ChatGPT humanizer prompts: make your drafts sound like you

How to make ChatGPT sound more human

"Sounding human" mostly means sounding like one person writing to one reader. The blader/humanizer project puts it plainly: a language model picks the choice that fits the widest range of readers, while a person chooses for one reader and one subject. Three inputs change that default more than any adjective in a prompt.

1. A voice sample. Two or three paragraphs you wrote yourself, ideally on a similar topic. A sample carries rhythm, sentence length and word choice that a description like "casual but professional" cannot.
2. A named reader and purpose. "A customer who waited two weeks for a refund" gives the model something to aim at. "A general audience" gives it nothing.
3. A banned list. Name the patterns you see in the draft. Wikipedia's Signs of AI writing guide lists the common ones: inflated significance ("a pivotal moment"), "not just X, but Y" contrasts, forced groups of three, promotional language, vague attributions ("experts say") and collaborative chat residue ("I hope this helps").

Then add one rule that protects meaning: keep every name, number, date and claim, and add none. A rewrite that sounds natural but drops a qualifier has made the text worse. My guide to testing rewrite quality sets out a quick check for that.

Set it once with personalization

If you want every chat to start closer to your voice, use the settings OpenAI documents rather than pasting the same instructions each time.

• Personality presets. OpenAI's article on customizing your ChatGPT personality lists Default, Professional, Friendly, Candid, Quirky, Efficient, Nerdy and Cynical. "Efficient" is the closest to plain, concise prose.
• Custom instructions. OpenAI's custom instructions article says you turn on Enable customization under Personalization and type your preferences. Changes apply immediately across all chats.

A custom instruction that works for most people is short: "Write in plain English. Use short sentences. Avoid stock phrases like 'delve', 'tapestry', 'game-changer' and 'in today's fast-paced world'. Do not end with a summary line or a question unless I ask. Never invent facts, quotes or numbers."

Prompts to humanize AI text

Copy these and change the bracketed parts. Each one is written for your own material: a draft you wrote with AI help, an email, a post or a page you are responsible for.

1. Voice match from a sample

Here are three paragraphs I wrote myself:
[paste your writing]

Rewrite the draft below so it sounds like the samples: same sentence
length, same level of formality, same word choices where they fit.
Keep every fact, name, number and date exactly as written. Add nothing.
Draft:
[paste draft]

2. Strip the tell-tale phrasing

Edit the text below. Remove: "not just X but Y" contrasts, groups of
three that the meaning does not need, inflated words (pivotal,
transformative, testament, landscape), one-line closing sentences that
repeat the point, and any "I hope this helps" style wrapper. Keep the
meaning and all facts. Show me the edited text, then list each change.
[paste text]

Asking for a list of changes makes the edit easy to review. You can reject any change that altered meaning.

3. Email in your own register

I need to send this email to [who] about [what]. I usually write short
emails, first names, no greeting fluff. Rewrite the draft in that style
in under [N] words. Keep the dates and the request exactly as they are.
[paste draft]

4. Marketing or product copy without the sales gloss

Rewrite this product description for [reader]. Say what the product
does and who it is for in plain words. Remove claims I cannot back up,
superlatives and words like "seamless", "cutting-edge" and "unlock".
Keep the specs and prices exactly as given.
[paste copy]

5. Non-native speaker smoothing

English is my second language. Fix grammar and make the phrasing
natural, but keep my ideas, my structure and my examples. Do not make
it more formal than it is. Mark anything you were unsure about.
[paste text]

This is one of the most defensible uses of a humanizing prompt: the ideas are yours and the tool fixes the surface.

6. A LinkedIn or social post that does not sound like a press release

Rewrite this post for [platform]. I write in first person, short
paragraphs, no hashtags in the body and at most one emoji. Open with
the specific thing that happened, not a lesson. Cut any line that
starts with "In today's" or ends with a question to the audience.
Keep every number and name exactly as written.
[paste draft]

Social posts are where readers spot generated text fastest, because they see dozens a day that open with the same "Here's what I learned" shape. Starting with the concrete event is the single change that does the most.

7. A report or summary that keeps its hedges

Tighten this summary for [reader] who has five minutes. Use plain
words and short sentences. Do NOT remove qualifiers such as "may",
"in some cases", "early data" or "about". Do not round numbers.
After the rewrite, list every qualifier from the original and show
where it appears in your version.
[paste summary]

Summaries are where tone edits do the most damage. A model asked to "make it punchier" will happily turn "early results suggest" into "results show". Asking it to account for every qualifier makes that drift visible.

