Write AI prompts that sound like your business by adding a short voice block to each one: three to five traits, each with a line you'd say and one you wouldn't, two or three real messages you've written, a list of banned words, and who's reading. Save that block in a shared Project or Gem so everyone uses the same one.
Adjectives on their own don't work. "Friendly, professional and approachable" describes almost every business, so the AI falls back on its default voice: upbeat, wordy and fond of exclamation marks. What moves the output is specifics: sentences you've actually written, words you'd never use, a length limit, and a clear picture of the reader. The difference is obvious the moment you put two outputs side by side.
Why "friendly and professional" produces generic copy
Here's a typical prompt from an illustrative two-chair barber shop announcing late openings, and the kind of caption it produces.
Write an Instagram caption for our barber shop's new late-night
Thursday openings. Tone: friendly and professional.Illustrative output: "Exciting news! We're thrilled to announce late-night Thursdays at [shop name]! Now open until 9pm so you can look your best after a busy day. Book now and level up your look! #barber #fresh #latenight #grooming #style #mensgrooming"
Nothing in it is wrong, and nothing in it is theirs. Four exclamation marks, "thrilled to announce", a stock phrase about looking your best, and six hashtags when Instagram has capped posts at five since December 2025. The owner, whose own posts read like texts to a mate, would never write any of it. The AI wasn't told what the shop sounds like, only what almost every business claims to sound like.
The voice block, filled in for a barber shop
A voice block is a paragraph you paste at the end of any prompt for customer-facing text. It has five parts: who the business is, how it sounds (traits with examples), real writing samples, banned words, and format rules. Here's the barber shop's:
VOICE (paste at the end of any prompt for customer-facing text)
Business: [shop name], a two-chair barber shop. Most customers are
regulars who've been coming for years.
How we sound:
1. Plain-spoken. Say what's happening: "Late opening Thursdays, till 9."
Not: "Exciting news!"
2. Dry humour, at most once per message: "Your beard called. It wants
Thursday." Never joke about anyone's looks or money.
3. Short. Captions under 40 words, emails under 120, texts under 25.
4. Talk to regulars. Assume they know us.
5. No hype. No exclamation marks (one allowed in a birthday message).
Never use: thrilled, excited to announce, elevate, look your best,
game-changing, pamper, treat yourself, fresh new look, don't miss out.
Examples of our real writing (match the tone, not the words):
- "Back from holiday Tuesday. Chairs are open, kettle's on."
- "Sorry you waited 20 minutes on Saturday. That's not how we do
things, and the next one's on me."
- "Price going up to $32 from 1 March. Still cheaper than a bad haircut."
Format: one emoji at most. Three hashtags at most, lower case.Three things make it work. The traits come with example lines, so "plain-spoken" means something specific. The banned list names the exact phrases the AI reaches for by default. And the writing samples are real, from the owner's own posts and messages, with an instruction to match the tone rather than reuse the words; without that line, AI tends to recycle a sample's phrases. If you haven't worked out your traits yet, building a brand voice guide that AI can follow takes you through finding them.
Writing your own voice block in about 30 minutes
- Collect 20 messages you wrote yourself (10 minutes): posts, emails, texts and review replies from the past year. Skip anything drafted by AI or by an agency.
- Star the five best (5 minutes): the ones that got replies, bookings or kind words. These are your voice at its best, and three of them become your samples.
- List what they have in common (5 minutes): typical sentence length, words you use a lot, whether you use exclamation marks, how you sign off, whether you joke.
- Turn each pattern into a trait with a "we say / we don't say" pair (5 minutes).
- Add the banned words (5 minutes): generate two or three drafts without the block and list every phrase you'd never write. Those are the AI's defaults, and naming them is the quickest way to switch them off.
Here's the trait table an illustrative wedding planner produced from that exercise. It reads nothing like the barber's, which is exactly right.
| Trait | We say | We don't say |
|---|---|---|
| Calm and organised | "Here's what happens between now and June." | "Don't panic, we've got this!" |
| Warm, not gushing | "It was a lovely day to be part of." | "OMG, the most magical day ever!!" |
| Specific about next steps | "We'll send the supplier timeline by Friday." | "We'll be in touch soon." |
| Honest about limits | "We can't guarantee the weather, but we always have a wet-weather plan." | "Your day will be perfect, whatever happens." |
Four rows is plenty. Most voices come down to three or four traits; a block with ten gets applied unevenly, because the AI can't weigh them all at once.
