← All AI output fixes

AI output doesn't sound like us

Last verified: Claude Fable 5 / GPT-5 era models · July 2026

The draft is fine. It just doesn't sound like you: it hedges, it reaches for buzzwords you would never say, it slips into American spelling, the register is off. So someone rewrites it line by line to sound like the company, and the time the AI saved goes straight back out. This page gives you a prompt that captures your voice as a reusable block, plus where to keep it so every draft starts in your voice.

Part 1 · The generator prompt

Paste this into a fresh Claude or ChatGPT chat. It interviews you about how your company sounds, then hands back a voice block you paste at the top of any writing chat so the output lands in your register instead of generic AI prose.

Generator prompt
You're going to build my company's VOICE BLOCK: a reusable instruction block I'll paste into any AI chat that writes for us, so the output sounds like our company without me rewriting it.

Interview me first, one question at a time. Where I don't know, propose a sensible default and move on:
1. What do we write most (client documents, emails, proposals, marketing, internal notes) and who reads it?
2. Register: formal or plain English? First person singular or plural? Do we address the reader as "you"?
3. Sentence and paragraph style: short and direct, or considered and detailed? Any habits we like (lead with the point) or hate (hedging, throat-clearing)?
4. Words and phrases we would never use (buzzwords, jargon, cliches), and any we always prefer.
5. Conventions: British or American spelling, how we write dates, numbers and currency, whether we use contractions.
6. Paste two or three sentences that sound exactly like us, and one that sounds wrong. If you have none, describe the difference.

Then output my VOICE BLOCK as one copy-paste block containing: (a) one line on who we are and who we write for, (b) register and person, (c) sentence and structure rules including lead-with-the-point and anything banned, (d) a banned-words list and any preferred words, (e) spelling, date, number and currency conventions, (f) a closing self-audit that rewrites any sentence that would not survive our style before showing me, then appends "Voice check: pass" or lists what it changed.

Keep it under 400 words so it fits any project-instructions field. After you output it, name the three ways the AI is most likely to still drift from our voice and how the block prevents each.

Part 2 · Where to keep it, and how to feed it examples

The block fixes the voice. This makes it automatic, and sharper over time.

  1. Save the voice block in your AI tool's project or custom-instructions field (a Claude Project, a ChatGPT project or custom GPT), so it applies to every message without pasting.
  2. Add two or three real paragraphs that sound exactly like you alongside the block. Examples teach voice faster than rules; the AI matches a sample better than a description.
  3. When a draft still reads off, don't rewrite it silently. Tell the AI which sentence was wrong and why, and add that rule to the block. The voice block gets more accurate each time.
  4. Keep one voice block per audience if you need to (client-facing versus internal), and name which one applies when you brief the AI.

Part 3 · A worked example

Here is the block the generator hands back, for our running example company, the fictional consultancy Fenwick & Cole.

Example: Fenwick & Cole, a boutique management consultancy
CONTEXT: Fenwick & Cole is a boutique management consultancy writing for mid-market retail and consumer clients. Most of what we write is client documents and proposals.

REGISTER: first person plural ("we"), plain professional English, address the reader as "you". Confident, not stiff. Contractions are fine.

STRUCTURE: lead with the point, then support it. Short sentences. One idea per paragraph. No throat-clearing openers ("In today's fast-paced landscape..."), no hedging ("it could be argued that"), no summarising what you are about to say before you say it.

BANNED WORDS: leverage, synergy, robust, seamless, delve, landscape, utilise, best-in-class, cutting-edge. Prefer plain equivalents: use, not utilise; strong, not robust.

CONVENTIONS: British spelling (organise, colour). Dates as 4 July 2026. Currency as £4,000. Numbers under ten in words, ten and above in digits. No emojis in client documents.

SELF-AUDIT: before output, rewrite any sentence that would not survive these rules. Append one line: "Voice check: pass", or list what you changed and why.

Common questions

Why doesn't AI sound like my brand?

The model writes to a generic average of everything it has read, because it has no record of how your company specifically sounds. It defaults to safe, hedged, buzzword-friendly prose. It will match your voice, but only once you give it your voice as an explicit instruction and, ideally, a few real examples to imitate.

How do I get AI to match my tone of voice?

Give it a voice block: register, sentence style, a banned-words list, spelling and number conventions, and two or three sentences that sound exactly like you. Tell it to rewrite anything off-voice before showing you. The generator prompt on this page builds that block, and pairing it with real examples matches your voice far better than rules alone.

How do I stop AI using buzzwords or American spelling?

Put the banned words and your spelling convention in the voice block explicitly (no "leverage" or "utilise"; British spelling), and add a self-audit line that makes the AI catch and fix them before output. Stored in your tool's project instructions, it applies to every draft instead of you catching the same words by hand each time.

The prompt is the aspirin. Stored context is the cure.

The voice block you just generated is exactly what an AI Brain stores as always-on context, tuned on your real writing. Paste it and this chat sounds like you; installed, every email, proposal and report does, because the voice lives in the system and improves as you correct it. This page proved it on your own words. The install is how your voice becomes the default output, not the thing you edit back in.

What else does AI get wrong that you fix by hand?Browse the other fixes →