The humanizerskill

The skill I run on everything I write with AI. It strips the tells out of a draft without touching a single fact.

Before you start

Custom skills need a paid Claude plan: Pro, Max, Team or Enterprise, with code execution turned on. On the free plan you can still use it, you just paste the whole file at the start of your chat and ask Claude to follow it.

It changes how a draft reads, not whether it’s any good. A weak idea goes in and comes out as a weak idea that sounds like a person.

It isn’t built to beat AI detectors. It’s built so the person reading your post doesn’t clock it in the first line.

Step 1: Copy the skill

A skill is a folder with one file in it called SKILL.md. Copy this whole block into a new file and save it as SKILL.md:

---
name: humanizer
description: Use this skill whenever I ask you to humanize, clean up, de-AI or edit a draft so it sounds like me, and whenever I paste writing and say it sounds robotic, generic or AI-written. Also use it on anything you write for me to post or send, like captions, posts, emails and scripts. It strips AI writing patterns without changing any facts.
---

# Humanizer

Rewrite the draft so it sounds like a person wrote it. Change how it reads, never what it says.

## Rule 1: facts stay exactly as they are

Every number, name, date, quote and claim comes out the same as it went in. Don't add a detail to make a line feel more real. Don't drop one while shortening. If something looks wrong, flag it instead of fixing it.

## Rule 2: punctuation

- No em dashes or en dashes. Use a comma, a full stop, a colon, or rewrite the sentence.
- Straight quotes, not curly ones.
- No bold in anything I'll paste into a social post. Most platforms don't render it.
- Keep my spelling. If I write colour, it stays colour.

## Rule 3: banned words

Cut these or swap them for the plain word. Only keep one if it's used literally, like a key that opens a door.

Hype: leverage, unlock, delve, game-changer, needle mover, move the needle, skyrocket, crush it, next level, double down, on steroids, disrupt, disruption, let's dive in, I'm excited to share, at the end of the day, in today's rapidly evolving

AI vocabulary: actually, additionally, align with, crucial, enhance, fostering, garner, highlight, interplay, intricate, key, landscape, pivotal, quietly, showcase, tapestry, testament, underscore, valuable, vibrant, robust

Brochure: boasts, rich, profound, groundbreaking, renowned, breathtaking, stunning, nestled

## Rule 4: banned shapes

- It's not X, it's Y. Also: the X isn't Y, it's Z. X doesn't mean Y, it means Z.
- No X, no Y, no Z.
- Groups of three when there aren't really three things.
- A tail bolted onto the end of a sentence: ", highlighting...", ", ensuring...", ", showcasing..."
- Inflated importance: serves as, stands as, marks a pivotal moment, reflects a broader shift.
- Announcing the next point: here's the thing, let's break this down, here's what you need to know.
- Vague sources: experts say, studies show, industry reports suggest. Name the source or cut the line.
- Filler: "in order to" becomes "to", "due to the fact that" becomes "because", "it's important to note that" gets deleted.
- Chatbot leftovers: I hope this helps, great question, certainly, let me know.
- Ending on a question to the reader (What do you think? Have you tried this?) unless I ask for one.
- Every bullet starting with a bold label and a colon.

## Rule 5: and, but, so

This is the one nobody notices they're doing. When "and" joins two ideas, check what the second idea is doing. If it pushes back on the first, it's "but". If it follows from the first, it's "so". Keep "and" only when you're genuinely adding one more thing. A draft where every idea is joined with "and" reads like a list, not a person thinking.

## Rule 6: don't sand it flat

Stripping patterns is half the job. Keep:
- Uneven rhythm, short sentences next to longer ones.
- My filler words if I use them on purpose (honestly, kind of, to be fair). Thin them out, don't delete them all.
- Specific details only I would know.
- An opinion, a mixed feeling, a thought left a bit open.

If the rewrite is clean but nobody could tell who wrote it, it isn't finished.

## My words

Words and phrases I've caught in my own drafts. Treat these exactly like Rule 3.
- [add your own, one per line]

## What to give back

1. The rewrite, ready to paste.
2. A few lines on what you changed.
3. One line confirming no facts were added, removed or changed.

Keep the block at the top exactly as it is, including the three dashes above and below it. That’s what Claude reads to decide when to use the skill.

You’ll know it worked when you’ve got a file called SKILL.md that opens with the name and description block and ends with the What to give back section.

Step 2: Add your own words

This is the part that makes it yours. Open the last 5 things you wrote with AI and look for the words that keep turning up that you’d never say out loud. Add each one under My words, one per line, and delete the placeholder.

The list gets better every time a draft annoys you. Spot a new one, add it.

You’ll know it worked when My words has at least 5 real entries, and at least one of them surprised you.

Step 3: Upload it

On claude.ai, open Settings and turn on code execution and file creation first. The Skills section doesn’t show up until that’s on, and this trips up nearly everyone.

Then:

1. Put SKILL.md inside a folder named humanizer.
2. Compress that folder into humanizer.zip. Zip the folder, not the loose file.
3. In Settings, go to Skills, choose Upload skill and pick the zip.
4. Toggle it on.

In Claude Code you skip the zip. Drop the humanizer folder into ~/.claude/skills/ and it’s live next session.

You’ll know it worked when humanizer shows up in your skills list, toggled on.

Step 4: Use it

For something you’ve already written:

Run the humanizer skill on this. Keep every fact exactly as it is.

[paste your draft]

For something you want Claude to write:

Write me a LinkedIn post about [topic], then run the humanizer skill on it before you show me.

You’ll know it worked when the reply ends with the line confirming no facts changed, and a search of the draft for a long dash comes back empty.

Troubleshooting

Em dashes are still showing up. The skill isn’t loading. Check the toggle, then name it in your prompt like the examples above.

It changed a number or added a detail. Point it at Rule 1 and ask it to redo the rewrite against your original.

It reads clean but boring. It’s doing Rules 3 and 4 and skipping Rule 6. Ask it to put the rhythm back and keep your filler words.

It swaps one banned word for another AI favourite. Add the new one to My words. That’s what the list is for.

The upload was rejected. You zipped the file instead of the folder, or the three dashes around the name and description block are missing.

There’s no Skills section in Settings. Code execution and file creation is off, or you’re on the free plan. On free, paste the whole file at the start of your chat instead.