Which prompt for which job

What is the best prompt to humanize AI text?

The best single prompt combines the three inputs above with a two-pass structure: rewrite, then critique, then rewrite again. That is the method the open-source blader/humanizer skill uses. Its SKILL.md tells the model to mark the tells it finds, draft a rewrite, check the draft against the original claims, then write the final version, and it says the model must ask for a missing detail rather than invent one. The README reports that in a blind test logged in the project's issues, judges preferred its rewrite over the original AI text 16 times out of 16. That is the maintainer's own result, not an independent study.

A compact version you can paste into ChatGPT:

Step 1: List every phrase in the text below that sounds generic or
machine-written, strongest first.
Step 2: Rewrite the text for [reader], matching this sample of my
writing: [sample]. Keep every fact. Add none.
Step 3: Reread your rewrite. List anything that still sounds generic
and anything you added or dropped. Fix it and show the final text.
[paste text]

Anthropic's prompt engineering overview groups the same ideas (clarity, examples, chaining steps) as core techniques, and they transfer to ChatGPT. My guide to writing better prompts covers them in more depth.

A worked example

Here is the kind of paragraph ChatGPT produces when you ask for "a short update for the team about the new reporting tool" with no other context. I wrote both versions below myself to show the pattern; neither came from a tool.

Example, before:

Effective reporting has never been more important for teams striving to stay ahead. Our new reporting tool isn't just an upgrade; it's a game-changer that will streamline workflows, enhance collaboration and unlock valuable insights. We're excited to embark on this journey together!

Example, after (voice sample plus banned list):

The new reporting tool goes live on Monday. It pulls the weekly numbers automatically, so nobody has to rebuild the spreadsheet on Friday afternoon. Sara is running a 20-minute walkthrough at 10am on Tuesday. If something looks wrong in your first report, tell her before you fix it by hand.

Notice that the "after" version contains details the "before" version did not: Sara, Tuesday, 10am. Those came from the person writing the update, not from the model. That is the honest limit of any prompt. ChatGPT can strip the generic framing, but it cannot supply the specifics that make writing sound like a person who was there. If you do not give it those details, the best it can do is a shorter, plainer version of nothing in particular.

My take: if a rewrite still feels hollow after a good prompt, the problem is usually missing information, not missing style. Add one real detail per paragraph and the "AI feel" mostly disappears.

Common prompt mistakes and how I'd fix them

A 10-minute workflow I'd use for any important draft

1. Collect your sample. Copy two or three paragraphs you wrote without AI on a similar topic. Keep them in a note so you can reuse them.
2. Write one line about the reader. Who they are, what they already know, what you want them to do.
3. Skim the draft for tells. Mark three to five phrases that sound generic. These go in your banned list.
4. Run the two-pass prompt with your sample, reader line and banned list.
5. Read the change list. Reject any change that altered meaning, softened a commitment or added a claim.
6. Add your specifics by hand. Names, dates, the real example. This is the step no prompt can do for you.
7. Read it aloud once. Anything you would not say out loud, rewrite or cut.

Steps 1 and 2 take a minute once you have done them for the first time. The rest scales with length. For a full manual version of steps 3 and 5, see my seven-step manual humanizing method. If you want to know whether a dedicated app does this job better than a prompt, my explainer on whether AI humanizers work sets out what the research shows.

Does ChatGPT have a humanizer?

No. OpenAI's help articles describe personality presets and custom instructions, not a humanizer feature. Third-party "humanizer" GPTs exist in the GPT Store; a January 2025 paper by Masrour and colleagues noted that two of the four most popular Writing GPTs at the time were humanizers calling outside services. A GPT built by a third party sends your text to that party, so read its privacy terms before you paste anything sensitive. My AI tool privacy checklist lists what to look for.

My advice is simple: for anything with client names, health details or unpublished numbers, skip third-party GPTs entirely and use the prompts on this page in ChatGPT itself, where OpenAI's own data controls apply. My guide on whether AI humanizers are safe covers the terms worth reading before you paste anything into any rewriting tool.

Prompt or dedicated humanizer?

A prompt is free, controllable and works on any chatbot. A dedicated humanizer adds a preset workflow and, in some products, extras such as a detector or word-count plans.

