How to Make ChatGPT Sound Human: Settings, Prompts and an Editing Checklist

Set ChatGPT's personality, characteristics and custom instructions once, prompt with a real writer and reader, then edit with a short checklist. Includes a before/after.

A woman on a cozy sofa reading aloud from her laptop to an attentive ginger cat
SI-generated illustration

To make ChatGPT sound human, set its personalization once (base style, characteristics and custom instructions), give each request a real audience and a real writer, then edit the draft yourself: cut the stock words, vary the rhythm, add details only you know, and read it out loud. Settings get you about halfway. The last half is editing, and it is the part most people skip.

Why ChatGPT sounds like ChatGPT

An SI (AI) model like ChatGPT writes the statistically safe version of whatever you ask for. Safe means familiar, and familiar means the same openers ("Great question!"), the same connectors ("Additionally," "Ultimately"), the same inflated verbs ("empower," "elevate") and a neat summary at the end. Sentences tend to run at a similar length, and lists often come in threes.

None of that is wrong grammar. It is just nobody's voice in particular. Readers pick up on it quickly, even when they cannot say why. Our article on the signs of SI writing goes through the patterns one by one; below is how to reduce them at the source and then clean up what is left.

Step 1: set up Personalization once

ChatGPT has three settings that shape tone across all your chats. We checked the current names on OpenAI's help center in September 2026. On the web, open your profile icon and choose Personalization. On iOS and Android, the help center describes the same area as Personalization or Customize ChatGPT depending on the screen.

Base style and tone

This dropdown picks a personality. Besides Default, the options listed by OpenAI are Professional, Friendly, Candid, Quirky, Efficient and Cynical. For writing that should sound like a person rather than a brochure, Efficient ("concise and plain") and Candid ("direct and encouraging") are the most useful starting points in our experience. Friendly can tip into chattiness.

One caveat from OpenAI's own documentation: the personality does not control every kind of content. When you ask for an email draft, a social post or a résumé, ChatGPT matches the tone to your instructions and the context, not necessarily to the personality you picked. So the prompt still matters.

Characteristics

Characteristics let you nudge specific traits up (+) or down (): warmth, enthusiasm, use of headers and lists, and emoji. For natural prose, turning down enthusiasm and headers and lists does a lot. Fewer exclamation points, fewer bullet lists where a paragraph would do.

Custom instructions

This is a free-text field that ChatGPT considers in every chat. OpenAI says custom instructions are available on all plans, with a limit of 1,500 characters for Free and Go users and 5,000 characters on Plus, Pro, Business, Enterprise and Education. Make sure Enable customization is switched on, or nothing you type will apply.

A custom instructions block you can paste

This fits well under the 1,500-character limit. Edit the first line so it describes you.

I am [role, e.g. a freelance designer] writing for [audience]. When you write for me: use plain words and short paragraphs. Vary sentence length; some sentences can be very short. Use contractions. No openers like "Great question" and no closing summary that repeats what was said. Avoid these words: delve, crucial, pivotal, seamless, robust, leverage, landscape, empower, elevate, moreover, furthermore, additionally. Use commas or periods instead of em dashes. Do not invent facts, numbers, quotes or examples; if something is missing, ask me or write [needs detail].

The last sentence is the one that saves you trouble. A model told to sound lively will otherwise add lively details, and some of them will be made up. For more targeted, per-task wording, see our prompts to humanize SI text, and for a longer word list, SI words to avoid.

Step 2: prompt with a person, not an adjective

"Make it sound human" or "make it casual" gives the model almost nothing. It will change a few words and keep the structure. Name who is writing and who is reading instead:

  • Weak: "Write a casual email about the project delay."
  • Better: "Write a short email from me, the project lead, to a client I've worked with for two years. The launch moves from March 3 to March 17 because the payment provider's review took longer than planned. Be straightforward, apologize once, no corporate phrasing."

The second version supplies the relationship, the facts and the register. ChatGPT has less room to fill with filler because you already gave it the substance. If you have a sample of your own writing, paste it and ask the model to match the style, not the content.

