How to keep one brand voice when several people are writing

Why brand guidelines do not survive contact with a writer

Open almost any brand guidelines document and you will find a page that says something like: confident but approachable, expert but never stuffy, human, never robotic.

Nobody has ever written a sentence differently because of that page. It is not a failure of effort — it is a category error. Those are descriptions of an effect the writing has on a reader. A writer needs instructions about choices: how long the sentences run, whether to use contractions, whether "we" appears at all, whether a paragraph may be one line.

The gap shows up immediately when you hire. A freelancer reads "confident but approachable", writes something perfectly competent, and it comes back wrong. Neither of you can say why. You edit it yourself, which takes longer than writing it, and you conclude the freelancer was not very good. Usually the freelancer was fine and the brief was empty.

The second failure is slower and worse. When there is no shared definition, the voice drifts towards whatever the most prolific writer does naturally — or, increasingly, towards the default register of whichever AI model people are drafting with. Both drifts are invisible month to month and obvious across a year.

What a voice actually consists of

Voice feels ineffable, but most of what makes two writers sound different is countable. You do not need software to do this — you can do it by hand on ten pieces with a highlighter — but you do need to look at the right things.

These are the dimensions that carry most of the signal:

• Sentence length variation — more revealing than average length. A voice that alternates a twenty-eight word sentence with a four word one reads nothing like a voice that runs every sentence to seventeen words. Flat variance is the single most common tell in machine-drafted copy.
• Contraction rate — "don't" against "do not". This moves register faster than any other single choice, and it is the easiest instruction to follow.
• Person — does the brand say "we"? Does it address the reader as "you"? Is it written from nowhere, in the passive? Many guidelines never answer this, and it changes everything.
• Punctuation habits — em dashes, semicolons, colons, parentheses, question marks used as section openers. Writers have strong unconscious preferences here and they fingerprint a voice quickly.
• Paragraph length — whether one-line paragraphs are allowed is a real editorial decision, not a formatting detail.
• Concreteness — how many named things, numbers, places and specifics appear per hundred words. Vague writing and specific writing are different voices even with identical vocabulary.
• Hedging — the rate of "may", "often", "generally", "can help to". Some brands hedge everything for legal reasons. Some never hedge. Mixing the two in one publication reads as inconsistency.
• Sentence openings — does the voice start on the subject, or lead with a subordinate clause? Repetitive openings are the fastest way to sound mechanical.

Notice that vocabulary is not on that list. Word choice is what people think voice is, and it matters least — a banned-words list will not save you if the rhythm is wrong.

Building a voice brief someone can actually hit

The method is unglamorous and takes about two hours per brand.

1. Assemble the canon. Pick eight to ten published pieces that everyone senior agrees sound right. Not the best-performing pieces — the most characteristic ones. If you cannot find eight, that is the finding: the voice does not exist yet, and you are about to invent it rather than document it.

2. Assemble the anti-canon. Pick three pieces that are wrong, and write one sentence each on why. This is the part people skip and it is worth more than the canon. "Wrong" is far easier to articulate than "right".

3. Measure the canon against the dimensions above. Ranges, not single numbers — a voice is a band, not a point.

4. Write the brief as rules with examples. Every rule gets a before and after drawn from real work. A rule without an example will be interpreted, and interpretation is the thing you are trying to eliminate.

5. Put the "never" list first. Writers read the top of a brief properly and skim the rest. Prohibitions are also easier to comply with than aspirations.

VOICE BRIEF — [Brand]
Last reviewed: [date] · Owner: [name]

NEVER
- [e.g. Never open a piece with a rhetorical question.]
- [e.g. Never use "solutions", "leverage", "seamless", "in today's
  fast-paced world".]
- [e.g. Never write a paragraph longer than five lines on screen.]
- [e.g. Never hedge a factual claim we can actually stand behind.]

ALWAYS
- [e.g. Always use contractions. "We're", not "we are".]
- [e.g. Always address the reader as "you". The brand is "we".]
- [e.g. Always name a specific thing in the first two sentences.]

RHYTHM
- Sentence length: mostly [12-20] words, with deliberate short ones for
  emphasis.
- Vary it. Two sentences of similar length in a row is fine; five is
  not.
- Paragraphs: [1-4] lines. One-line paragraphs are [allowed / not
  allowed].

REGISTER
- Contractions: [always / sometimes / never]
- Person: [we / you / impersonal]
- Humour: [none / dry, sparing / warm]
- Jargon: [assume the reader knows X but not Y]

PUNCTUATION
- Em dashes: [freely / sparingly / never]
- Semicolons: [yes / no]
- Exclamation marks: [never]

SOUNDS RIGHT
[Paste 3 short passages from the canon. No commentary — let them do the
work.]

SOUNDS WRONG
[Paste 3 passages, each with one line on what is off about it.]

WHO DECIDES
Ambiguities go to [name]. Anything unresolved after one round ships as
drafted and gets logged, rather than waiting.

That last section matters more than it looks. Most voice problems are actually decision-rights problems wearing a disguise.

Reviewing without becoming the bottleneck

If you are the only person who can tell whether a draft is right, you have not built a voice — you have built a dependency on yourself. The brief above is the first half of the fix. The second half is changing what you review.

Split the review in two. Conformance — does it follow the brief? — is checkable by anyone, including the writer, before it reaches you. Judgement — is this the right argument, is this claim true, does this actually help the reader? — is the part that genuinely needs you.

Most senior editors spend the majority of their review time on conformance, which is the half that could have been handled upstream. Pushing conformance to the writer usually cuts review load substantially, and it has a second effect: writers improve when the feedback is specific and immediate rather than arriving as a rewritten document three days later.

Catching drift before the client does

Voice drift is not a single event. It is a slow averaging, and it has three common engines:

• A dominant writer. Whoever produces most of the volume gradually becomes the house style, regardless of what the brief says.
• Model default. When drafting starts in an AI tool, the output arrives in a recognisable register — even, hedged, tidy, structurally repetitive. Editing removes the obvious words and leaves the rhythm, which is the part readers actually notice.
• Cross-contamination. An agency running ten clients will find that all ten start sounding like the agency, because the same three people are writing them.

The defence is a periodic check rather than a permanent vigilance. Once a quarter, take five recent pieces and re-measure them against the canon. If contraction rate has halved and sentence variance has flattened, you have your answer, and it took twenty minutes to find.

Do this per client, not per agency. A single average across a roster hides exactly the problem you are looking for.

Where a tool fits, and where it does not

Everything above works with a spreadsheet and some discipline. What a tool changes is the cost of doing it repeatedly — measuring ten pieces by hand is a morning; measuring every draft is not something anyone will keep up.

Be clear about what a tool cannot do, though. It cannot tell you whether an argument is any good, whether a claim is true, or whether the piece was worth writing. It measures the surface. The surface happens to be where voice lives, which is why it works for this particular problem and not for most others.

This is what Wordcheck was built for: it holds a measured profile per client, built from the canon you paste in, and scores drafts against that profile rather than against a general idea of "good writing". Agencies get a distinct voice per client and a report they can hand over. There is more on that on the agencies page.

Common questions