I compare dedicated tools field by field in my best AI humanizers ranking. As one example, Walter Writes AI pairs a humanizer with a built-in AI detector and supports 80 or more languages; the Walter Writes AI review lists its plans and limits. No tool, and no prompt, can guarantee a detector result.

My take: for one person writing a few emails and posts a week, a saved prompt and a voice sample are enough, and I would not pay for anything. A dedicated tool starts to earn its place when you rewrite long documents every week, work across several languages or want a detector in the same account. If you are on a budget, compare the free tiers first in my list of free AI humanizers.

What prompts cannot do

They cannot make a detector result predictable. Detection research moves in both directions. Krishna and colleagues showed in 2023 that a paraphrase model cut DetectGPT's accuracy from 70.3% to 4.6% at a 1% false-positive rate, and that a retrieval defense caught 80% to 97% of paraphrased text in their settings. Turnitin added AI bypasser detection on 27 August 2025, aimed at text modified by humanizer tools. My explainer on how AI detectors work covers why scores shift.

They cannot change the rules. Schools, publishers and employers set their own AI policies. If AI use must be disclosed, a rewrite that hides it is still a breach. My guide on whether using an AI humanizer is cheating quotes real university policies. My guide on humanizing versus paraphrasing explains the difference, and the AI detector false positives guide covers what to do if honest writing is flagged.

They cannot check facts. A model rewriting for tone can drop "may", change "some" to "most" or round a number. Read the result against the original every time. My guide to how to humanize AI text by hand sets out a manual editing method, and if you use Claude rather than ChatGPT, my Claude humanizer skill guide covers a free agent skill built on Wikipedia's list of AI writing patterns.

They cannot add expertise. For web content, a prompt can fix tone but not substance. Google's guidance rewards pages that show first-hand knowledge, and no rewrite supplies that. If you publish AI-assisted pages, my guide to humanizing AI content for SEO covers what to add beyond tone.

How we checked this

• Read on 29 October 2026: OpenAI Help Center articles on ChatGPT personalities and custom instructions; Wikipedia's Signs of AI writing guide; the blader/humanizer README, SKILL.md (version 3.1.0) and CHANGELOG on GitHub; Anthropic's prompt engineering overview; Turnitin's 27 August 2025 bypasser announcement; Krishna et al. 2023 (arXiv 2303.13408); Masrour et al. 2025 (arXiv 2501.03437).
• Scope: I read OpenAI's documented ChatGPT settings, a published open-source prompt and named research, and wrote the prompts and the worked example on this page myself. GitHub star counts change daily; the figure above is as displayed on 29 October 2026.

More guides on writing with AI are in the HumanRank guides hub, and model differences are covered in the ChatGPT review and the ChatGPT vs Claude comparison.

Frequently asked questions

What is a ChatGPT humanizer prompt? It is a prompt that asks ChatGPT to rewrite text so it reads like a specific person rather than generic model output. The strongest versions include a sample of your writing, a named reader, a list of phrasing to avoid and a rule to keep every fact.

How do I make ChatGPT sound less like AI? Give it a sample of your own writing, ban stock phrases by name and tell it not to add summary closers or questions. Setting the Efficient personality and a short custom instruction under Settings, Personalization makes the change apply to every chat.

Does ChatGPT have a built-in humanizer? No. OpenAI documents eight personality presets and custom instructions. Humanizer GPTs in the GPT Store are made by third parties and send your text to their services.

What is the blader/humanizer prompt? An open-source, MIT-licensed skill on GitHub that rewrites text against 26 patterns drawn from Wikipedia's Signs of AI writing guide. Version 3.1.0 was current on 29 October 2026. Its README says getting past AI detectors is not a goal.

Will a humanizer prompt make text undetectable? No prompt guarantees that. Studies show detector results shift after rewriting in both directions, and vendors such as Turnitin retrain against rewriting tools. Schools set their own AI rules, and those rules apply whatever a score says.

Can I use these prompts for school work? Only where your course allows AI help, and with the disclosure it asks for. The prompts here are for your own drafts, emails and content.

Is a dedicated humanizer better than a prompt? It depends on volume. A prompt is free and fully controllable for one-off pieces. A dedicated tool helps when you rewrite long documents often, need many languages or want a detector in the same account.

REVIEW NOTEI checked this guide on 29 October 2026 against OpenAI's help articles on personalities and custom instructions, Wikipedia's Signs of AI writing guide, the open-source blader/humanizer prompt and published research on rewriting and detection.