Make AI writing sound like you with Humanizer

When a draft sounds like AI, most people ask the chatbot to "make it sound more human". The second draft keeps the same habits: the dramatic one-line ending, the "it's not just X, it's Y", the run of three adjectives. Readers notice them even when they couldn't tell you what's wrong.

Humanizer is a free skill for Claude that names them. It checks your text against 26 patterns from Wikipedia's guide to signs of AI writing, the list its editors use to catch machine-written articles, and rewrites what it finds. Give it a few paragraphs you wrote yourself and it copies your rhythm too. In about ten minutes you learn the five strongest tells, install the skill, and set it up to sound like you.

Before you start

• A Claude account. Skills work on every plan, free included (checked September 2026 on Anthropic's help page). On Team or Enterprise, an admin has to allow them first.
• Code execution on. Settings, then Capabilities, then "Code execution and file creation".
• Two or three paragraphs you wrote without AI. An old email, a post, a note to a colleague.

The plan · 10 minutes, 5 prompts

1. Learn the five tells
2. Install the skill
3. Run your first rewrite
4. Teach it your voice
5. Save your voice so it sticks

1. Learn the five tells

Humanizer ranks its 26 patterns strongest first, and treats any one of the top five as reason enough to rewrite a sentence.

The other 21 are smaller habits, like "delve", bold on every other phrase, and "I hope this helps!". You can hunt for the top five before you install anything, in ChatGPT, Gemini or Claude:

Check the text below for these five signs of AI writing:

1. Not X but Y ("it's not just X, it's Y", "this doesn't mean X, it means Y")
2. One-line closers that repeat the point ("That's the real win.")
3. Sayings that sound deep but say nothing specific
4. A staged run-up before the point ("Here's the thing", "Honestly?")
5. Arguing with an objection nobody raised

For each one you find, quote the sentence and name the tell. Don't rewrite anything yet. If a tell isn't there, say so. End with the two sentences that sound most like AI.

TEXT:
[PASTE YOUR TEXT]

Doing it by hand once shows you which habits are yours, and makes Humanizer's rewrite easier to judge.

2. Install the skill

A skill is a set of written instructions Claude loads when a task needs them. Humanizer is shared on GitHub under the MIT licence, free for anyone to use.

In Claude on the web or the desktop app:

1. Open github.com/blader/humanizer, click the green Code button, then Download ZIP.
2. In Claude, go to Customize, then Skills.
3. Click +, then Create skill, then Upload a skill, and choose the ZIP.
4. Check that Humanizer shows in your skills list with its switch on.

In Claude Code, type:

/plugin marketplace add blader/humanizer
/plugin install humanizer@humanizer

That needs Claude Code 2.1.142 or newer. On older versions, run npx skills add blader/humanizer --global --agent claude-code in your terminal instead.

Heads up: Safari unzips downloads on its own. If you get a folder, right-click it, choose Compress and upload that ZIP. And only install skills you trust: this one's SKILL.md is plain English, so read it first.

3. Run your first rewrite

Start a new chat and ask for the skill by name:

Use the humanizer skill on the text below. It's a [TYPE OF TEXT, E.G. "LINKEDIN POST" OR "EMAIL TO A CLIENT"] for [WHO WILL READ IT].

Keep every fact, name and number. If a sentence needs a detail you don't have, ask me for it instead of making one up.

TEXT:
[PASTE YOUR TEXT]

You get a first draft, a short critique of what still sounds artificial, and the final version. Read the critique: it shows what your drafts keep doing. If Humanizer stops to ask for a detail, answer it. Often the missing detail is what made the original sound empty.

4. Teach it your voice

Without a sample, Humanizer writes clean English that could have come from anyone. With one, it matches your sentence length, word choice and quirks, even habits it would normally remove. If you use dashes, it keeps them.

Use the humanizer skill.

Here's a sample of my writing. Match its sentence length, word choice, punctuation and the way I open and close paragraphs:

[PASTE 2 TO 3 PARAGRAPHS YOU WROTE YOURSELF]

Now humanize this text in that voice:

[PASTE THE TEXT TO REWRITE]

After the final version, list three things you copied from my sample, so I can check you read it right.

Tip: Match the sample to the job. A casual email makes a poor model for a report, so keep one sample for each.

5. Save your voice so it sticks

A Claude project keeps instructions for every chat inside it. Create one (Projects, then New project), call it "Writing", and paste this into its instructions:

MY CONTEXT
I write [WHAT YOU WRITE, E.G. "CLIENT EMAILS AND A WEEKLY LINKEDIN POST"] for [WHO READS IT].
Here is a sample of how I write:
[PASTE 2 TO 3 PARAGRAPHS YOU WROTE YOURSELF]

YOUR JOBS
1. When I paste a draft, use the humanizer skill and match the voice in my sample.
2. When I ask you to write something new, write it in that voice from the start.

RULES
- Never add a fact, number, name or quote I didn't give you. Ask instead.
- Keep my opinions. Don't soften them.
- Show me the final version only, unless I ask for the critique.

Before you send anything from that project, run one last check:

Read your last version once more as a skeptical editor. Quote any sentence that still has one of the five strongest tells (not X but Y, a one-line closer, a saying that sounds deep, a staged run-up, arguing with no one). If you find none, say "clean". Don't rewrite anything unless I ask.

Common mistakes

• Using a sample that AI helped write. Humanizer copies its AI habits. Use something you wrote alone, even if it's rough.
• Humanizing text you haven't fact-checked. Wrong facts come out sounding more believable. Check first, then polish.
• Pasting the whole document at once. Long texts get a lighter touch. Go a section at a time.

The honest part

• It won't get you past AI detectors. The author says that isn't a goal, and detectors still flag most of its output. It's built for people reading your text.
• It doesn't check facts. A wrong number stays wrong in better prose.
• Keep it away from contracts and legal text. A rewrite that sounds better but means something slightly different is a real risk where the exact wording matters.
• Watch what you paste. Black out client names, account numbers, health details and anything about other people first.
• It's made by an independent developer, not Anthropic, and it can change or stop. The patterns come from Wikipedia's public list, so you can keep checking for them without the skill.
• Disclosure rules still apply. If your school, publisher or employer asks you to say when you used AI, a rewrite doesn't change that.

Your action plan

1. Run "Find the tells" on the last thing you wrote with AI. No install needed.
2. Install Humanizer on the web or in the desktop app.
3. Run "Match my voice" on one draft, with two or three paragraphs you wrote yourself.
4. Once the result sounds like you, save the sample in a "Writing" project.

Start with step 1 on today's emails.