Make Claude write in your brand voice

Why a skill instead of a prompt

Pasting “write like us” instructions into every chat works until you forget, paste an old version, or a colleague uses a different one. A skill fixes that. It lives in Claude, it has a short description that tells Claude when to use it, and Claude reads the full instructions only when a task matches. You set it up once, and every request for copy gets the same voice.

Skills use the open Agent Skills format, which Anthropic released as a standard. The same folder also works in Claude Code and in other tools that read skills, such as OpenAI Codex, Cursor and GitHub Copilot.

What a brand voice skill contains

A skill is a folder with a file called SKILL.md. The file starts with two required fields between --- lines, then the instructions in Markdown:

---
name: acme-brand-voice
description: "Writes and edits copy in Acme's brand voice. Use for any Acme content: web pages, blog posts, emails, social posts, product copy."
---

# Acme brand voice

One paragraph on what makes this voice different.

## Voice traits
### Confident, never loud
- Do: State the outcome first, then how.
- Don't: Stack adjectives.
- Sounds like: "a real line from your site"

## Writing mechanics
- Median sentence: 11 words. No exclamation marks.

## Vocabulary
## Brand facts
## Before you deliver

• name: lowercase letters, numbers and single hyphens, up to 64 characters, and the same as the folder name. On Claude it can't contain the words “claude” or “anthropic”.
• description: what the skill does and when to use it. The spec allows 1,024 characters; Claude's help center asks for 200 or fewer, so stay under that.
• The body: the instructions. The spec recommends keeping SKILL.md under 500 lines and moving long material, like annotated examples, into files such as references/examples.md that Claude opens when it needs them.

What makes the instructions work

The difference between a voice skill that changes Claude's writing and one that doesn't is specificity.

• Behaviour, not adjectives. “Be confident” does nothing. “State the outcome first, then the mechanism; no hedging words like might or could” does.
• Numbers where you have them. Median sentence length, whether you use exclamation marks at all, sentence case or Title Case headings. Claude can hold to a number.
• Your real words. A short list of words you actually use, product names spelled exactly, and a few words that would sound wrong coming from you.
• Quotes as calibration. Three to six real passages, labelled as examples to learn from rather than text to reuse.
• Facts kept apart from style. A short list of brand facts, and an instruction to ask rather than invent when a piece needs a number or a claim that isn't there.

Saylike builds exactly this from your website and writing samples, and checks every quote against the source before it goes into the file.

Install it in Claude (web and desktop)

1. Skills need code execution. In Claude, open Settings, then Capabilities, and turn on code execution and file creation. On Team and Enterprise plans an owner enables skills for the organisation first.
2. Open Customize, then Skills, select +, then Create skill, then Upload a skill.
3. Upload the ZIP. The skill folder must be at the root of the ZIP, so the archive contains acme-brand-voice/SKILL.md, not a loose SKILL.md. Saylike's download is already packed this way.
4. Ask for something: “Write a launch email for our new feature.” Claude picks the skill up from its description.

Skills you upload in the Claude app are personal. They don't sync to the Claude API or to colleagues automatically.

Install it in Claude Code

Unzip the folder into ~/.claude/skills/ to have it in every project, or into .claude/skills/ inside a repository to share it with everyone who works there. The folder name must match the name field.

Other tools that read skills

The Agent Skills format is supported well beyond Claude. OpenAI Codex reads skills from ~/.agents/skills/ and a repository's .agents/skills/. Cursor, GitHub Copilot, VS Code and Gemini CLI also load skill folders; check each tool's documentation for its folder. Assistants without skills support can use the same instructions as plain text: see brand voice in ChatGPT.

When the upload fails

Keep it current

Treat the skill like a style guide: edit it when you notice Claude getting something wrong (“we never say ‘solution’”), and regenerate it when your site's voice changes. Because it is a plain file, you can keep it in version control and review changes like any other document.