Prompts to make ChatGPT sound human

Why ChatGPT sounds like ChatGPT

Any prompt to make ChatGPT write like a human is working against one habit. Language models write by predicting likely next words, so ChatGPT's default prose settles into the safe middle: similar sentence lengths, the same transitions, a tidy summary at the end. Readers spot it quickly, and so do AI detectors.

A good prompt pulls the model away from that middle. It can't change how the model works, but it can give it a voice to copy, a reader to write for and a list of habits to drop. Below are prompts that do that, then their limits, and when it's faster to humanize AI text with a tool built for the job.

Copy-paste prompts to make ChatGPT write like a human

Use them one at a time or stack several into one message. Replace anything in square brackets with your own details. Two ideas from OpenAI's own prompting advice run through the list: models respond better when you show them what you want, and it helps to say what to do instead of only what not to do.

1. Give it your voice. “Here are two paragraphs I wrote: [paste them]. Match their sentence length, word choice and tone when you write [the piece]. Don't reuse any phrases from my sample.” Why: a real sample tells the model more than any adjective like “casual” or “friendly”.
2. Name the reader. “Write this for [a specific reader, such as a second-year nursing student with five minutes to spare]. Use the words they'd use and skip what they already know.” Why: a named reader forces concrete choices about vocabulary and detail.
3. Ban the stock words. “Don't use these words: pivotal, multifaceted, intricate, testament, underscore, showcase, foster, vibrant, meticulous, furthermore, moreover. Use the plain, everyday word instead.” Why: models lean on a small set of showy words, and asking for the plain word gives them something to do instead.
4. Vary the rhythm. “Vary sentence length. Mix short sentences of five to eight words with longer ones, and don't put three sentences of similar length in a row.” Why: even rhythm is one of the first things readers and AI detectors notice.
5. Ask for specifics. “Back every general claim with one concrete example: a number, a name, a place or a short scene. If you don't know a real fact, write [CHECK] instead of guessing.” Why: specifics are what generic text lacks, and the [CHECK] tag stops the model inventing details to please you.
6. Use contractions. “Write the way a careful person talks: use contractions like it's, don't and you'll, and plain verbs.” Why: “it is” and “do not” in every sentence read stiff and formal.
7. Cut the summary. “Don't end with a summary or a conclusion paragraph. Stop when the last point is made.” Why: the recap paragraph is the most predictable part of an AI draft, and readers skip it anyway.
8. No lists unless needed. “Write in paragraphs. Use a bulleted list only for real lists of separate items, like steps or ingredients. No bold headings inside the text.” Why: ChatGPT turns everything into bullets, while connected prose reads like a person thinking.
9. Take a side. “Take a clear position and say why. Don't balance every point with ‘on the other hand’ unless the evidence really is split.” Why: hedging everything is a model habit, and people usually have a view.
10. Fix the worst lines. After the draft: “List the five sentences that sound most like generic AI writing and rewrite only those.” Why: a targeted second pass gives the model a clear job instead of a vague “make it more human”.

Save the rules so you don't paste them every time

ChatGPT's custom instructions (Settings, then Personalization) apply to all your chats. Put the rules that never change there: your voice sample, the ban list, contractions, no summary paragraph. Keep the reader, topic and length in the chat itself, since those change with every task.

Add “don't use em dashes” while you're there. OpenAI's CEO said in November 2025 that ChatGPT now follows that instruction when it's in your custom instructions. Check the output anyway, because one slip is easy to miss.

An all-in-one prompt

If you'd rather paste one message, this combines the prompts above:

Write [the piece] for [reader]. Match the voice of my sample below: its sentence length, word choice and tone. Use contractions and plain words, and avoid pivotal, multifaceted, testament, underscore, showcase, furthermore and moreover. Vary sentence length. Give one concrete example for each general claim, and write [CHECK] where you'd need a fact you don't have. Write in paragraphs, take a clear position, and don't end with a summary. My sample: [paste two paragraphs you wrote].

Where prompts stop working

Prompting reduces the patterns. It doesn't remove them. After a few tries you'll run into the same limits.

• The model drifts back. The rules tend to hold in the first paragraphs and then fade. By the end of a long answer, the stock transitions and the even rhythm creep back in.
• Longer outputs get more uniform. A short reply can sound fairly natural. A full article written in one go tends to settle into the same sentence shapes, paragraph after paragraph.
• Bans create substitutes. Ban one showy word and the model often reaches for its nearest cousin, which sounds just as stock.
• It can't know your life. A prompt can ask for specifics, but the model can only invent them or leave gaps. Real examples have to come from you.

Detectors see the same thing. They look at wording and rhythm across the whole text, so a lively opening doesn't help much if the rest is uniform. Our guide on how AI detectors work explains what they measure.

How to make ChatGPT sound more human after it writes

The prompts get you a better first draft. These steps take it the rest of the way.

1. Generate in pieces. Ask for one section at a time instead of the whole article, and repeat the key rules in each message so they don't fade.
2. Edit by hand. Read it out loud. Cut an opening line that only warms up, delete the recap at the end, split one long sentence, merge two short ones, and add one detail only you know.
3. Or use a humanizer. LumenWrite rewrites the rhythm and word choice of a draft and keeps every fact, number and citation where you put it. Our ChatGPT humanizer page shows how it handles ChatGPT drafts. The free plan gives you 10 rewrites a day, up to 1,000 words each.
4. Check the result. Paste the final text into our free AI detector to see which sentences still read as AI, then rework those.

No prompt or tool can honestly guarantee that a text passes every AI detector, and a detector score is a signal, not proof. If the writing is for school, follow your institution's rules on AI use.