AI text sounds stiff because it aims for the safest, most average way to say everything. To fix it, show ChatGPT a few samples of your own writing, tell it what to avoid, and then edit the draft with your own details and words. Reading it aloud is the quickest test.
You ask ChatGPT to write a short note to your neighbor, and it comes back sounding like a press release. "I hope this message finds you well. I wanted to reach out regarding…" Nobody talks like that. If your AI drafts feel stiff, you are not doing it wrong. There are simple fixes, and none of them need technical skill.
Why AI writing sounds like AI
ChatGPT does not know you. Unless you tell it otherwise, it aims for the safest, most average way to say something. Journalists and researchers who have compared AI text with human text, including an analysis by The Economist, found some common habits: longer sentences that all have a similar length, fancier words than people usually use, the "it's not this, it's that" pattern, and lists of exactly three things. The details change as AI models are updated, so do not treat any single trick as proof. But the overall feel is the same: smooth, correct and impersonal.
The good news is that the fixes are the opposite of those habits. Be specific, be plain, and let your own details show.
Show it your own writing
This is the biggest single improvement. Find two or three short things you wrote yourself: an email, a text to a friend, a Facebook post. Paste them into ChatGPT and say:
"Here are examples of how I write. Notice my tone, how long my sentences are and the words I use. Then write a short note to my landlord about a broken heater in the same style. Do not copy my sentences."
Use writing you are happy with, and leave out anything private, such as full names, addresses or account numbers. If you are unsure what is safe to paste, see what not to tell an AI chatbot.
Tell it what to avoid
AI follows instructions well when they are specific. Add a line like this to your request:
"Write the way a friendly person talks. Use short sentences and everyday words. No buzzwords, no lists of three, no flowery openings such as 'I hope this finds you well'. Do not end with a summary."
If the first draft is still too formal, just say so: "Make it warmer and shorter." You can go back and forth as many times as you like. For more on asking well, see how to write prompts.
Save your style in Custom Instructions
If you do this often, you do not have to repeat yourself every time. ChatGPT has a setting for this. On the website or in the app, open Settings, then Personalization, then Custom instructions. Turn it on and you will see two boxes: one for what ChatGPT should know about you, and one for how you want it to answer.
In the second box you could write: "Write in a warm, plain, conversational tone. Short sentences. Avoid corporate words and dramatic phrasing. Do not start by repeating my question."
Custom instructions are available on all plans and apply to your new chats. In the phone app the entry is usually called Customize ChatGPT, in your profile menu. Menu names can change with updates, so if you do not see them, look around the Settings menu. If you would like ChatGPT to remember things about you as well, our guide to ChatGPT memory settings explains how to control that.
Add the details only you know
This is what separates your writing from anyone's. AI cannot know that your neighbor's dog barks at 6 a.m. or that you visited the hardware store on Saturday. So give it those facts in your request, or add them afterward. One real detail per paragraph does more than any style trick. Vague, general statements are the clearest giveaway, so replace them with something real: a name, a place, a number, a memory.
Read it aloud and edit
Treat the draft as a first draft, not the final one. Read it out loud. Wherever you stumble, or think "I would never say that", change it. Cut the opening line, shorten the long sentences, and swap the fancy words for ones you would use with a friend. It is fine to leave in small imperfections, such as a short sentence or a "well" or "anyway". They are part of how people write.
A quick before and after
Before (typical AI): "I hope this message finds you well. I wanted to reach out regarding the heating issue in my apartment, which has been a significant concern for several days."
After (with a sample of your writing and a few details): "Hi Mr. Novak, the heater in my apartment stopped working on Monday and it is getting cold at night. Could someone come by this week? I am home after 5."
Same message, but the second one sounds like a person wrote it.
When you should not hide the AI
Making a draft sound like you is about clearer, warmer communication, not about fooling anyone. If a school, employer or publication asks you not to use AI, follow the rule. And always read what you send: AI can get facts wrong, so check any names, dates and numbers. You cannot reliably tell whether a text was written by AI just by reading it, and neither can other people, as we explain in how to tell if a text was written by AI. Honesty with the people you write to matters more than any trick. For everyday emails, see also how to write a professional email with AI.
Sources
- How to spot AI writing — Dataconomy
- Forget em dashes: Viral report on AI writing has surprising new clues — Fast Company
- ChatGPT Custom Instructions — OpenAI Help Center
