# Brand voice and tone: how to write rules a stranger can follow

> A brand voice only holds if a new writer can apply it without asking you. Write a few rules that can be checked (sentence length, who the sentence is about, words you never use), keep adjectives for judgement calls, and move tone by the reader's state. The research on tone is thinner than blogs suggest.

By LAIN, 25 Mar 2026. 10 min read. Category: Branding. Canonical: https://lainagency.com/resources/brand-voice-and-tone-guide

A brand voice is only real if someone who has never met you can write in it. That is the test most guides skip. They list three adjectives, "friendly, professional, innovative", and two writers read them two ways. This article is about the other kind of guide: rules short enough to remember and exact enough to check.

The short version: **voice is a set of repeatable choices**, tone is how you set those choices for the moment, and the choices worth writing down are the ones a stranger could get wrong.

## Voice stays, tone moves
Mailchimp's public content style guide draws the line most teams use. The voice is constant. The tone changes with the reader's emotional state: you would not talk to someone who is scared or upset the way you talk to someone who is laughing. Apple's WWDC24 session on UX writing makes the same split, with voice as the consistent expression of a brand's values and tone as what changes by situation. A write-up of that session notes that Apple's Activity app leans on exclamation marks while its Health app stays sober, and nobody would say those are two companies.

So the practical question is never "which tone is ours". It is "which choices stay fixed, and which move".

## What the research does and does not show
Almost every article on this topic cites the same source, so it is worth reading it closely. In 2016 Nielsen Norman Group (NN/g) published a study of tone built on four dimensions: funny to serious, formal to casual, respectful to irreverent, enthusiastic to matter-of-fact. They wrote pairs of near identical pages for four kinds of organisation (auto insurance, checking accounts, home security and a hospital), changed only the tone, and asked 50 US adults to rate each organisation from 1 to 5.

Three details matter more than the headline.

- **Casual, conversational and enthusiastic versions scored best**, even for the hospital. The formal, serious version scored 3.6 for friendliness, 4.0 for trustworthiness and 3.5 for desirability. The casual, enthusiastic one scored 4.1, 4.1 and 3.7. Formality did not buy trust, but the trust gap was 0.1 of a point.
- **Trustworthiness predicted willingness to recommend**, which is why the trust column is the one to watch.
- **NN/g itself warns that two samples per industry cannot support reliable conclusions** about any one industry. It is a small, US only study of short samples, not a law.

The honest reading: tone changes how people rate an organisation, a stiff default does not protect you, and the best tone for your audience is something to test, not assume. The four dimensions are also best used on a piece of writing ("this reminder is casual and matter-of-fact"), not on a company ("we are casual"). A voice is the range you allow on each dimension. Tone is where a specific piece lands inside it.

A second, older source is Jennifer Aaker's 1997 study of brand personality in the Journal of Marketing Research, which grouped 114 traits into five dimensions (sincerity, excitement, competence, sophistication, ruggedness) from ratings of 37 US brands. It is a useful word bank for choosing traits. It is not universal: later cross-cultural work found only three of the five dimensions in Japan, Spain and the US alike, and ruggedness was specific to the US. If your audience is not American, treat the list as a prompt.

## Write rules a stranger can check
Sort every rule into one of two kinds.

| Kind | Example | How it is checked |
|---|---|---|
| Mechanical | Average sentence under 14 words. Never open with "We". No "please be advised". | A script or a second pair of eyes in ten seconds |
| Judgement | Be warm without being cute. Use humour only when the reader is relaxed. | A person, against examples |

Mechanical rules do most of the work, because they are the ones that survive a new freelancer, a new agency or a tired Friday. The guides people trust are full of them. Google's developer style guide asks for a conversational, friendly tone, then makes it checkable: prefer "you", avoid slang and jargon. Microsoft's style guide reduces its voice to three traits: warm, crisp and helpful.

If you want the rules enforced rather than remembered, a prose linter such as Vale reads rules from small YAML files. A rule file is a few lines: `extends: substitution`, a message, a level, and a `swap` map such as `"we regret to inform you": "sorry"` and `"customer service department": "support team"`. Vale also takes a vocabulary folder with an accept list and a reject list.

You do not need a linter to start. The point is that a rule you could paste into one is a rule a stranger can follow.

## A worked example you can copy
Take one ordinary message, a late delivery, and write it three ways. Then measure what you can measure.

