Humanizer + VOICE.md: Make Claude Write Like You

I just installed something in Claude Code that fixes the one thing everyone hates about AI writing. You know the tone. Em dashes everywhere. "Leverage" and "comprehensive" in every sentence. That stiff, too-polished feel that makes anyone reading it think "a robot wrote this." Gone.

Most people fight Claude on this. They write a draft, then spend 20 minutes editing the AI tells out of it. I used to do the exact same thing. Then I installed one skill called humanizer and the whole problem disappeared. It catches the 29 most common AI patterns and rewrites them out automatically.

Here is what the workflow looks like once it is set up. You ask Claude Code to write something. It drafts in its normal voice. Before the draft hits you, the humanizer skill runs and strips out the AI tells. Then you layer your VOICE.md on top, which is a file that tells Claude exactly how YOU sound. The output reads like you wrote it. Not even exaggerating.. people will not be able to tell the difference.

Install the humanizer skill from the blader/humanizer GitHub repo into Claude Code at user scope so it is available across every project on my machine. Walk me through the exact commands to run, explain what user scope means, and confirm it is installed at ~/.claude/skills/humanizer. After it is installed, tell me how to invoke it on any future draft (the slash command and how to feed it a writing sample for voice calibration).

That alone makes Claude sound 80% more human. But this is where most people stop, and they leave the best part on the table. The next step is what takes it from "sounds human" to "sounds like ME."

Help me create a VOICE.md file for this project that captures exactly how I sound when I write. I want Claude Code to use this every time it writes anything for me, so it sounds like me and not like AI.

Walk me through it section by section. Ask me:
1. What do I write the most (captions, emails, blog posts, guides)?
2. What words or phrases do I use a lot?
3. What words do I never use?
4. How do I punctuate (do I use dashes, semicolons, exclamation points)?
5. Do I contract (it's, you're) or write everything out?
6. What is my tone (casual, professional, hype, calm)?

After my answers, draft the VOICE.md file with rules at the top and a "Gold Standard Examples" section at the bottom where I can paste 5 to 10 of my actual writing samples. Save it to the project root as VOICE.md.

Now you have both pieces. Humanizer kills the AI tells. VOICE.md teaches Claude your style. Run them together and the output is genuinely indistinguishable from something you would type yourself.

Write a [post / email / caption / whatever you need] about [topic]. After you draft it, run the humanizer skill on it using my VOICE.md file as the voice calibration sample. Show me the final humanized version, not the first draft. I want it to sound exactly like me.

A few things to know. The humanizer skill only runs when you tell it to or when another skill calls it. It is not always-on by default. You either invoke /humanizer manually after a draft, or you build it into your CLAUDE.md so Claude knows to run it on every piece of writing.

Examples beat rules every single time. If your VOICE.md has 50 lines of rules and zero examples, Claude will still write generically. Five solid samples of your actual writing teach it more than every rule combined.

VOICE.md is per-project by default. If you want Claude to use the same voice across every project on your machine, put it in your global ~/.claude/CLAUDE.md instead, or reference your VOICE.md from there.

That is the whole thing. Install humanizer. Build a VOICE.md. Run them together. Your AI writing stops sounding like AI and starts sounding like you. Forever.

The full course shows you the exact VOICE.md I use for ryanfrizelle.com plus how I wire humanizer into every guide page, email, and caption on autopilot.