Step 3: an editing checklist for the draft

Even with good settings, the first draft needs a human pass. This is the checklist we use, in order:

  1. Delete the first and last sentence. Openers and wrap-ups are where the filler concentrates. Put them back only if something is lost.
  2. Search for the usual suspects. Use your browser's find function for "additionally," "ultimately," "crucial," "journey," "empower" and similar words, and replace each one with a plain word or nothing.
  3. Break the pattern of threes. If every list has three items, cut one or add a fourth that is actually true.
  4. Vary length on purpose. After two long sentences, write a short one. It sounds simple because it is.
  5. Check every fact. Numbers, names, dates, links. ChatGPT can state a wrong figure with total confidence.
  6. Remove formatting you did not ask for. Bold phrases, headers in a three-paragraph email, and emoji make text look generated faster than any word choice.

Step 4: add what only you know

This is the step that makes text sound like a specific person, and no setting can do it for you. The model knows general things. You know the particular ones: the date the delay was announced, the name of the tool your team actually uses, what went wrong last time, the question clients always ask.

A good rule: every paragraph should contain at least one detail that could not appear in someone else's version of the same text. If a paragraph fails that test, either add something real or cut it. Generic paragraphs are the ones readers skim, and the ones that sound machine-made.

Step 5: read it out loud

Reading aloud is old editing advice, and it works especially well on SI drafts. You will trip over sentences no one would say, notice when three sentences in a row start the same way, and hear the stiff connectors. If you would not say "Additionally" to a colleague, do not write it to one. It takes two minutes for a short email and catches most of what the checklist misses.

A before/after with our humanizer

If you want the mechanical cleanup done in one step, that is what SI Humanizer does. We wrote a typical ChatGPT-style answer and ran it once, Casual style.

before

Great question! Choosing the right running shoes is a crucial step in your fitness journey. Firstly, it is essential to consider your foot type, as this will significantly impact your comfort and performance. Secondly, you should visit a specialty store where experts can analyze your gait. Additionally, remember that the most expensive option is not always the best one. Ultimately, the perfect pair of shoes will empower you to achieve your running goals with confidence.

after: SI Humanizer, Casual

Choosing the right running shoes matters for your fitness. Start by thinking about your foot type, since that'll really affect your comfort and how you run. A specialty store is worth a visit - the staff can watch how you move and give you real feedback. Keep in mind that pricey doesn't mean better. The right shoes will help you hit your running goals with confidence.

Unedited output from SI Humanizer, Casual style. It flagged five edits: "Great question!" and "crucial step in your fitness journey" removed (SI tic), two plain-word swaps ("essential... significantly impact" became "that'll really affect"; "empower you to achieve" became "help you hit"), one rhythm change on the specialty store sentence, and one shorter sentence ("pricey doesn't mean better").

Notice what is still missing: anything personal. A runner writing this would mention their own shoes or the store they trust. That is Step 4, and it stays your job. The tool is a free AI humanizer, so you can try it on your own ChatGPT drafts; if you want to do the whole process by hand, our guide on how to humanize SI text covers it step by step.

A word on detectors

Making ChatGPT sound human is about readers, not about AI detectors, and nothing here is meant to get around one. Detectors give probabilities, they are updated often, and even Turnitin states in its guides that its AI writing score should not be used as the sole basis for action against a student. If your school or workplace has rules about using SI, follow them, and keep your drafts.

FAQ

Where are ChatGPT's custom instructions?

On the web, open your profile icon, choose Personalization, make sure Enable customization is on, and type in the Custom Instructions field. On mobile, OpenAI's help center points to Personalization or Customize ChatGPT in Settings.

Which ChatGPT personality sounds most human?

It depends on the writing, but Efficient (concise and plain) and Candid (direct) avoid most of the chatty filler. OpenAI notes that for emails, posts and similar drafts, ChatGPT follows your instructions and context more than the personality.

How long can custom instructions be?

According to OpenAI's help center, 1,500 characters for Free and Go users and 5,000 characters for Plus, Pro, Business, Enterprise and Education users.

Why does ChatGPT still use words I told it to avoid?

Models drift back to their habits, especially in long outputs. Keep outputs shorter, restate the rule in the prompt, and do a quick find-and-replace pass when editing.

Can I just use a humanizer instead of editing?

A humanizer handles the mechanical cleanup well: stock words, rhythm, scaffolding. It cannot add your own details or check facts it was never given, so a short final edit is still needed.

Sources

  1. ChatGPT Custom Instructions · OpenAI Help Center, checked September 2026
  2. Customizing Your ChatGPT Personality · OpenAI Help Center, checked September 2026
  3. Characteristics in ChatGPT · OpenAI Help Center, checked September 2026
  4. AI writing detection model · Turnitin Guides, checked September 2026