Same request, before and after the voice block
The same caption request, with the voice block pasted underneath:
Illustrative output: "Late opening Thursdays starts this week. Chairs free till 9, so no more sprinting from work with half a fade. Book online or turn up and chance it. #barbershop #skinfade #thursday"
That's recognisably the shop: short, dry, specific. Still, run the usual checks before posting. Is it really "this week"? Does the booking system actually take walk-ins on Thursday evenings, or does "turn up and chance it" promise something the chairs can't deliver after 7pm? A good voice makes wrong facts more convincing, not less, so the voice block never replaces a fact check.
What changed wasn't the model or the request. It was 150 words of specifics about how the business talks. That block took the owner about 30 minutes to write, most of it spent scrolling his own old posts for the three best examples.
Three more prompts with the voice built in
Each prompt below ends with the voice block. The outputs are illustrative, and each one still needed a fix, which is the point: voice gets you close, and you finish.
Replying to a mixed review
Reply to this Google review for [shop name]. Thank them, answer the
specific point they raised, and don't argue. Under 50 words. If they
complain about waiting, apologise once and don't offer compensation.
Review: "Good cut but waited 25 mins past my booking. 3 stars."
[VOICE BLOCK]Illustrative output: "Thanks for the honest review. Twenty-five minutes past your booking isn't on, and I'm sorry. Saturday got away from us. Glad the cut was right, and next time you'll be in the chair on time."
What to fix: the AI guessed "Saturday"; the booking record says it was a Friday, so that goes. And "next time you'll be in the chair on time" is a promise the shop can't guarantee. The edited version ends: "We're keeping a gap between bookings on busy days now." Voice intact, no invented facts, no promise it can't keep.
A reminder text in a tattoo studio's voice
Voice blocks differ by business, and a quick contrast shows why. An illustrative tattoo studio's traits are calm, precise and reassuring: it explains what happens next, never jokes about pain, and keeps aftercare language plain. Its block produces a very different register from the barber's for the same kind of job.
Write a text reminding a client of their tattoo appointment tomorrow at
2pm with [artist]. Include: eat beforehand, bring photo ID, the $50
deposit is paid. Moves within 72 hours use the deposit. Under 45 words.
[TATTOO STUDIO VOICE BLOCK]Illustrative output: "Hi [first name], see you tomorrow at 2pm with [artist]. Please eat beforehand and bring photo ID. Your $50 deposit is paid. If something's come up, just reply and we'll sort it out."
What to fix: "We'll sort it out" softens the rule into something the client might read as a free reschedule. Precise was one of the traits, so the last line becomes: "If you can't make it, reply and we'll talk it through. Moves this close to the day use the deposit."
Announcing a price rise to regulars
Write a short email to our regulars: a cut goes from $30 to $32 on
1 March, because rent and product costs have gone up. Under 90 words.
No long apology, no hype. Sign off from [owner's first name].
[VOICE BLOCK]Illustrative output: "Hi [first name], a quick heads-up: from 1 March a cut goes from $30 to $32. Rent and product costs have both gone up, and we'd rather tell you straight than cut corners. Kids' cuts and beard trims stay the same. See you in the chair. [Owner's first name]"
What to fix: "Kids' cuts and beard trims stay the same" wasn't in the prompt. The AI added a reassuring detail that sounds exactly like something the shop would say. On the price list, beard trims are also going up by $1. The sentence goes, and the owner adds the real change. For customer email replies more generally, drafting customer email replies with AI that sound like you covers the inbox side.
Where the voice block goes in a prompt, and why order matters
Put the parts of a prompt in this order: the task, the facts, the constraints, then the voice block, then the output format. Facts come before voice for a reason. When the voice instructions come first and the facts are thin, the AI tends to invent details that fit the voice, which is how the price-rise email above gained a reassuring line about beard trims. Giving it the facts first, with an explicit "use only these facts", keeps the voice working on the wording rather than the content.
