Your voice isn’t a vibe.It’s a spec.
Every team says it has a brand voice. Then a draft comes in and the only feedback is “feels a little off,” followed by a heroic rewrite by the one person who actually holds the voice in their head. The drafts get worse the more AI tools you add to the workflow, because the assistants default to corporate filler unless something tells them not to.
The Brand Voice Engine codifies your voice into rules a machine can check — not adjectives it has to guess. Five prompts (Extractor, Style-Guide Builder, On-Brand Writer, QA Checker, Voice Transfer), one 8-section portable Voice Spec, and a 100-point rubric that turns “feels off” into a score and a flagged-line list. One-time $59.
“Sound on-brand” is the least enforceable instruction in marketing.
Most brand-voice work lives in a PDF style guide written for a human designer, plus a folder of past work that the one writer who actually holds the voice points new hires at when something feels off. It worked when the writing was all done by humans. It doesn’t work when half of your content goes through an AI assistant first.
The default behavior of every generic AI prompt is corporate filler — “leverages cutting-edge,” “in today’s fast-paced economy,” “empowers teams to seamlessly,” “robust solution,” “delve.” The model isn’t lying — it’s doing what a generic prompt asks for. The fix is at the prompt layer: codify the voice into specifics and contrast pairs and banned words that a model can actually check, and enforce them at QA time.
The least actionable note in copyediting. A PDF style guide and a vibe-check don't tell the writer (or the model) what specifically broke. The QA Checker scores the draft 22/100 and quotes the five exact lines that miss.
“Leverage,” “seamless,” “robust,” “in today’s fast-paced,” “delve” — the words that sound competent in a vacuum and announce “AI wrote this” the moment they hit. Banned-words lists make these hunt-able instead of inevitable.
Anyone can ship the same offer. Voice is the part competitors can't copy and AI tools can't default into. Making it specific enough to enforce is the difference between sounding like every other newsletter and sounding like you.
Clear about the lane. No inflated promises.
- A five-prompt engine plus an 8-section portable Voice Spec template — your voice codified into rules a machine can check.
- A 100-point QA rubric that turns “feels off” into a score, a banned-words hit-list, and quoted off-brand lines with rewrites.
- Model-agnostic — works in Claude, ChatGPT, Gemini, or any chat-capable assistant.
- Three run modes — paste-and-go, hands-free Claude Skill, or shared Claude Project for whole-team consistency.
- Built around evidence (your samples) rather than abstract adjectives — the Voice Extractor explicitly refuses to invent traits the samples don't support.
- 12 months of prompt updates as the engine evolves.
- A content management system. Use Notion / Webflow / WordPress / your CMS.
- A grammar / spelling checker. Use Grammarly / Hemingway / your editor's built-in.
- A branding agency. The kit is the operating layer; positioning, naming, and identity work still benefit from a specialist.
- An AI-detection tool. Different job; this enforces your voice rather than scoring how AI-like a draft is.
- A subscription. One-time $59 with 12 months of prompt updates.
- A guarantee your team uses it. Adoption is on you; the kit is engineered to make the right thing the easy thing.
Four rules. Built into every prompt.
The engine’s IP isn’t the prompts; it’s the four rules below that make them work. The prompts are the packaging — these are the discipline.
"Friendly" is useless — every brand thinks it's friendly and no model can apply it. "Uses contractions, second person, one-line paragraphs, no exclamation points" is checkable. The spec is built from specifics so the model has rules to follow, not vibes to guess at.
"Direct, NOT blunt" tells a model where the line is in a way "direct" alone never will. The 4-5 contrast pairs in the attributes section do most of the heavy lifting — they mark exactly where each attribute stops being on-brand and starts being something else.
The fastest way to sound generic is the filler everyone uses — "leverage," "seamless," "robust," "in today's fast-paced world," "delve." Name them so the QA Checker can hunt them. A banned-words list is one of the highest-leverage sections of the entire spec.
"Feels off" is not a verdict; a flagged line with a fix is. The QA Checker quotes every off-brand line, names the rule it breaks, and provides an on-brand rewrite. The output is actionable, not advisory.
Eight sections. One page. The portable artifact.
The Voice Spec is the product. Everything else uses it. Fill it once (or let the Voice Extractor fill it for you), then paste it into any prompt, Claude Project, custom GPT instructions, or hand it to a writer. Designed to stay under one page so a model — and a human — actually use it.
The north star. If a draft violates this, nothing else matters.
4-5 "We are X, not Y" pairs. The "not Y" half marks exactly where the line is.
Phrases you actually use, and the filler / cliches / AI tells to hunt. The QA Checker scans for both.
