Brand Voice Codex: Make Every AI Tool Write Exactly Like You
Every creator who uses AI to write captions, emails, or scripts eventually hits the same wall: the output sounds like AI, not like them. The Brand Voice Codex for Claude and ChatGPT fixes this at the root. It analyzes your real content across five dimensions (tone, vocabulary, sentence architecture, humor, and platform adaptation) and produces a portable voice profile with a copy-paste AI prompt snippet that any tool can use to write in your voice. This guide covers how the analysis works, what the output looks like, and why 'be casual and friendly' is not a voice description.
Every creator who's used AI to write a caption or email eventually says the same thing: "It doesn't sound like me."
That's not a complaint about AI quality. It's a description of the problem. You haven't told the AI who you are. You've given it a topic and a format, and it's written the average version of that topic in the average version of that format. The average is always going to sound like AI.
The Brand Voice Codex changes that. It analyzes the writing you've already published (scripts, captions, newsletters, tweets, community posts) and produces a voice profile so detailed that any AI tool reading it will write output you look at and think: "Yeah, that sounds like me." Including a portable prompt snippet you can paste into any AI tool you use, once, to make every output in that tool match your voice by default.
Why "Be Casual and Friendly" Doesn't Work
Most creators who try to customize AI output start with instructions like "write in a casual, friendly tone" or "sound like a real person." These don't work, not because they're wrong, but because they're not specific. "Casual and friendly" describes half the internet. It doesn't describe you.
What actually distinguishes your voice is the specific words you gravitate toward, the sentence structures you default to, the way your energy shifts when you're explaining something technical versus reacting to something that surprised you, the humor pattern you use (dry asides? hyperbolic comparisons? none?), the words you never use even when everyone else does.
A generic voice instruction gives AI your latitude and longitude. A Voice Codex gives it your address.
How the Analysis Works
The Brand Voice Codex analyzes your content across five dimensions. You provide 3-5 real content samples: ideally from different formats (a video transcript and a set of tweets reveals more than five video scripts, because you can see how your voice shifts across contexts).
Tone Spectrum: Six spectrums, each scored 1-10: formality, gravity, authority, energy, warmth, optimism. The skill doesn't average these: someone can be simultaneously high-energy and skeptical, or calm and warm. More importantly, it notes the range, not just the position. A creator who swings between 3 and 9 on energy (measured, until they're recommending something they love) has a completely different voice than someone who sits at a consistent 6.
Vocabulary DNA: The specific words and phrases that make your writing sound like you. Focus on the distinctive ones. Your signature phrases, your power words, your avoidance patterns (the words you never use, even when everyone else would). This includes AI-isms you'd never say (no "delving," no "tapestry," no "it's worth noting") because the codex has to work against default AI behavior.
Sentence Architecture: Average sentence length, sentence fragment usage, how often you ask rhetorical questions, how you move between ideas, how you start sentences. The skill tracks 15-20 sentence openers to find your patterns. A creator who starts sentences with "Look," and "Here's the thing" has a completely different feel than one who starts with actions and evidence.
Humor and Personality: Self-deprecating? Dry? Observational? Absurdist? None? More importantly: when does humor appear: woven throughout, or saved for the end? Do you tell stories? How do you enter them? The skill identifies the specific type of personality that shows through your writing and documents it precisely enough to replicate.
Platform Adaptation: Your voice changes with the format: how you talk when you're writing a YouTube script vs. an Instagram caption vs. a community post. The codex captures each register separately (length preference, structural habits, formality shift) so the AI prompt snippet works correctly whether you're writing for your newsletter or your TikTok.
What the Output Looks Like
The Voice Codex produces a structured document with seven sections:
Voice Identity Summary: Two or three sentences that capture the essence of how you communicate. Not "casual and friendly." Something like: "Direct and opinionated with a dry sense of humor that shows up mostly in asides and parentheticals. Talks to the audience like they're already competent: explains the 'why' without over-explaining the 'what.' Gets genuinely fired up about tools and workflows, which breaks through the otherwise measured delivery."
