01 Claude skill

AI brand voice generator for Claude

Extracts a brand's real voice from human-written samples and bottles it as a brandVoiceProfile YAML — evidence-mapped traits, PII redaction, and naturalness rules that keep AI output sounding human, not generated.

Claude skill bundle · SKILL.md + schema reference intermediate Free · No upsell
10 traits, 1 YAML
Every voice trait scored 1–100 and mapped to a real, redacted snippet — portable to any LLM.
Free download · Claude skill bundle

Get the bundle

Two fields. We send the files and a setup note — that's it.

No spamUnsubscribe anytimeGDPR-friendly
02 The leak

If this sounds familiar, the claude skill is for you.

01

Every prompt drifts back to the same generic, AI-sounding voice because nothing tells the model how the brand actually writes.

02

Your tone-of-voice doc is a slide of adjectives no model can actually apply.

03

Voice falls apart across tools — the email bot, the social writer and the support replies all sound like different companies.

03 What's inside

A working system, not a prompt.

Each piece extracted from a live engine — same prompts, same scaffolding, same rules.

  • 01
    SKILL.md — the Brand Voice Linguist (v3.3): a step-by-step protocol that reads your communication samples and outputs one brandVoiceProfile YAML, nothing else
  • 02
    references/schema.md — the full YAML contract: voiceSummary, 10 scored coreTraits, writingStyle, lexicon, interactionPatterns, naturalnessRules, aiTellAvoidance, doDont, evidence, confidence
  • 03
    10 core voice traits scored 1–100, each anchored to a redacted snippet pulled from your own writing — falsifiable, not adjective soup
  • 04
    Adaptive AI-tell avoidance: it detects which LLM 'tells' actually show up in your samples and only bans the ones you don't use
  • 05
    Built-in PII redaction: names, emails, phone numbers and IDs become placeholders before anything reaches the profile or its evidence snippets
  • 06
    A setup interview that pins register, emoji policy, persona and brand-speaker first, plus minimum-data thresholds so it refuses to guess on thin input
Method · Brand Voice Linguist v3.3

Most tone-of-voice docs are adjectives. This extracts the patterns a model can actually obey — sentence rhythm, diction, the AI tells your brand does and doesn't use — and writes them down as structured YAML.

— Marc-Aurele Legoux · Marcus-Aurelius Digital

10 traits
Core voice dimensions, each scored 1–100 and cited to a real evidence snippet
Level A
PII redaction applied before any sample enters the profile
Adaptive
AI-tell avoidance keyed to your source — it bans only the tells your brand never uses
04 Audience

Built for a specific operator. Not for everyone.

Who it's for
  • Operators who want one Claude session to capture a client's voice and hand a reusable YAML profile to every downstream automation
  • Founders who want Claude — or any LLM — to write in their own voice without re-explaining tone in every prompt
  • Agencies onboarding a brand whose emails, posts and replies all need to sound like the same person across surfaces
Probably not for you if
  • Anyone without real writing samples — the skill won't invent a voice, and asks for at least ~300 words and ~20 messages first
  • Anyone who just needs one on-brand email — write it directly instead of extracting a full profile
  • Teams expecting finished marketing copy — this produces the voice profile, not the campaign
05 · Get it

The same claude skill I run for paying clients. Yours, free.

Two fields. The bundle and a short setup note land in your inbox.

No spamUnsubscribe anytimeGDPR-friendly

06 · FAQ

Common questions

Do I need to feed it my own writing?

Yes. It extracts voice from real human-written samples — chats, emails, DMs, call transcripts. It won't invent a voice, and it asks for at least ~300 words and ~20 messages before it'll produce a confident profile.

What do I actually get out of it?

One brandVoiceProfile YAML file. Drop it into any LLM's context — a Claude Project, a system prompt, an n8n node — and the model writes in the captured voice.

Is my data safe to paste in?

The skill redacts PII — names, emails, phone numbers, IDs — into placeholders before anything is written into the profile or its snippets, and it never echoes back full message histories.

How is this different from telling Claude to 'write in a friendly tone'?

Adjectives like 'friendly' don't transfer. This captures the mechanics — sentence-length variance, signature phrases, punctuation habits, which AI tells the brand does and doesn't use — so output sounds like a specific person, not a vibe.

Marc-Aurele Legoux, founder of Marcus-Aurelius Digital
Marc-Aurele Legoux · Founder

“I build AI growth engines for luxury villas, boutique hotels, and high-ticket service brands. The skills I give away are the ones I run for paying clients. If you build it yourself, great — if you’d rather have it built and run, talk to me.”

More claude skills