Length, structure, openers, fragments, what to avoid. The shape of how you write, codified.
Headers, lists, emoji policy, capitalization, oxford comma, punctuation habits, links/CTAs.
Person (I / we), tense, how you address the reader, stance (peer / authority / coach).
The 2-4 recognizable tics that make it unmistakably you. Hard to fake; impossible to ignore.
The non-negotiables. "Never overpromise," "Never fabricate a stat," "Never bury the point past the first sentence."
One engine. Five jobs.
Two starred prompts do most of the work — the Voice Extractor (run once to build your spec) and the Voice QA Checker (run on every draft before it ships). The other three solve the writing, documentation, and cleanup jobs that come up around them.
Run this first. Paste 3-5 samples of writing that sound unmistakably like you; out comes a structured Voice Spec calibrated to the actual evidence in your samples. The prompt explicitly forbids inventing traits the samples don't support.
Expands the Voice Spec into a one-page style guide a new hire or freelancer can follow without you explaining anything. DO / DON'T examples per rule, plus a 10-item pre-publish checklist drawn from the hard rules and banned words.
Writes new content (LinkedIn post, email sequence, landing hero, whatever) using the spec exactly — attributes, vocabulary, rhythm, formatting, POV, signature moves. Ends with a one-line self-check naming which spec rules it leaned on most.
The enforcement layer. Scores any draft against the spec on the 100-point rubric, flags banned words, quotes every off-brand line with the rule it breaks and an on-brand rewrite, identifies the top 3 fixes, and ships a fully rewritten 90+ version with the new score.
De-robotize existing generic / AI / vendor copy in one pass. Changes only how it sounds — not the facts, offer, or structure. Strips banned words, applies the rhythm and signature moves, keeps it the same length or shorter, and lists the 3 biggest changes it made and why.
Five dimensions. One score. Cited fixes.
The QA Checker scores a draft against the spec on five dimensions worth 100 points total. Banned-word hits, off-brand lines, and weak rhythm all cost real points. The output isn’t a vibe; it’s a number, a list, and a rewrite.
Does the draft feel like the voice as defined by the contrast pairs in the spec? Each attribute that lands earns points; each pair where the draft tips into the “not Y” side loses them.
Signature phrases used where they fit, banned words avoided. Banned-word hits cost real points — the rubric explicitly refuses to pass filler just because it's grammatical.
Length distribution, structure variety, openers, fragments where called for, and the “no sentence over X words” rule from the spec actually being honored.
Headers, lists, emoji policy, capitalization, oxford comma, punctuation habits, link/CTA structure — applied exactly as the spec defines them.
Person (I / we / you), tense, stance (peer / authority / coach) — consistent with the spec across the whole draft, no slipping into third person mid-paragraph.
Match score, banned words hit, every off-brand line quoted with the rule it breaks and an on-brand rewrite, the top 3 highest-impact fixes, and a fully rewritten 90+ version. Worked example: 22 → 92 on a generic AI draft.
The integrity moat.
Exactly what you get for $59, and what you don’t.
- Five prompts — Voice Extractor, Style-Guide Builder, On-Brand Writer, Voice QA Checker, Voice Transfer.
- 8-section Voice Spec template (the portable artifact).
- 100-point QA rubric (Tone 35 / Vocab 25 / Sentence & rhythm 20 / Formatting 10 / POV 10).
- Bundled SKILL.md for hands-free voice + QA in Claude.
- Quickstart guide for first-time spec extraction.
- Worked example: samples → spec → 22/100 generic draft → 92/100 on-brand rewrite.
- 12 months of prompt updates.
- A CMS or publishing platform. Use Notion / Webflow / WordPress / your CMS.
- Grammar or spelling checking. Use Grammarly / Hemingway / your editor.
- A branding agency or positioning consultancy. Different specialty.
- An AI-content detector. The job is to enforce your voice, not score AI-likelihood.
- Automated publishing or scheduling. The prompts produce drafts; you ship them.
- A subscription. One-time $59 with 12 months of updates.
Three run modes deserve their own callout — the kit deploys equally well as a paste-and-go prompt for any assistant, as the bundled SKILL.md installed into Claude (hands-free voice + QA), or pasted into a Claude Project ($69) for shared whole-team voice. The Claude Project path is the one most teams settle into.
Paste the Voice Spec at the top of any prompt, then the Writer / QA / Transfer prompt under it. Zero setup, any assistant. The fastest path to first value.
Install the bundled SKILL.md into Claude.ai (Settings → Capabilities → Skills) or your Claude Code skills directory. Just say "write this in our voice" or "does this sound like us?" and the skill fires hands-free against whatever Voice Spec is in scope.