A skeleton that works for most customer-facing jobs:
TASK: [what to write, for whom, on which channel]
FACTS (use only these, add nothing):
- [fact 1]
- [fact 2]
CONSTRAINTS: [length, what not to promise, anything to avoid]
[VOICE BLOCK]
FORMAT: [one paragraph / three bullet points / subject line + body]If the output still invents something, add the invented item to the constraints as a "don't mention" line and rerun. After a few rounds you'll know which details your AI tends to add for your kind of business.
Picking the real examples you paste in
The writing samples do more work than the traits, so choose them with care:
- Use your best writing, not your most recent. Scroll back for the messages customers replied to warmly.
- Cover different jobs: one announcement, one apology, one everyday message. A block with three announcements makes every reply sound like an announcement.
- Never use AI-written text as a sample. You'll teach it its own habits.
- Remove customer details before pasting: names, numbers, anything personal.
- Keep them short. One to three sentences each. Long samples get copied rather than imitated.
If the output starts repeating a sample's exact phrases ("kettle's on" in every message), swap that sample out or add "don't reuse phrases from the examples". It's a sign the samples are carrying the voice on their own.
One voice, adjusted for each channel
A voice isn't one fixed register. The barber shop sounds like itself in a text and in an email, but not identically. Add a short channel table to the block so the AI adjusts length and formality without losing the character:
| Channel | Adjustment | Barber shop example |
|---|---|---|
| Text message | Under 25 words; no hashtags; first name only | "Hi [first name], you're booked Thursday 7:30 with [barber]. Reply C to cancel." |
| Instagram caption | Under 40 words; one joke allowed; three hashtags at most | "Late opening Thursdays starts this week. Chairs free till 9." |
| Under 120 words; a clear subject line; signed by a person | "Subject: Thursday late opening, till 9" | |
| Review reply | Under 50 words; no jokes if the review is negative | "Sorry about the wait. We're keeping a gap between bookings on busy days now." |
| Website copy | Fuller sentences; still no hype; facts first | "Open late on Thursdays until 9pm. Walk-ins welcome when a chair is free." |
The rule of thumb: the traits never change, only the length, the format and how much humour the moment can take.
Keeping one shared voice in Projects, Gems or skills
A voice block pasted from someone's notes app drifts within weeks: one person trims it, another adds their own favourite words. Keep a single copy where the whole team's AI tool reads it automatically.
- ChatGPT Projects: put the voice block in the project's instructions and upload your writing samples as files. On Business, Enterprise and Edu, a project can hold up to 40 files and be shared with colleagues by email or link, with up to 100 collaborators. OpenAI's help article on projects has the current sharing options.
- Claude: project instructions hold the block for a shared project, and they're available on every plan, including free accounts (limited to five projects). Claude's skills can also carry tone and format rules, including patterns taken from your own writing, and switch on when a task calls for them. Setting up Claude Projects as a shared team assistant walks through the setup.
- Gemini Gems: save the block as a Gem's instructions. When you share a Gem, recipients can see its instructions and uploaded files, so don't put anything in it you wouldn't show the whole team. Building Gemini Gems for repeat business tasks covers the rest.
- Gmail: since May 2026, Help me write can match the tone of your past emails on Business Starter, Standard and Plus and on Google AI plans. It's useful for quick replies, but it learns from everything you've sent, including your rushed emails, so keep the voice block for anything important.
Don't build a new custom GPT for this. OpenAI is retiring custom GPTs across ChatGPT plans, and they stop running on 11 December 2026. OpenAI's migration turns a GPT into a plugin, with its instructions becoming a skill. If your voice currently lives in a custom GPT, copy the instructions into a project now. And keep the master copy of the block in a shared document you control as well, so it outlives any one tool; a shared prompt library for your team is the natural home.
A blind test to see whether the voice stuck
You're too close to your own writing to judge. Run a blind test instead:
- Generate three messages with the voice block: a caption, an email and a reply.
- Take three real messages you wrote for similar jobs.
- Mix all six, remove anything that gives the game away, and ask a colleague or a friendly regular to pick the three AI ones.