![Three cards showing the same late delivery message as direct, warm and formal, with reading ease scores of 94, 92 and about 47 and the word where the news arrives.](https://lainagency.com/images/blog-art/brand-voice-and-tone-guide-fig2.webp)

The numbers come from the Flesch Reading Ease formula (Rudolf Flesch, 1948): 206.835, minus 1.015 times the average words per sentence, minus 84.6 times the average syllables per word. Higher is easier, and 60 to 70 is the usual band for general adult readers.

| Version | Words | Average sentence | Reading ease | The news ("late") arrives on word |
|---|---|---|---|---|
| Direct | 20 | 6.7 words | about 94 | 4 |
| Warm | 28 | 9.3 words | about 92 | 6 |
| Formal | 34 | 11.3 words | about 47 | 14 |

Two things stand out. The warm version is not slower to read than the direct one, so warmth did not cost clarity. The formal version buries the news under fourteen words of ceremony and drops far below the 60 to 70 band. That is the difference you can write a rule against: "state the news in the first six words" and "reading ease above 70 for anything a customer must act on". Reading ease measures surface features only, never meaning, and its constants come from English, so use it as a smoke alarm, not a judge.

Whichever version your team signs becomes an example in the guide. Choose by which one the team would be happy to see sent in their name.

## Set tone by the reader's state
Keep the voice rules fixed and move the dials by what the reader is going through. Use the table below as a starting point and edit it from your own support inbox.

| Moment | The reader is | Move the dials |
|---|---|---|
| Good news, a launch | Curious | More enthusiastic, humour allowed |
| Receipt or confirmation | Wanting certainty | Matter-of-fact, shortest sentence you have |
| Something went wrong | Worried, annoyed | Serious, own it once, give the next step |
| Complaint | Upset | Slow down, name the problem, no humour |
| Money, legal, health | Careful | Precise first, warmth second |

Two sourced rules belong in the error row. NN/g's error message guidelines ask for a positive, nonjudgemental tone and say to avoid words that imply fault, such as "invalid", "illegal" or "incorrect". They also say to avoid humour in errors, because a joke wears thin when people hit the message often. Mailchimp's guide ends up in the same place from a different direction: it values clarity over entertainment, and says to keep a straight face when unsure. You have to pick a tone for the state, but you do not have to be solemn.

## Voice in other languages
Voice does not translate word for word. In German and French one of the first decisions is not vocabulary but address: "Sie" or "du", "vous" or "tu". Decide it per locale and write it into the guide, or each translator will decide it for you. Localisation vendors report that translators given no direction drift to stiff, formal copy. That is a vendor view, not measured evidence, but it matches the problem you can see in any multilingual site. For a line meant to persuade, localise it. For a line meant to define the brand, such as a tagline, ask a writer to transcreate it, which means rewriting for the effect rather than the words.

## Where it goes wrong
- **The guide describes, it does not decide.** Cause: adjectives feel safe because nobody can disagree with them. Detect: ask two people to apply "friendly" to the same sentence. Fix: every trait gets a do and a do not, with a real sentence as each example.
- **A committee averages the voice into nothing.** Cause: each reviewer removes the sentence that offended them, and the survivors are the ones no one minds. Detect: remove the logo from five pieces and ask three people which company wrote each. If they cannot group them, you have no voice yet. Fix: one owner who can say no, and a short list of decisions that are not up for review.
- **The playful voice meets a refund request.** Cause: the voice was written for marketing and applied to everything. Fix: the tone table above, with "humour allowed" as an explicit column.
- **Nobody owns the small copy.** Cause: error messages, invoices, auto-replies and the 404 are written by whoever built the screen. Fix: list every template that sends words, give each one an owner and one example.
- **The rules live in a PDF.** Cause: a guide nobody opens cannot guide. Fix: put the mechanical rules where the writing happens, in a linter, a snippet library or the checklist of the template.

## Words are part of the system
Vocabulary is the most checkable part of a voice. The [Syntra](https://lainagency.com/works/syntra) case page shows it neatly: its design lists nine words for what goes wrong in a system (Error, Failure Point, Breakdown, Collapse and others), and the "A clear job" chapter turns each into the word it should have been, such as Reliability or Performance. Whatever you think of the metaphor, it is a voice rule you can apply with a search: name the state you want, not the one you fear. [Morgana](https://lainagency.com/works/morgana), a crypto platform, took a different line: its headlines are short and declarative, such as "Turning noise into clarity" and "Access is easier. Understanding is not."

