per context. The same brand can be warm in an onboarding note and clipped in a validation error, and the document should say which is which rather than pretending one register covers everything.

Notice what is absent: mood boards, competitor voice audits, brand archetypes, and a paragraph explaining that the voice is like a knowledgeable friend. None of that survives contact with a Tuesday afternoon and a queue of microcopy. If your messaging and verbal identity work has produced a beautiful artefact nobody opens, the artefact is the problem, not the team.

Rewrites are the part people actually use

Take twenty sentences that are live on your own site and in your own product right now. Not invented examples, not competitor copy. Yours, with all their flaws. Rewrite ten of them side by side and write the rule underneath each pair. The rule is the deliverable; the pair is the proof that the rule does something.

Read the right-hand column on its own. That is the voice, and it took no adjectives to describe. A writer who has seen those six pairs can produce a seventh without asking anyone, which is the only test of a voice document that matters: does it let someone write a sentence you have not written yet.

A voice document is only real when a sentence dies because of it. Until then it is decoration with a table of contents.

Ban constructions, not words

Word bans are weak. Somebody always finds a use for the banned word that is genuinely correct, the ban gets an exception, then two, and after a quarter it is advisory. Ban the construction instead, because a construction is a pattern a reviewer can point at without arguing about meaning.

• The excitement opener. Any sentence that begins with how thrilled, delighted or excited we are. The reader did not open the email to hear about our mood.
• The category sentence. A description that names the software category and its beneficiaries without describing a single thing the product does. It is interchangeable with every competitor, which is a fact worth testing by swapping the logo.
• The double hedge. Two softeners in one clause: may sometimes, can potentially, might help to. One hedge is honesty. Two is fear.
• The passive apology. Inconvenience caused, issues experienced, delays encountered. Nobody caused anything in these sentences, which is exactly why they are written that way.
• The abstract noun stack. Three nouns doing the work of one verb. Solution delivery optimisation is a phrase with no subject and no action, and it can always be rewritten as somebody doing something.
• The punctuation shout. Exclamation marks anywhere near money, errors, deletion or cancellation. Enthusiasm in a failure state reads as contempt.

Keep the list short enough to remember. Six bans that everyone can recite beat thirty that live in a document nobody has opened since onboarding. Add a new ban only when you have caught the same construction twice in review, and retire one when it stops appearing.

Write down decisions, not preferences

Half the time a team spends on copy is spent re-deciding things it already decided. Is it sign in or log in. Do we capitalise the product name mid-sentence. Do we say customers, clients or users. Do we write per cent or use the symbol. None of these answers is interesting, and all of them cost an hour each time they come up unanswered.

So keep a decision log with a date against each entry and no rationale longer than a line. The rationale is what invites the argument back. A decision that is written down and dated is a decision; a decision that is remembered is a preference, and preferences drift with whoever is writing. This is the same instinct that makes a positioning statement worth freezing before the copy starts rather than after it.

The sequence we would run

1. Collect twenty live sentences. Pull them from the homepage, a pricing page, three product screens, two emails and one error message. Breadth matters more than volume here.
2. Rewrite ten of them in public. Do this in a room with the people who will have to live with the result. The arguments that surface are the document.
3. Extract the rules from the rewrites. Never the other way round. Rules invented in the abstract are the adjectives problem again, in longer form.
4. Turn repeated corrections into bans. If the same fix appeared three times across ten rewrites, it is not a fix. It is a pattern, and it belongs on the ban list.
5. Record the decisions with dates. Spelling, capitalisation, person, product naming, numerals, contractions. One line each, no essays.
6. Test it on someone who was not in the room. This is where most voice work is quietly proven useless, and it is the cheapest test in the whole process.

The stranger test

Hand the document to a competent writer who has never worked on your brand. Give them a brief for a page you have not written yet, an hour, and no other guidance. Then compare what comes back with what your own team would have produced.

If the stranger’s draft sounds like you, the document works. If it sounds like a competent draft of a generic brand, the document is describing a category rather than a company. If the stranger asks four clarifying questions before starting, the answers to those four questions are the sections you forgot to write. Every outcome is useful, which is rare for a test that costs an afternoon.

Both tests belong in the same conversation as a brand audit, because the gap between what a brand says about its voice and what its live surfaces sound like is usually the most concrete finding available. It is also the one people can act on the same week.

Where it lives, and who enforces it

A voice document stored where copy is not written is a document that does not exist. Put it beside the work: in the content workspace, in the design system next to the component that renders the label, in the pull request template if that is where microcopy lands. The distance between the rule and the writing predicts compliance more reliably than any amount of training.

Enforcement needs a name against it. One person who can send a sentence back, and whose decision does not get overturned by seniority. That role is closer to a sub-editor than to a brand guardian, and it works best when it is scheduled rather than heroic, in the same way a content calendar survives a busy quarter only when someone owns the slots.

Small surfaces, where voice does all the work

The smaller the surface, the more the words carry. On a profile site assembled from structured widgets, of the kind Nichevio produces, there is no layout system and no photography budget doing any lifting; the voice is nearly the whole differentiator. That is worth remembering when deciding whether a personal brand or a company brand should carry the business, because a person’s voice and an organisation’s voice are governed very differently.

The same question scales up. If you are running more than one name, each one needs its own bans and its own decisions, or they converge into a single house style and the second brand stops being a second brand. We have written about whether you can carry a second brand at all, and our parent company has covered the structural side in brand architecture for a small company.

Where we would start, and when to skip it

Start with the error messages. They are the sentences most likely to be read by someone in a bad mood, they are almost always written by whoever built the feature, and they are short enough that ten of them can be fixed in a morning. Fixing them proves the rules work on the hardest surface first. Marketing copy is the easy case, and starting there is how voice work becomes an exercise nobody outside the brand team notices. It is also where the accessibility guidance and the voice guidance agree: an error has to identify what went wrong in text, so vagueness is a defect rather than a style choice.

The concession, and it is a real one: none of this fixes a voice problem caused by a company that has not decided what it sells. If three people in the room would describe the offer three different ways, the rewrites will keep producing sentences that contradict each other, and the document will be blamed for a decision that was never made. Sort the claim first, then the sentences.

There is also a case for skipping the document entirely. A team of two who write everything themselves and agree instinctively do not need governance; they need to keep writing. The document becomes necessary at the moment a third person starts producing copy, or the moment the founder stops reading every sentence before it ships. That threshold arrives without announcing itself, and it usually shows up alongside the symptoms behind the things that get sold off when a brand stops working. If you are past it and the copy has started to wander, send us the sentences that annoy you most.