Blog
5 min read
Turning Homeschool Coaching Into a Bookable Product: International School of America
Jun 12, 2026
Your business already has a voice — it's in the email clients forward to their spouses, the review reply that made a stranger trust you, the way you explain a price without flinching. What it probably doesn't have is that voice written down. That used to be a nice-to-have. Now an AI drafts your posts, your replies, and your newsletters every week, and it answers the question "what does this business sound like?" with whatever you gave it. Vague in, beige out.
Quick answer: Mine your own best writing for the voice you already have, write it down as three traits with do/don't example pairs plus a banned-words list — two pages, not a brand bible. Then install it twice: into your AI's workspace memory so every draft starts in your voice, and into your team's onboarding so humans follow the same page. Review stays human either way.
Why "professional but friendly" fails twice
Most voice guidance fails because it describes every business: professional, friendly, authentic, passionate. Nobody aims for unprofessional and hostile. Vague traits fail humans slowly — each hire interprets them their own way, and the voice drifts one inbox at a time. They fail AI instantly: a model told to be "professional but friendly" produces the statistical average of everyone else's professional-but-friendly, which is exactly the LinkedIn-flavored beige you've learned to scroll past.
The fix is the same one we keep arriving at for profiles and policies: decision rules, not adjectives. A voice guide earns its existence the same way — every line must be able to change a sentence. "We're warm" changes nothing. "We use the customer's first name once, never twice" rewrites actual drafts.
Find the voice you already have
Don't invent a voice in a workshop; excavate the one that's working. Three exercises, one afternoon:
- Pull your five best pieces of writing. The email a client replied to with "this is why we hired you." The review response that won the audience. The text that defused the angry Tuesday. Highlight the phrases only your business would say — those are your raw material.
- Run the bar test. For each sentence you're tempted to put in marketing: would you say it out loud to a customer at a bar? "We leverage cutting-edge solutions" fails. "We'll have it sorted before the weekend" passes. Your spoken register is usually your written voice trying to get out.
- List the banned words. Every business has them — the words that appear in your drafts but never in your mouth. Leverage, solutions, excited to announce, best-in-class, journey (the metaphor, not the email kind). The banned list does more work per line than any other section, because it's the part both humans and AI can apply mechanically.
Write it as rules and pairs — two pages, max
The guide's structure matters less than its texture: every rule travels with a before/after pair, because examples are the only instruction both a new hire and a language model reliably follow.
| Section | What it holds | Example |
|---|---|---|
| Three traits | Each with a do/don't pair | "Plain-spoken: 'It'll cost $400 to fix properly' — not 'pricing varies depending on several factors'" |
| Vocabulary | We say / we never say | "We say 'clients,' never 'users.' We say 'the fix,' not 'the solution.'" |
| Mechanics | The fiddly calls, decided once | Contractions yes; exclamation points one per message; emoji in social, never in invoices |
| Channel dials | Same voice, different volume | "Social is us at the bar; email is us across the desk; invoice notes are us in writing — same person everywhere" |
Two pages is the ceiling for a reason: a guide nobody can hold in their head is a guide nobody applies on a busy Thursday. If a rule never changes a real draft, cut it — the voice guide obeys its own bar test.
Install it where the AI works
A voice guide in a drive folder governs nothing — the same filing-cabinet problem as every other document, with higher stakes because the AI ships words daily. Installation, not existence, is what changes output:
- Put it in workspace memory. Your workspace's AI memory is where standing facts about your business live — the voice guide belongs there, so every chat and draft starts from it instead of from the internet's average. For one-off work with extra nuance, attach the guide as context directly.
- Bake recurring formats into skills. The formats you produce weekly — newsletter sections, booking confirmations, social captions — deserve more than re-explaining each time. A workspace AI skill encodes the format and the voice once: "newsletter section, our voice, 120 words, one practical tip, no exclamation points." When the voice guide evolves, review and publish the skill change once and every future draft inherits it.
- Watch it propagate. The month-of-social workflow we've covered — theme in, drafts out — quietly becomes a different product when the drafts arrive pre-voiced: review shifts from "rewrite everything" to "check the judgment calls," which is the entire economics of AI content that still sounds like you.
The human half, and the drift audit
The same two pages are your team's writing onboarding — which is the point: one document, two readers. A new hire answering customer email applies the same banned list and bar test the AI does, and when either drifts, you point at the page instead of having a vague conversation about tone.
Two standing disciplines keep it alive. First, the review gate doesn't retire: AI drafts get human review before publishing, because voice-correct and judgment-correct are different properties — the draft can sound exactly like you and still promise the wrong deadline. Second, run a quarterly drift audit: reread one month of outbound — newsletters, social, review replies, AI and human both — in one sitting, against the guide. Drift is invisible message-by-message and obvious in bulk; the same trick as reading a month of reviews as one document. Update the guide with what you find — a voice guide that never changes isn't stable, it's abandoned.
Key takeaways
- Vague voice fails AI instantly: "professional but friendly" produces the internet's average of professional-but-friendly — vague in, beige out.
- Excavate, don't invent: your five best client emails, the bar test, and a banned-words list surface the voice that's already working.
- Rules with pairs, two pages max: every line must be able to change a sentence, and every rule travels with a before/after example — the only instruction humans and models both follow.
- Installation beats existence: workspace memory makes every draft start in your voice; skills bake voice into recurring formats once, and publish updates to everything downstream.
- One document, two readers: the same guide onboards your hires and your AI — when either drifts, you point at the page.
- Review and audit stay human: voice-correct isn't judgment-correct, and drift is only visible in bulk — reread a month of output quarterly against the guide.
Frequently asked questions
What if I genuinely don't have good writing to mine yet?
Then mine your mouth instead. Record yourself explaining your service to a friend — what you do, what it costs, why the cheap alternative disappoints — and transcribe it. The phrasing in that transcript is your voice; the guide's job is just protecting it from what happens when you sit down to "write marketing." New businesses don't lack a voice; they lack the confidence that the talking voice is allowed in print. It is — it's the whole advantage.
Should the AI sound exactly like me, or just consistent?
Aim for "unmistakably from your business," not "forensically you." A voice guide gets a model inside the right fence — your vocabulary, your register, your banned list — and that's what readers actually notice. The last few percent of personal idiosyncrasy comes from your review pass, where you'll find yourself changing a phrase or two per draft. That division is healthy: the guide does consistency at scale, you do personality at the margin.
We're three people with three styles. Whose voice wins?
The business's — which is usually the founder's voice, sanded. Personal style survives in 1:1 relationships (clients know which of you they're talking to), but everything that ships under the business's name — site, newsletter, social, review replies — follows the guide. The practical compromise: write the guide from the best existing material regardless of author, then let everyone keep their signature in their own client threads. What you're preventing isn't individuality; it's the website sounding like three different companies.
How is this different from just editing the AI's drafts harder?
Editing fixes one draft; the guide fixes the generator. Without it, every draft starts at the internet's average and your review pass does the same heavy rewrite forever — which is how AI workflows quietly die of editing fatigue. With the voice installed upstream, drafts start close and review gets cheap, which means it actually keeps happening. If you find yourself making the same correction three times, that correction is a guide line waiting to be written down.
Ready to write the two pages once? Faster gives your voice somewhere to live — workspace AI memory, reusable skills for your recurring formats, and a review step before anything ships. Start free and run the bar test on your homepage tonight.