## The one page
1. **Three traits**, each with one do and one do not, and one real sentence.
2. **Five mechanical rules**, written so a script could check them.
3. **A word list**: ten words you use and ten you never do.
4. **The tone table** for your six most common moments.
5. **Locale notes**: address form and any words that must stay in English.
6. **Three before and after rewrites** of your own copy, and one named person to ask.

This page becomes the voice section of your [brand guidelines](https://lainagency.com/resources/brand-guidelines-what-to-include) and follows from the personality words in your [brand strategy](https://lainagency.com/resources/brand-strategy-for-small-business). If your words have outgrown one writer and no longer sound like one company, [talk to us](https://lainagency.com/contact) about a voice and brand system as part of a [branding project](https://lainagency.com/about#logo-brand-identity).

## Frequently asked questions

### What is the difference between brand voice and tone?

Voice is the set of choices that stay the same: word choice, sentence length, how you address the reader. Tone is how you set those choices for a moment and for the reader's state. Mailchimp's style guide and Apple's WWDC24 UX writing session both describe voice as constant and tone as adaptive.

### Does a formal tone make a brand look more trustworthy?

Not reliably. In Nielsen Norman Group's 2016 test with 50 US adults, casual and conversational versions scored as well or better on trust, even for a hospital: 4.1 against 4.0 on a 5 point scale. NN/g warns that two samples per industry cannot support firm conclusions, so test with your own audience before choosing.

### How do you make a brand voice measurable?

Write mechanical rules that can be checked, such as average sentence length, the word where the news appears and a list of banned phrases. Reading ease (Flesch, 1948) is a quick smoke test: the same late delivery message scored about 94 in a direct version and about 47 in a formal one. Keep judgement rules, such as humour, for a short list with examples.

### Should error messages use the brand's humour?

Usually not. NN/g's error message guidelines say to avoid humour because it goes stale when users see the message often, and to avoid blaming words such as "invalid" or "incorrect". Mailchimp's guide advises a straight face when in doubt. Keep the voice, drop the jokes.

### How do you keep brand voice consistent across languages?

Decide the form of address per locale, for example "Sie" or "du" in German and "vous" or "tu" in French, and write it into the guide. Give translators examples and a word list, and have a writer rewrite taglines for effect rather than translate them. Without direction, translators tend to default to formal copy.

### How many traits should a brand voice have?

Three, each with a do, a do not and one real example sentence. More traits make the guide vaguer, not richer. Brand personality research such as Aaker's 1997 study can supply words to choose from, but its five dimensions came from US brands and did not all appear in Japan or Spain.

## Sources and fact check

Checked on 28 Oct 2026.

- [Mailchimp Content Style Guide: Voice and Tone](https://styleguide.mailchimp.com/voice-and-tone/)
- [Nielsen Norman Group: The Four Dimensions of Tone of Voice](https://www.nngroup.com/articles/tone-of-voice-dimensions/)
- [Nielsen Norman Group: The Impact of Tone of Voice on Users' Brand Perception](https://www.nngroup.com/articles/tone-voice-users/)
- [Nielsen Norman Group: Tone-of-Voice Samples](https://www.nngroup.com/articles/tone-voice-samples/)
- [Nielsen Norman Group: Error Message Guidelines](https://www.nngroup.com/articles/error-message-guidelines/)
- [Apple Developer: Add personality to your app through UX writing (WWDC24)](https://developer.apple.com/videos/play/wwdc2024/10140/)
- [Create with Swift: Writing better copies for your apps](https://www.createwithswift.com/writing-better-copies-for-your-apps/)
- [Google for Developers: Voice and tone](https://developers.google.com/style/tone)
- [Microsoft Learn: Microsoft Writing Style Guide](https://learn.microsoft.com/en-us/style-guide/welcome/)
- [Vale documentation: substitution check](https://docs.vale.sh/checks/substitution)
- [Aaker (1997): Dimensions of Brand Personality, Journal of Marketing Research](https://gwern.net/doc/economics/1997-aaker.pdf)
- [University of Reading: Brand personality across cultures (JPBM)](https://centaur.reading.ac.uk/88534/1/Brand%20personality_JPBM_prefinal_.pdf)