How to make ChatGPT sound more human

Ask ChatGPT to “write a blog post about remote work” and you’ll get something competent, organised and forgettable. That isn’t a flaw you can fix with one magic prompt. It’s how language models work: with little to go on, they produce the most typical version of what you asked for.

The good news is that most of the robotic feel comes from missing information. Give the model better input, set clear constraints and edit the output, and the result can sound much more like a person, specifically like you.

Why ChatGPT sounds generic by default

Large language models generate text by predicting likely next words based on patterns in their training data. When a request is vague, the most likely response is a broad, safe overview in a neutral tone: the average of countless similar articles. That’s why default outputs share the same openers, transitions and tidy structure.

Two things change that: context (what you know that the model doesn’t) and constraints (how you want it written). The rest of this guide shows how to supply both.

1. Give it the context only you have

Before asking for any text, tell the model:

• Who it’s for: “new managers at a 40-person software company”, not “a professional audience”.
• What the reader should do or understand afterwards: “approve the budget request” or “know which plan to pick”.
• What you know: your data, experience, opinions and examples, even as rough bullet points.
• What to leave out: points the reader already knows or topics that aren’t relevant.

Raw notes are fine. “We tried async standups for three months; meetings dropped from 6 to 2 a week; two people hated it” gives the model far more to work with than any amount of style instruction.

2. Show a sample of your writing

Describing a voice is hard. Showing one is easy. Paste a few paragraphs you wrote, such as an email or a post you’re happy with, and ask the model to match its sentence length, vocabulary and level of formality. Tell it to copy the style, not the content.

3. Describe your voice in concrete terms

Instructions like “make it engaging” or “sound human” are too vague to help. Specific, checkable instructions work much better:

• “Write in first person. Use contractions.”
• “Keep most sentences under 20 words, with the occasional longer one.”
• “No exclamation marks. No rhetorical questions.”
• “Explain it the way you would to a smart colleague from another department.”

4. Tell it which phrases to avoid

You can steer the model away from its habits by naming them. Add a line such as: “Avoid these words and phrases: delve, leverage, seamless, robust, in today’s world, it’s important to note, furthermore, moreover, in conclusion.”

It won’t catch everything, but it removes many of the most obvious tells. Our list of common AI writing patterns is a good source for your own avoid list.

5. Ask for specifics, and permission to say “I don’t know”

Generic writing is full of claims that could apply to anything. Ask for concrete examples, numbers and steps. Just as important, tell the model not to invent them: “If you need a fact or statistic you don’t have, write [NEEDS SOURCE] instead of making one up.” That gives you a clear list of things to research and reduces the risk of confident-sounding errors.

6. Work in smaller pieces

Long outputs drift toward filler. You’ll get better results by working in stages:

1. Ask for an outline and edit it until the structure is right.
2. Generate one section at a time, adding notes for each.
3. Ask for two or three alternative versions of weak paragraphs and pick the best parts.

A prompt template you can adapt

I’m writing [type of content] for [specific audience]. After reading, they should [goal].

My key points and notes: [bullet points, data, examples, opinions].

Match the style of this sample I wrote: [paste 2 or 3 paragraphs]. Write in first person with contractions. Vary sentence length. Get to the point in the first sentence. Don’t use a summary conclusion.

Avoid: delve, leverage, seamless, robust, “in today’s world”, “it’s important to note”, furthermore, moreover.

Don’t invent facts, quotes or statistics. If something needs a source, write [NEEDS SOURCE].

7. Save your preferences

If you use ChatGPT often, its custom instructions (or similar personalisation settings in other assistants) let you store standing guidance about your role, audience and style so you don’t have to repeat it in every chat. Keep them short and specific, and review them now and then as your needs change.

8. Always finish with an edit

Even a well-prompted draft needs a human pass. At a minimum:

• Check every fact and fill in each [NEEDS SOURCE] marker with something you’ve verified.
• Cut the parts that sound like anyone could have written them, and add a detail only you know.
• Read it aloud and rewrite anything you’d never say.

Our step-by-step guide to editing AI-generated writing covers this in more depth. For the sentence-level polish, you can paste a section into the humanizer, use “Show changes” to review each edit, and keep only what fits your voice.

A word on honesty

Making AI-assisted writing sound natural is reasonable. Using it to mislead people isn’t. If your school, employer or publisher requires you to disclose AI use, do so. If an assignment is meant to be your own unaided work, don’t submit AI writing, however human it sounds. See our responsible use policy for more.