Paste the Voice Spec into a Claude Project's custom instructions. Every chat in that Project inherits the voice automatically. Share the Project on Team / Enterprise and the whole team writes in one voice without anyone pasting anything.
Pairs naturally with the Founder’s Positioning Forge ($59) — Positioning defines the strategy (what you say, to whom, why it matters); the Voice Engine defines how you sound saying it. Together they form the complete narrative spine.
For agencies shipping client work across multiple brands, pair with the Agency Operators Skills Pack ($89) — one Voice Spec per client, the QA Checker runs at the “before we send it” step of every status report and deliverable.
The questions teams actually ask before codifying their voice.
Five prompts that form the engine — Voice Extractor (turns 3-5 of your best writing samples into a structured Voice Spec), Style-Guide Builder (expands the spec into a one-page guide a new hire can follow), On-Brand Writer (writes new content using the spec), Voice QA Checker (the 100-point rubric — scores any draft, flags off-brand lines, rewrites them), and Voice Transfer (de-robotizes existing generic copy by applying the spec). Plus the 8-section Voice Spec template — the portable artifact you fill once and reuse everywhere. Plus a quickstart guide and a worked example showing samples → spec → 22/100 generic draft → 92/100 on-brand rewrite.
Any capable assistant — Claude, ChatGPT, Gemini, or anything else with a chat interface. The prompts are model-agnostic. The Voice Spec is a portable plain-text artifact you paste at the top of any prompt, drop into a Claude Project, paste into a custom GPT's instructions, or hand to a human writer. For Claude users, the bundled SKILL.md installs hands-free voice + QA — just say “write this in our voice” or “does this sound like us?” and the skill fires.
A traditional brand guide is written for a human designer or copywriter. The Voice Spec is written for both a human AND a model — every rule is specific enough that an AI assistant can apply it correctly without guessing. “Friendly” is useless to a model (and arguably to a human); “uses contractions, second person, one-line paragraphs, no exclamation points” is checkable. The QA Checker enforces the rules by scoring a draft against the rubric and flagging specific lines, which a PDF style guide can't do. Same purpose, very different operating model.
Yes — fill the Voice Spec template by hand using the filled mini-example as a guide. Even a rough spec written from scratch beats “sound on-brand” as an instruction. The contrast pairs (“direct, NOT blunt”) and banned-words list are the highest-leverage sections; if you only fill those two, you've already moved the needle. Then run the On-Brand Writer for a few drafts, look at what feels right, and refine the spec from the patterns you actually like.
Five dimensions worth 100 points total: Tone & attributes match (35 points), Vocabulary — signature used and banned avoided (25 points), Sentence & rhythm (20 points), Formatting & mechanics (10 points), Point of view (10 points). Output includes the match score, the banned words hit, every off-brand line quoted with the rule it breaks and an on-brand rewrite, the top 3 highest-impact fixes, and a fully rewritten on-brand version with the new score. The worked example shows a generic AI draft scoring 22 and the rewrite scoring 92 — that gap is the product.
Common case — and the kit handles it two ways. Path 1: extract a single brand-level Voice Spec from samples that span the team's best work; the spec captures the shared spine. Path 2: build separate Voice Specs per channel or per writer (a CEO LinkedIn spec, a product blog spec, a support reply spec) and route each piece of content to the right one. The QA Checker scores against whichever spec you paste, so the routing is just prompt hygiene. The Claude Project deployment makes this trivial — one Project per voice.
That's the canonical workflow. Generate a draft with the On-Brand Writer (or with any other tool), run the Voice QA Checker on the draft, it produces the rewritten on-brand version. Most teams use this for content they didn't generate themselves — a freelancer's draft, a generic AI tool's output, an old marketing-page paragraph — to bring it onto the spec without rewriting from scratch. The Voice Transfer prompt does the same job in one pass when you don't need the scored audit.
30-day no-questions refund. Extract a Voice Spec from your real samples, run the QA Checker on five drafts your team has shipped recently. If the rubric doesn't flag at least one off-brand pattern you'd quietly been tolerating — and if the rewritten versions aren't tangibly more like your best work — email and we refund. Refunds across the catalog are countable on one hand; for voice work specifically, the QA Checker earns its keep on the first scored draft.
Stop saying “feels off.”
Start shipping 92s.
Five prompts, one portable Voice Spec, a 100-point rubric, and three run modes. The worked example takes a generic AI draft from 22 to 92 in one QA pass. The product is the discipline; the prompts just enforce it.
Sold by RedHub AI LLC · Secured by Stripe · redhub.ai