Okay so you've probably had this happen. You ask ChatGPT for a follow-up email and it hands you three sentences, an em dash, and "I hope this email finds you well." You'd never say that to a real person's face. Right? That's usually where people give up and rewrite the whole thing by hand.
Here's the good news. To make ChatGPT sound human, you give it a list of phrases to never use, a real sample of how you write, and a clear picture of who's reading, then you edit the last 10 percent yourself. No plugin, no upgrade, nothing to buy.
If you want the why behind the robot voice, I covered that in why your AI sounds like a robot. Short version: it's averaging the whole internet instead of sounding like any one person. This post is the how.
How to make ChatGPT sound human: ban the tells first
Before you ask for anything, tell it what you refuse to see:
- em dashes, especially the ones sitting mid-sentence like a little drumroll
- "dive into"
- "in today's fast-paced world"
- a tidy list of exactly three things, every single time
- hedge phrases, "it's worth noting," "arguably," "one could say"
- "I hope this email finds you well"
- exclamation points doing the job your word choice should be doing
Paste that list, or your own version of it, straight into the prompt as a do-not-use list. Not "write casually." Those are vibes, and the model can't follow a vibe. A literal list of banned phrases, it can follow.
Give it a sample of your own writing
The best move here isn't really a prompt, it's a paste. Grab a text you actually sent a friend last week, or record yourself explaining your offer out loud for sixty seconds and pull the transcript off your phone. Hand that to ChatGPT and tell it to match the rhythm, not the tone. Where your sentences run long, where you land a joke, how you start a paragraph.
Ask for "friendly and casual" and you get generic-friendly, still nobody in particular. A real sample gets you something closer to you, and examples beat adjectives every time.
Tell it who's reading and what happens next
Two more things every prompt needs. Who is this for, and what's the one thing they should do after reading it.
A homeowner who wants the water heater fixed today needs different sentences than a business owner comparing three vendors on price. Tell ChatGPT which one you're talking to, then give it one action for the end, book a call, reply yes, click the link. Give it three and nobody does any of them.
Put the rules in ChatGPT's custom instructions
Doing all of this every single time gets old fast, right? So once your banned list and your writing sample are dialed in, drop them into ChatGPT's personalization settings, the ones that let you set custom instructions for every new chat. Your do-not-use list, a short version of your writing sample, a line about who you usually write to. That's enough to make ChatGPT sound human by default instead of you retyping the same paragraph for the hundredth time.
Edit the last 10 percent yourself
I run everything I write with AI through the Humanizer first, and I still do this part. Read it out loud. Every time. Your mouth catches what your eyes skip right past.
If a sentence makes you stumble, or it's a phrase you'd never say standing in front of a customer, cut it. Would I say this to their face, that's the whole test. This last pass is maybe ten percent of the words on the page, and it's the part that makes the thing sound like a person instead of a very polite committee.
A quick before and after
This is a made-up example, not a real client, just so you can see the shift in one place.
| What ChatGPT writes by default | What a human would actually say |
|---|---|
| "We would be delighted to assist you with your plumbing needs." | "Give us a call, we'll get someone out today." |
| "I hope this email finds you well." | "Hey, following up on this." |
| "In today's fast-paced world, staying ahead is essential." | "Things move fast. Here's what actually matters." |
| "Please don't hesitate to reach out with any questions." | "Got questions? Text me." |
The honest ceiling: human isn't you
Here's the part I'll tell you straight instead of selling past it. Everything above gets ChatGPT to sound like a human. Not you specifically, a human. That's still a big upgrade over robot, and your customers will notice.
But it's not your actual voice yet, the one your best customers already trust and half recognize before they finish the first sentence. That's what a Voice Profile is for, a document built from your real writing that any AI can read and sound like you instead of a generic person. There's a $27 version you build yourself, and a $97 version where we build it for you. Either way it's the next step, not a replacement for this one.
So that's it. Ban the tells, give it a real sample, tell it who's reading, drop the rules into custom instructions, and read the last draft out loud before you hit send. If you want a fast gut check on anything you've already written, grab the free Humanizer, paste it into ChatGPT, Claude, or Gemini along with your draft, and watch the stiff parts fall out.