If they spot all three instantly, the block needs work, and their reasons tell you where ("too many adjectives", "you never say 'lovely'"). If they're guessing, it's doing its job. Alongside the test, score each AI draft against a simple sheet:
| Check | Passes if |
|---|---|
| Banned words | None appear |
| Exclamation marks | Within your rule |
| Length | Inside the limit for that format |
| Specific detail | At least one concrete fact (a time, a price, a day) rather than a general claim |
| Closing line | No stock call to action you'd never write yourself |
| Read aloud | Sounds like the owner on a good day |
| Facts | Nothing added that wasn't in the prompt or your sources |
Why the voice drifts, and how to pull it back
Even a good block weakens over time. The usual causes:
- Long conversations. After many back-and-forth turns, instructions from the start carry less weight. For anything customer-facing, start a fresh chat inside the project.
- Model updates. When a vendor changes its default model, the default style changes with it. Rerun the blind test after any major update you notice.
- Edits to the block. Give the master copy one owner and a "last updated" date.
- Staff overriding it. Someone adds "make it more exciting" to a prompt and the hype returns. Put your two or three non-negotiables at the very top of the block, where they're hardest to ignore.
Review the block every three months with fresh samples of your best recent writing. The voice of a business shifts too, and the block should follow the business, not the other way round.
Further reads
- Tone of Voice Examples: AI Copy Before and After Editing — More before-and-after edits of AI copy for tone.
- How to Write a House Style Guide Your Team and AI Both Follow — Turn your voice block into a house style guide.
- How to Write Instagram Captions With AI That Sound Like You — Captions that sound like you, post after post.
- Custom GPT vs Claude Project vs Gemini Gem: Which Should You Use? — Choose where your shared voice should live.
- Brand Voice Tools Compared: Jasper, Writer and Custom Instructions — Dedicated brand voice tools against plain instructions.
- ChatGPT Prompts for Small Business Owners: 50 Tested Examples — Fifty tested prompts to pair with your voice block.
- How to Give AI Your Business Context So Answers Fit — Build a one-page business profile and task instructions, store them where each AI tool reads them, and test that answers fit your business.
- ChatGPT Prompts for Pet Sitters and Dog Walkers That Win Clients — Twelve ready-to-paste prompts for pet sitters and dog walkers, from profile bio to daily updates, each with what to feed in and what to check.
- How to Write a Salon Service Menu and Price List With AI — Price each service from your chair-hour cost, then use three AI prompts to structure, describe and stress-test the menu before it goes live.
- Shopify Magic vs ChatGPT for Product Descriptions — Shopify Magic is free and sits in the description box; ChatGPT handles bulk rewrites and brand voice. A decision table, a fair test and sample outputs.
- Can a B&B Use an AI Chatbot and Keep the Personal Touch? — Which guest messages a B&B chatbot should answer, which ones only you should, and how the bot can make your own replies more personal, not less.
- Using AI in a Yoga or Pilates Studio While Keeping Your Voice — Why AI makes every studio sound the same, and the voice sheet, before-and-after edits and teacher-bio method that keep yours recognisably yours.
- Should You Let AI Auto-Post Replies to Google Reviews? — A sorting rule for which Google reviews an AI may answer on its own, why star ratings make a poor trigger, and the guardrails for the auto-post lane.
- How Agencies Keep Brand Voice Consistent in AI-Written Content — A filled-in voice card, two client voices side by side, a sample-led prompt and a five-line scoring rubric for keeping AI copy on-brand.
- How Freelance Writers Use AI Without Losing Their Voice — Keep control of the angle and first draft, then use targeted prompts to find weak passages without flattening your writing.
- How Course Creators Use AI to Build a Course in Weeks, Not Months — A week-by-week plan: AI handles outlines, scripts, edits and quizzes; you supply the expertise, examples and recording. Six weeks for a 6-10 hour course.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI help articles on projects in ChatGPT and on the custom GPT retirement; Claude help article on personalisation (instructions, project instructions, skills); Google help on sharing Gems; Gmail Help me write tone matching and Instagram's hashtag limit from the site's verified fact sheet. Checked September 2026.