Why AI Writing Sounds the Same for Everyone
Ask five different business owners to write a product launch email with ChatGPT, and you'll often get five emails with the same rhythm: a punchy opening line, three benefit bullets, and a closing nudge that says something like "don't miss out." The words change, but the voice doesn't. That's because most people prompt AI with instructions about what to write, not how they sound when they write it.
The fix isn't a magic prompt. It's a small, repeatable process: collect real examples of your voice, turn them into rules, build those rules into a reusable prompt, then test and refine. This article walks through that process in order, with templates you can copy today.
Step 1: Build a Voice Reference File
Before you can teach AI your voice, you need to see it clearly yourself. Most business owners have never written down how they sound. It just lives in their head.
Start a single document (a Google Doc works fine) and paste in 8 to 12 examples of writing that actually sounds like you. Pull from:
- Emails you wrote yourself, not ones a freelancer drafted
- Social posts that got genuine replies, not just likes
- Text messages or Slack messages to customers (with names removed)
- Any page on your site you're proud of
- A voicemail or call transcript, if you have one, turned into text
Try to include a range: something persuasive, something informational, and something casual. A single polished sales page won't teach AI how you sound when you're explaining a late shipment.
Label each example with context: "This is a cold outreach email that got a reply" or "This is how I talk to customers who are upset." Context matters as much as the text itself, because voice shifts slightly by situation, and you want AI to understand that range, not flatten it.
Step 2: Turn Examples Into Explicit Rules
This is the step most people skip, and it's the one that matters most. AI can't infer your voice from examples alone nearly as well as it can follow explicit instructions. You need to translate "this sounds like me" into rules a model can actually apply.
Read through your reference file and answer these questions in writing:
- Sentence length. Do you write short, punchy sentences, or longer ones with asides?
- Contractions. Do you say "don't" and "we're," or do you spell everything out?
- Humor. Is there any, and what kind (dry, self-deprecating, none at all)?
- Words you'd never use. Every brand has them. Maybe it's "synergy," maybe it's "amazing," maybe it's any exclamation point.
- How you open. Do you lead with a question, a statement, a bit of empathy?
- How you close. Hard call to action, soft suggestion, or just information with no push?
Write the answers as a short list of rules, not a paragraph. For example:
- Sentences under 20 words, mostly
- Always use contractions
- One light joke per email, max
- Never use "excited to announce" or "game-changing"
- Open with a direct statement, not a question
- Close with a specific next step, not "let us know if you have questions"
This list becomes the backbone of every prompt going forward.
Step 3: Build a Reusable Prompt Template
Once you have examples and rules, combine them into a single prompt template you reuse every time, instead of rebuilding instructions from scratch. A template that works well in our experience looks like this:
Voice context: "You are writing as [business name]. Here are the voice rules: [paste your rule list]. Here are three examples of this voice in action: [paste two or three short examples]."
Task: "Using that exact voice, write [specific thing: a 100-word product description, a follow-up email, a social caption]."
Constraint: "Do not use these words: [your banned word list]. Keep sentences short. Avoid generic marketing phrases like 'take your business to the next level.'"
Save this template somewhere you can paste it quickly: a note app, a doc, or a prompt library if your AI tool has one. The goal is to never start from a blank prompt again. Every piece of writing should start from the same voice foundation.
A Before and After Example
Here's what a generic prompt produces versus a voice-trained one, using a plumbing company as an example.
Generic prompt: "Write a social post announcing we now offer emergency weekend service."
Typical output: "Exciting news! We're thrilled to announce that we now offer emergency weekend service. Don't let a plumbing problem ruin your weekend, call us today!"
Voice-trained prompt (with rules like "short sentences, no exclamation points, dry and practical tone"): "We now do weekend emergency calls. A burst pipe doesn't care what day it is, and now neither do we. Call anytime."
Same information, completely different feel. The second one sounds like a person who fixes pipes, not a template.
Step 4: Run the Swap Test
After you generate a few pieces of content, run what we call the swap test. Take the AI output and swap the business name for a competitor's. If the writing still sounds believable under a different name, it's generic, no matter how good the grammar is.
This test is fast and brutally honest. Run it on:
- Email subject lines
- About page copy
- Social captions
- Product descriptions
If something fails the swap test, go back to your rule list and figure out what's missing. Usually it's one of two things: the rules are too vague ("friendly tone" isn't specific enough) or the examples you fed it were already a little generic to begin with.
Step 5: Keep a Feedback Log
Voice training isn't a one-time setup. Keep a short running log, even just five or six lines, every time you edit AI output. Write down what you changed and why. Over a few weeks, patterns show up: maybe you keep removing the word "seamless," or you keep shortening the opening line.
Feed those patterns back into your rule list every month or so. This turns editing time into training time instead of just cleanup.
Quick Summary: Steps to Train AI on Your Voice
- Collect 8 to 12 real examples of your own writing across different situations
- Turn those examples into a short, explicit list of voice rules (sentence length, banned words, tone, openings, closings)
- Build one reusable prompt template that includes your rules and examples every time
- Run the swap test: if the writing works under a competitor's name, it's too generic
- Keep a feedback log of edits and update your rules monthly
Do this once and every future prompt gets faster and sounds more like you. Skip it, and you'll keep editing generic drafts into shape by hand, every single time.