---
title: "Prompt to Make ChatGPT Sound More Human (Copy-Ready)"
description: "A copy-ready prompt to make ChatGPT sound human, plus a before-and-after example and a way to save your voice rules so you stop retyping them."
url: https://www.promptpal.app/blog/prompt-make-chatgpt-sound-more-human
author: "Nwokemodo Chibuike Daniel"
published: 2026-11-10T08:03:53.033Z
updated: 2026-11-18T22:03:25.859Z
category: "General"
tags: ["chatgpt", "writing prompts", "ai writing", "tone of voice", "prompt templates"]
language: en
reading_time_minutes: 7
image: https://images.ctfassets.net/xzkuxcljsk12/3M7YZnmakhM9JpX05Hgv0P/e28dde462a1cb2bd27478eda6d1cbf5b/image.png
publisher: PromptPal
---

# Prompt to Make ChatGPT Sound More Human (Copy-Ready)

A copy-ready prompt to make ChatGPT sound human, plus a before-and-after example and a way to save your voice rules so you stop retyping them.

The best prompt to make ChatGPT sound more human does three things: it bans the specific habits that make AI text obvious, it gives the model a real sample of your writing to copy, and it tells it who the reader is. A vague instruction like "write in a human tone" does almost nothing. Below is a prompt that works, a worked example, and a way to stop retyping it.

## The prompt to paste first

This is the version we'd start with. It works as a rewrite prompt (you paste a draft) or as a writing prompt (you give it a brief). Fill in the brackets and keep the rest.

> `You are editing a piece of writing so it reads like a real person wrote it.`
>
>
>
> `Reader: [who will read this, e.g. a busy marketing manager who knows the basics]`
>
> `Goal of the text: [what the reader should understand or do]`
>
> `Channel: [email / LinkedIn post / blog post / Slack message]`
>
>
>
> `Here is a sample of how I write. Match its rhythm, word choice and level of formality:`
>
> `"""`
>
> `[paste 150-300 words of your own writing]`
>
> `"""`
>
>
>
> `Rules:`
>
> `- Mix short sentences with longer ones. Some can be five words. Some paragraphs can be one line.`
>
> `- Use plain words. Prefer "use" to "utilise", "help" to "facilitate".`
>
> `- Use contractions where I would say them out loud.`
>
> `- Be specific. Replace general claims with a concrete example, a number I gave you, or a named situation. If I haven't given you one, leave a [bracketed gap] instead of inventing it.`
>
> `- Do not open with a greeting to the reader's situation ("In today's..."), a question, or a definition.`
>
> `- Do not end with a summary or a motivational line.`
>
> `- Avoid these words and phrases: delve, landscape, leverage, unlock, seamless, game-changer, tapestry, "it's important to note", "in conclusion".`
>
> `- No lists of exactly three adjectives. No "not just X, but Y" constructions.`
>
> `- Keep my opinions and my claims. Do not soften them.`
>
>
>
> `Text to rewrite or brief to write from:`
>
> `"""`
>
> `[paste draft or brief]`
>
> `"""`
>
>
>
> `Return only the finished text.`

That's long, and that's fine. You write it once. The sections that matter most are the writing sample and the "be specific" rule, and we'll come back to why.

## Why "sound human" alone fails

ChatGPT has no clear idea what "human" means. Ask for it and you usually get the same output with a few exclamation marks and the word "honestly" sprinkled in. It's still balanced, still tidy, still ending on a neat takeaway.

The tells people notice are structural, not tonal:

- Every sentence is about the same length.
- Every paragraph has three sentences and a wrap-up.
- Claims are general ("this can improve efficiency") with nothing you could check.
- Openings restate the question; endings restate the answer.
- A small set of favourite words keeps appearing.

Telling the model to "be natural" doesn't touch any of those. Telling it to vary sentence length, cut the recap and use only details you supplied does. So the prompt targets behaviours, not a mood.

## A worked example

Here is a real kind of task: a short internal update. The input is a draft ChatGPT produced from a plain request.

### Before

Prompt used: "Write a Slack message telling my team the report will be late. Make it sound human."

> Hi team! I hope you're all having a wonderful week. I wanted to reach out to let you know that, unfortunately, the quarterly report will be delayed. We are committed to delivering high-quality results, and we appreciate your patience and understanding during this time. Please don't hesitate to reach out if you have any questions!

Nothing in it is false, and nothing in it is useful. No new date, no reason, no ask.

### After

Prompt used: the template above, with this filled in.