Tone Settings: The six spectrum scores presented as a reference card, with ranges and trigger notes. If you're normally a 5 on energy but spike to 9 when recommending something you love, that's documented.
Vocabulary Guide: Two columns: 15-25 words and phrases to use (with context), and 10-15 words to avoid (including AI-isms that don't fit your style). This column is often the most useful single reference.
Sentence Patterns: 5-8 representative sentence structures pulled directly from your writing, with annotations explaining what makes them characteristic. The pattern isn't abstract: it comes from actual sentences in your content.
Platform Notes: Voice adaptation guide for each platform you use.
"Sounds Like Me" vs. "Doesn't Sound Like Me", Three or four side-by-side example pairs covering the same idea, one in your voice, one in generic AI voice. This is the proof of work section, and if the examples don't feel clearly distinct, the analysis wasn't sharp enough.
The Portable AI Prompt Snippet: This is the deliverable most creators use immediately. A 200-400 word self-contained block of text you can paste into any AI tool's system prompt, custom instructions, or project knowledge. Read it as an AI instruction and it should produce noticeably different output than no instructions at all. Specific enough that a creator can look at a paragraph generated with it and recognize themselves.
The Portable Snippet in Practice
Once you have the snippet, you paste it into:
- Claude Projects: In your project's custom instructions, so every new conversation in that project writes in your voice
- ChatGPT Project: In the project instructions so any content you generate there starts from your voice
- Gemini Gems: In the system instructions for your custom Gem
From that point on, when you use any of those tools for captions, emails, scripts, or posts, the AI writes in the profile it learned from your actual content.
You can also drop the snippet into a single conversation as context: "Here's my voice profile, write in this style: [paste snippet]", and get a single piece of content that matches your voice without any permanent setup.
When to Build Your Voice Codex
Before scaling AI content production: If you're going to use AI to help write captions, newsletters, video outlines, or email sequences, building the codex first means every piece of output starts from your voice. The difference compounds over time.
When onboarding a team or editor: The codex doubles as a voice guide your team can use to write in your style without a lengthy approval process for every piece of content.
When you feel like AI content doesn't sound right: This is the most common trigger. The codex diagnosis usually reveals exactly what's happening: the AI is writing a generic 6 on the authority scale when you're naturally an 8, or it's using warm "we" language when you're more direct.
After significant content evolution: If your voice has changed over two years of creating, a codex built from old content will be off. Rebuild it with recent samples.
Who Gets the Most Out of This Skill
Creators who write across multiple platforms: Managing YouTube scripts, Instagram captions, newsletters, and LinkedIn posts means four different registers. The codex documents all of them in one reference.
Creators working with editors or ghostwriters: The voice codex is the brief. Handing it to someone new to your content replaces months of "this doesn't sound like me" feedback loops.
Creators who find AI content sounds flat: If your AI output consistently feels generic, this is the fix. The codex doesn't improve AI capability: it gives AI accurate instructions to follow.
Creators building a brand over time: Voice consistency compounds. A codex locks in what you've built so it doesn't erode as you experiment with new formats.
Pricing and Where to Get It
The Brand Voice Codex is $7, one-time. Works in Claude and ChatGPT: load the skill into your project, then paste in your content samples to generate the codex in one session.
Pair It With
- Long-Form Script System: Once your voice codex is built, the Long-Form Script System can use it to write YouTube scripts that sound like you authored them, not like AI filled in a template.
- Newsletter Conversion Engine: Apply your voice codex to newsletter copy to ensure your email list reads a consistent voice across every issue.
- Caption Chain Generator: The voice codex becomes the style layer for multi-platform caption chains, ensuring the repurposed version of a video sounds like you on every platform.
Generic output is what happens when an AI doesn't know who it's writing for. The Brand Voice Codex solves that problem once, and then the AI knows.
About the author
Content, CreatorSkills
The CreatorSkills team publishes practical guides on AI workflows for content creators.
About CreatorSkills
