A skill that makes Claude build a brand voice guide the way a senior verbal identity writer does, from evidence instead of adjectives. Share five to ten pieces of your own writing, or answer a short interview if the brand is new. Claude sorts the samples, measures them with code (sentence length, contractions, who the text talks to, reading grade and what never appears) and writes three or four traits as "X, not Y", each with a do and a don't from your own copy. Then come the word lists, rhythm rules with numbers, a tone map for ten to fourteen situations and a library of before and after rewrites. After that, it rewrites any text in the voice.
Inside are five reference files. One covers sorting samples and the interview, one holds a tested measuring script with working ranges, and one sets the rules for traits, tone and rewrite pairs. A rules file covers US claim words such as non-toxic, Made in USA and kills germs, the FTC rule on fake reviews, WCAG 2.2 for error messages, and SMS length, each with its source and checked on 17 October 2026. The last is a complete worked example for a fictional cleaning brand. Claude works one stage at a time and ends each with one question and a sensible default. It never invents a fact, a review or a customer quote, and a rewrite keeps every price, claim and legal line exactly as written.
It is a writing skill, so it makes no images. Measuring needs code execution switched on; without it, Claude counts by hand on twenty sentences and labels every figure as an estimate. It reads your site only if web search is on, and the legal notes are US rules, not legal advice. A thin set of samples gives a provisional guide, so share as much real writing as you can, especially support replies and the pieces you wish you had never sent. When the guide is done, paste its prompt block into a Claude project's instructions, so every chat in that project writes in the voice.
Claude skill · version 1.0
Brand voice builder
Checking your access.
In the download · 10 files
- SKILL.mdThe method Claude follows: gather, measure, traits, words and rhythm, tone map, the guide, then rewrites, one stage at a time.
- README.mdHow to install it in claude.ai or Claude Code, how to start, and the licence.
- references/samples-and-interview.mdWhat counts as a sample and how much is enough, the interview for a brand with nothing written yet, and a check for an existing voice guide.
- references/measuring-voice.mdA tested Python script that measures sentence length, rhythm, contractions, pronouns, reading grade and absences, with working ranges and the checks a rewrite must pass.
- references/traits-and-tone.mdHow to write traits as X, not Y, the four tests, the four tone dials, the tone map with a stakes ladder, and the rules for rewrite pairs.
- references/rules-and-sources.mdUS claim words (FTC, EPA), the rules on fake reviews and endorsements, WCAG 2.2 rules for error messages and links, style books and SMS length, each sourced and checked on 17 October 2026.
Install it and begin
01Install it in Claude CodeTerminal mkdir -p ~/.claude/skillsunzip ~/Downloads/brand-voice-builder.zip -d ~/.claude/skills/# Start it with /brand-voice-builder. If this is your first skill, type /reload-skills once.On claude.ai there is nothing to unzip: upload the zip as it is. Every step is on Install and use a Claude skill.
02Your first message to ClaudeText Use the brand voice builder skill. The brand is [BRAND], selling [WHAT YOU SELL] to [WHO BUYS IT] in [COUNTRY]. I've attached [NUMBER] pieces of our writing: [WHICH PIECES, FOR EXAMPLE SITE, EMAILS, PACKAGING]. Sort and measure them first, then tell me what is missing.Attach five to ten real pieces from at least three places, including two or three support replies with customer details removed. No writing yet? Say so and it will interview you instead.
Worked example
Building a voice guide from your own writing
- 01Gather five to ten real pieces from at least three channels, including two or three support replies.Support replies show how a brand handles bad news, which is where most voices break.
- 02Measure them before describing them: sentence length, contractions, who the text talks to, and what never appears.Numbers turn "we're fairly casual" into a rule a freelancer can check, and an absence is as telling as a habit.
- 03Write three or four traits as "X, not Y", each with a do and a don't taken from your own copy.The "not" stops a trait being overdone; "direct" on its own drifts into curt.
- 04Fix the words: what you say, what you never say and what to say instead, plus the claim words that need proof.A banned word with no replacement gets ignored, and a word like "non-toxic" on a vocabulary list spreads a claim into every advert.
- 05Map the tone across ten to fourteen situations, with at least three where something has gone wrong.The voice stays put while the tone moves; a joke that suits a launch post is wrong in a refund refusal.
- 06Build twelve to twenty before and after pairs, then a one-page summary and a prompt block for Claude.Writers learn a voice from examples, and the single page is the part that actually gets opened.
- 07Test it on a real draft: list every fact, rewrite, then score before and after against each trait.If the same fault turns up in three rewrites, the guide needs a new rule rather than more fixes by hand.
What to do it with
- ClaudeFree tier, paid above it
A language model that writes, edits, reads long documents and runs tools, including agents that work inside a codebase.
Bad at It is agreeable by default, so it will improve a bad idea instead of telling you it is bad. Ask it to argue against you and the quality of the answer changes.
- Claude CodeSubscription
An agent that reads a whole repository, edits files, runs the tests and explains what it changed.
Bad at It is fast enough to make a large mistake quickly. Work on a branch, read the diff, and never let it push. Watch the starting mode: on Pro, Max and Team plans it opens in auto mode, which Anthropic's own list says may read a .env file and send the key in it to the API that key belongs to. Switch it off.
- Hemingway EditorFree tier, paid above it
Marks long sentences, passive constructions and words with simpler alternatives.
Bad at Follow it slavishly and every writer ends up sounding the same, which is the opposite of verbal identity.
Why it matters
Ask a chat for a brand voice and you get four adjectives any competitor could sign, with no rule a writer can follow on Monday. This skill makes Claude measure what your writing already does, tie each trait to evidence and an example, say how the voice changes when something goes wrong, and leave facts, claims and customer words untouched when it rewrites. A freelancer can then check a draft against the guide line by line.
- Type
- technique
- Period
- ongoing