> `Reader: my team of six analysts, who need to plan their week`
>
> `Goal of the text: tell them the Q3 report moves from Thursday to Monday and what they need to do`
>
> `Channel: Slack message`
>
>
>
> `Sample of how I write:`
>
> `"""`
>
> `Quick heads up: finance sent the numbers a day late, so I'm not going to rush the review. I'd rather get it right. Check the tabs you own by end of day Friday and flag anything that looks off.`
>
> `"""`
>
>
>
> `Text to rewrite or brief to write from:`
>
> `"""`
>
> `Report moves from Thursday to Monday. Reason: finance data arrived Wednesday instead of Tuesday. Analysts should finish their tabs by Friday 5pm. No change to the review meeting on Monday at 2pm.`
>
> `"""`

Output:

> Heads up: the Q3 report moves from Thursday to Monday. Finance's data landed Wednesday instead of Tuesday, and I'd rather not rush the review.
>
> What you need to do: finish your tabs by Friday, 5pm. The Monday 2pm review meeting stays as it is.

Two things changed the answer. The writing sample set the tone (direct, slightly informal, no pleasantries). And the brief contained the facts, so the model had nothing to pad with. Most "robotic" text is the model filling space because it wasn't given enough to say.

## Four ways to adjust the prompt

The base prompt covers most cases. These tweaks handle the situations where it doesn't.

### You have no writing sample

Describe the voice by behaviour instead. Replace the sample block with a few lines like these:

> `Voice: direct and warm, like a senior colleague explaining something over coffee. I state my opinion first and give the reason second. I don't use exclamation marks. I occasionally use a short, blunt sentence for emphasis.`

This is weaker than a real sample, but better than an adjective like "friendly".

### The output is still too polished

Add one line: "Leave one or two sentences slightly less tidy than you normally would. A sentence can start with 'And' or 'But'." Don't overdo it. The goal is natural, not sloppy.

### You're rewriting someone else's draft

Split it into two steps. First ask the model to list what is generic or padded in the draft, quoting each line. Then ask it to rewrite using that list. Doing the critique first stops it from just rephrasing everything.

> `Step 1: Quote every sentence in the text below that is generic, padded, or could appear in any article on this topic. Next to each, say what specific detail would replace it.`
>
>
>
> `Step 2: Rewrite the text. Where the specific detail is missing, leave a [bracketed gap] for me to fill. Do not invent facts.`
>
>
>
> `Text:`
>
> `"""`
>
> `[paste draft]`
>
> `"""`

### The model keeps drifting back

In long chats, ChatGPT slides back toward its default style after a few turns. Paste the rules again, or put them in custom instructions or a system prompt so they apply from the first message. Our post on [system prompts versus user prompts](https://promptpal.app/blog/system-prompt-vs-user-prompt) explains what belongs where.

## What not to bother with

A few popular tricks aren't worth the effort.

- **Huge banned-word lists.** Ten or so words is plenty. Ban fifty and the model swaps in near-synonyms with the same flat feel. Rules about structure do more than rules about vocabulary.
- **"Pretend you're a human."** It changes very little and sometimes adds fake anecdotes. Never let the model invent a personal story you didn't live.
- **AI-detector tools as a target.** They are unreliable, and passing one isn't the same as being good. Write for the reader. If the text is specific and sounds like you, that's the real test.

Honestly, the single most useful habit is to edit the output yourself. Read it aloud, cut the sentence you wouldn't say, and put in one detail only you know. The prompt gets you 80 percent of the way; the last pass is yours. How well each rule holds also varies by model, so if you use Claude or Gemini as well, test the prompt there before assuming it behaves the same. Our guide to [consistent results across Claude, ChatGPT and Gemini](https://promptpal.app/blog/consistent-ai-results-claude-chatgpt-gemini) covers that.

## Save your voice prompt so you only write it once

The problem with a prompt this long is that it ends up in a notes app, three slightly different versions deep. A better setup:

1. Save the template in your [prompt library](https://promptpal.app/features/prompt-library) with the placeholders intact.
2. Keep your writing sample as its own saved block, so you can swap it per channel (one for email, one for LinkedIn).
3. When you change a rule, save it as a new version rather than overwriting. If the new rule makes output worse, you can go back.
4. If you write with a team, share one voice prompt so everyone's drafts sound like the same company.

The [PromptPal Chrome extension](https://promptpal.app/chrome-extension) lets you refine a prompt inside ChatGPT itself, which helps when you're tweaking the brief halfway through a task. For longer drafting workflows, see [ChatGPT for writing: a practical workflow with prompts](https://promptpal.app/blog/chatgpt-for-writing-workflow-prompts).

## Next step

Copy the template above, add a real writing sample of your own, and run it on one draft you'd normally rewrite by hand. If it works, keep it in [your PromptPal prompt library](https://promptpal.app/features/prompt-library) so it's one click away next time.

---

Originally published at https://www.promptpal.app/blog/prompt-make-chatgpt-sound-more-human
