One prompt turns Wikipedia's signs of AI writing guide into a Claude skill. After that, you say humanize this and your draft comes back sounding like you.
The overused words, the perfectly even sentences, the "it's important to note." Your audience spots them in 2 lines, and they stop reading.
Wikipedia has a volunteer team that cleans up AI written articles. After reviewing thousands of them, they published every pattern they keep seeing in a guide called Signs of AI writing. It is the best list of AI tells anywhere, and it gets updated as the patterns change.
This prompt turns that guide into a skill inside Claude. Build it once, and any caption, email, or page that sounds machine-made gets rewritten with the tells stripped out. The meaning stays. The facts stay. The robot voice goes.
I built the skill already, with the full guide baked in plus 2 rules the prompt alone won't give you: it keeps every fact exactly as you wrote it, and it flags anything it can't verify instead of making something up. Download it and upload it to Claude, done.
New to Claude skills? Watch how to install it. Takes under 2 minutes.
Builds the skill in one message. You paste the prompt, then paste the guide text where the brackets are.
Create a skill called Humanizer. When I give you text and ask you to humanize it, rewrite it to remove every sign of AI writing listed in the guide below. Keep the meaning and the facts, and don't add anything new.
Here is the guide:
[paste the full text of Wikipedia's Signs of AI writing guide here]
On ChatGPT or Gemini? Same idea. Paste the guide into your custom instructions and ask it to avoid every pattern on the list when it writes for you.
Want to check a finished piece instead of rewriting it? The Anti-AI writing audit does that.