Brand Voice Definition

Brand Voice Definition

Most voice guides are a list of adjectives - "confident, warm, clear" - that describe every brand and constrain none. A writer cannot check a draft against "warm", and neither can an AI. What they can check is a reading level, a sentence length, whether the reader is "you", whether contractions are used, and which words never appear.

This skill gets those rules from the writing you already agree sounds right, rather than from a workshop. Give it three or more on-brand pieces and it measures them, checks that they agree with each other (if they do not, you do not have one voice yet), contrasts them with anything you mark off-brand, and writes the measurable half of the guide. The output includes a profile the brand-voice-governance engine checks new content against, so the guide is enforced from the day it is written. The adjectives, the do/don't list and the examples are still yours to write; this gives them a spine.

When to use

The user is writing a voice guide for the first time, rewriting one nobody uses, briefing an AI tool or an agency on how the brand sounds, or trying to explain why one piece of copy is "off".

Do not use it to check content; brand-voice-governance does that with the profile this produces. Do not use it to decide what the brand stands for; positioning-statement and brand-product-context come first.

Before you start

1. Pick three to six pieces the team agrees sound like you, each 100 words or more: a homepage section, an email a customer replied to, an FAQ, a post the founder is proud of. Agreement matters more than volume.
2. Pick one or two pieces you do not want to sound like, if you have them: the old press release, the agency draft that got rejected.
3. List any words already off-limits. The engine adds AAJ's house list of buzzwords unless your on-brand writing uses them.

Method

Measure, do not describe. Per sample: Flesch-Kincaid grade, average and longest sentence, "you" / "we" / "I" per 100 words, contractions per 100 words, exclamation and question marks as a share of sentences, hedges per 100 words (very, really, just, perhaps...), buzzwords per 100 words, adverbs, and a passive-voice estimate. Samples under 100 words are shown but not counted.

Check the samples agree. On-brand samples more than three grades apart, or more than eight words apart in average sentence length, are not one voice (engine rules). The verdict is "not one voice" until you decide which end is the brand and drop the rest.

Write the rules from the medians. Reading level: the on-brand median grade plus one of headroom (an engine rule). Sentences: the median average, with the longest kept under twice it. Person: whichever of you / we / I the samples use most, if they use one at all. Contractions: used or not, from the rate. Exclamation marks: none if fewer than 2% of sentences. Hedges: at or below the on-brand rate. Passive voice is called out if it passes 15% of sentences.

Build the avoid list honestly. The house buzzwords and your candidates go on the list unless the on-brand samples use them; a word the brand uses is flagged for a decision, not banned behind your back. Buzzwords that appear only in off-brand samples are added.

Contrast with off-brand. The metrics where off-brand writing differs most are the rules worth writing down first. If it differs in none of them, what makes it off-brand is not measurable here, and belongs in the do/don't list in words.

Workflow

Step 1 - Collect. Three to six on-brand pieces, one or two off-brand, the off-limits words.

Step 2 - Run. If the verdict is "not one voice", argue about which samples stay, then re-run.

Step 3 - Write the guide. The rules go in as they come out; add the do/don't list, the adjectives and two example rewrites beside them. The Brand Voice Guide template has the structure.

Step 4 - Wire it in. Pass the profile to brand-voice-governance and check the next ten pieces the team or an AI produces.

Step 5 - Re-derive yearly, or when the on-brand pile changes.

Run the tool

node resources/voice-derive.js --demo

run({ brand, samples, avoidCandidates }) returns each sample's measurements, the on-brand and off-brand medians, the spread and whether it is consistent, the person, the written rules, the profile object for governance, findings and a verdict.

Present the result

1. The verdict: profile ready, not one voice, or too thin.
2. The rules, as written, ready to paste into the guide.
3. The contrast with off-brand writing in one line: where it differs most.
4. The profile object, and the instruction to run the next drafts through governance.

Guardrails & common mistakes

• Never derive a voice from one writer's favourite piece. Three samples the team agrees on, minimum.
• Never ban a word the brand actually uses without deciding to. The engine flags it; you decide.
• Do not mistake the measurable rules for the whole guide. They are the spine; the do/don't list and the examples carry the character.
• Do not include off-brand samples as on-brand to "balance" the voice. That is how you get a grade-9 rule nobody follows.
• Reading grade is a ceiling, not a target. Simpler than the ceiling is always allowed.

Related AAJ resources

• Why the guide needs a spine, and how to write the rest: https://aajconsult.com/playbooks/brand-voice-playbook
• The guide template these rules paste into: https://aajconsult.com/resources/brand-voice-guide
• Check copy against the rules in a browser: https://aajconsult.com/tools/brand-voice-checker
• The problem this solves: https://aajconsult.com/blog/why-ai-content-doesnt-sound-like-you

Related skills

• brand-voice-governance - checks content against the profile this produces
• brand-product-context - where the voice section of the brief lives
• aaj-copywriting - writing inside the rules
• positioning-statement - what the voice is in service of