Humanize Like Me

Your voice card

Humanize AI in your own voice, with a voice card you can see

Most humanizers guess at “human”. This one learns you, shows you what it learned, and lets you change it.

Emails, essays, posts, messages. More is better: about 1,200 words gets a fully trained card. We measure sentence length, rhythm, contractions, punctuation and small-word habits in your browser; an AI model writes the short summary.

0 words
you

your card appears here

Stored in this browser only. Sign in later if you want to sync it.

In short · How can I make AI text sound like me?

Build a voice card in Humanize Like Me: paste 150+ words of your own writing, take the 2-minute calibration, or fill a short form. The card shows sentence length, formality, contractions, rhythm and your never-use words, and every rewrite follows it. Free: 1 card, no sign-up.

Three ways to build your card

  1. Paste your writing. 150 words or more of something you wrote yourself, or upload a .docx, .txt or .md file. Emails and old essays work well.
  2. Take the calibration. Ten rounds of picking which of two sentences sounds more like you. It takes about 2 minutes. See how it works.
  3. Fill the short form. Tell us about you, your native language, your field and your audience when you don't have samples handy.

What's on the card

Nothing is hidden. If the card says you love semicolons and you don't, flip the toggle.

  • Sliders for sentence length, formality, contractions, vocabulary level and rhythm.
  • Toggles for em dashes, semicolons, parentheses and exclamation marks.
  • Signature phrases you actually use.
  • A “never use” list, the words you'd never say out loud.
  • A few of your own sentences, kept as style anchors.

Use your card anywhere

Export it as a plain-English prompt and paste it into ChatGPT or Claude, so the model writes closer to you from the first draft. Or download it as a stamp-style PNG card.

Plus holds up to 3 cards, so an Essay voice and an Email voice can differ. Free holds 1.

The “sounds like you” score

Each rewrite gets a likeness score from 0 to 100. It compares sentence length, rhythm, contractions, word level, punctuation habits and small-word habits (and, but, so) against your samples, and checks the never-words. It's rule-based, and the reasons are written out in plain English. Read the full method on the About page.

Where your card lives

Cards are stored in your browser by default. If you sign in with Google you can sync them to the cloud, and “delete everything” removes them.

Questions people ask

How much writing do I need to build a voice card?

At least 150 words of your own writing. More helps, especially if you write in different registers. If you have no samples, the 2-minute calibration or the short context form builds a card without any.

Can I edit what the voice card learned?

Yes, that's the point. Every setting is visible and editable: sliders, punctuation toggles, signature phrases, never-use words and style anchors. Change them and the next rewrite follows. The calibration builds a card with no writing at all, so you can start from nothing and refine it later.

Does it copy my writing?

It doesn't paste your sentences into the output. It uses patterns like length, rhythm and word level, plus a few of your sentences as style anchors for the model to match.

Can I use my voice card in ChatGPT?

Yes. Export the card as a plain-English prompt and paste it into ChatGPT or Claude. It's a description of how you write, not a file the other tool needs to read.

How many voice cards can I have?

One on the free plan. Plus gives you 3, for example one each for Essay, Email and Social, so a LinkedIn post doesn't sound like your lab report. Pasting your text builds the card in your browser, and nothing goes to an AI provider until a rewrite runs.

Is my writing sample uploaded to a server?

The card is built and stored in your browser by default. Text is sent to an AI model provider only when you run a rewrite. Optional cloud sync is off unless you sign in and turn it on.

Related

Updated 2026-10-29. We don't promise to beat any AI detector — see our integrity note.