Short answer: AI writes more like you when it works from a description of your habits, not from the instruction “write in my style”. Collect a few pieces you wrote yourself, turn them into a short profile covering sentence length, vocabulary, person, punctuation, connectors and paragraphing, then give the model that profile with every request. Afterwards, read the result against your original, because a style change must not change what the text says.
This guide shows how to choose the samples, what a useful profile contains, a prompt you can use in any chatbot, and two rewrites of the same passage so you can see what style does and does not change.
Why “write in my style” rarely works
Most people have never had to describe how they write. They know a sentence sounds wrong, but they couldn’t tell you why. So the instruction stays vague, and the model falls back on its default voice with a few surface changes. Three things usually go wrong:
- No sample. “Make it sound like me” gives the model nothing to go on, so it guesses at something casual and friendly.
- One sample, used as a template. Paste a single email and the model tends to borrow its topic and its memorable phrases instead of the habits underneath them.
- The wrong kind of sample. Your text messages and your lab reports were written by the same person in very different styles. A profile built on one won’t suit the other.
A written description fixes most of this. It’s specific, you can read and correct it, and it carries no content from the samples into the new text.
How to choose your samples
The profile can only be as good as what it’s built from. Before you paste anything, run each candidate through these checks:
- You wrote it yourself. Leave out anything drafted or heavily edited by AI, or you’ll get a description of the model’s habits with your name on it. Older work is a safe source.
- It matches what you’re about to write. For a report, use reports. For a newsletter, use newsletters. If you write in two very different registers, keep a profile for each.
- It’s continuous prose. Bullet lists, tables and heavily quoted passages say little about your sentences. Remove long quotations so you’re describing your words and not your sources’.
- It’s long enough to show a pattern. A paragraph or two won’t reveal much about your rhythm. Several hundred words is a better starting point.
- It’s safe to share. Samples go to a model provider for analysis. Remove names, client details and anything confidential, and don’t use text that belongs to someone else.
What a useful style profile contains
A good profile describes habits you could count or point to. “Clear and engaging” tells a model nothing. “Mostly short sentences, with a long one every few lines” gives it something to follow.
| Feature | Vague | Useful |
|---|---|---|
| Sentence length and rhythm | “Good flow” | “Mostly 8–15 words, with an occasional sentence over 25” |
| Vocabulary | “Professional” | “Plain everyday words, technical terms only where needed” |
| Person | “Personal” | “First person singular, addresses the reader as ‘you’” |
| Punctuation | “Correct” | “Commas and full stops, rare colons, no semicolons” |
| Connectors | “Varied” | “Mostly ‘but’, ‘so’ and ‘because’, often at the start of a sentence” |
| Paragraphs | “Well structured” | “Two to four sentences, main point first” |
A two-step prompt for any chatbot
You can do this by hand. Keep the two steps separate, because the first produces something you can check before it affects any text.
Step 1: build the profile.
“Below are samples of my own writing. Describe my style in under 150 words: sentence length and rhythm, vocabulary register, grammatical person, punctuation habits, typical connectors and paragraphing. Don’t quote the samples and don’t mention their topics or any facts from them. [samples]”
Read the answer. If it says you use semicolons and you never do, correct it now. Save the final version somewhere you can reuse it.
Step 2: apply it.
“Rewrite the text below so it follows this style description. Keep the meaning, every fact, number and name, and the same level of certainty. Don’t add examples, opinions or details that aren’t in the text. Style description: [profile]. Text: [text]”
The second instruction matters more than it looks. Asked to sound personal, a model may invent a small anecdote, because that’s what personal writing often contains. An experience you never had is a false statement, however well it matches your voice.
One passage, two writers
Here’s the same content under two different profiles. The rewrites are our own illustrations of what to aim for.
Original: “Effective time management is crucial for students, as it facilitates better academic outcomes and reduces stress. Furthermore, utilizing a planner can significantly improve productivity.”
Writer A uses short sentences, plain words, “you”, and no semicolons.
“Managing your time well matters when you’re a student. It helps you get better results and cuts stress. A planner can also make you much more productive.”
Writer B uses long sentences with subordinate clauses, an impersonal voice and the occasional semicolon.
“Because good time management helps students achieve better academic results and reduces their stress, it matters a great deal; keeping a planner can also raise productivity considerably.”
Compare them claim by claim. Both versions say that time management matters to students, that it helps towards better results and reduces stress, and that a planner can raise productivity by a large amount. “Can” survives in both, and “helps” keeps the caution of “facilitates”, so neither is more certain than the original. Writer A split one long sentence into shorter ones and moved to “you”. Writer B turned “as it facilitates” into a “because” clause and joined the two ideas with a semicolon. The words and the rhythm changed, and the claims stayed the same. One small shift is worth catching: “matters” in Writer A is a little weaker than “crucial”, so if that emphasis counts, restore it.
That’s the check to run on your own results. If a rewrite adds a new detail, or turns “can improve” into “improves”, the model has gone beyond style. Our guide on checking that an AI rewrite kept your facts has a full checklist.
Using “Write like me” in humai
Write like me builds this profile inside the humanizer, and it’s available on the Pro plan. The free humanizer works without it.
- Samples: paste one to three samples of your own writing. Each needs 200 to 3,000 characters, with a limit of 6,000 in total, so three samples can average up to 2,000 characters each.
- Profile: humai describes your style in up to 900 characters, covering the features in the table above. It’s asked to leave out quotations and facts from your samples. You can read the result and delete it at any time.
- Tone: the tone you pick still sets the register, and the profile refines it. Academic tone with your profile should stay academic while moving closer to your habits.
- Punctuation: whatever the profile says, the humanizer removes semicolons and long dashes from the final text, so Writer B’s semicolon wouldn’t survive. If they’re part of your style, add them back by hand.
- Privacy: samples and the profile aren’t stored on humai’s server. The profile is kept in your browser and sent with each text, and both go to our AI model provider to be processed. Because it lives in that browser, you’ll need to create it again on another device or after clearing your site data.
The profile describes style and nothing more. It doesn’t guarantee a particular result, and a model won’t match every habit each time. Treat the output as a draft to edit, and change anything that doesn’t sound like you.
What a style match can’t do
A text that sounds like you isn’t necessarily your work. If a chatbot supplied the ideas, the structure and the conclusions, adopting your rhythm doesn’t change where they came from. For coursework or a thesis, follow your university’s rules on AI, and disclose your use where they ask for it. Our guide on how to disclose AI use in an essay or thesis has example statements.
A style profile also says nothing reliable about how an AI detector will score a text. Scores like the one from humai’s AI detector are estimates, not a verdict, and they can flag careful human writing as well.
One practical test works better than any score. Read the result aloud and ask whether you’d say it that way. If a phrase makes you stop, rewrite it yourself. Those small corrections are also a useful guide to what your next profile should say.