# Stop asking Claude to "Humanise This" Do this instead!

https://www.youtube.com/watch?v=k8JY2cDXB1Y

[00:00] Most people try to fix AI flops after Claude is done.
[00:02] That's the wrong time.
[00:04] The fix happens before Claude writes a single word.
[00:08] That's what this video is about.
[00:08] Give me 6 minutes.
[00:09] You'll never have to type humanize this into Claude again.
[00:12] Okay, so before we build anything, let me show you what we're actually dealing with [music] here.
[00:17] Open Claude.
[00:17] Ask it, "Write a LinkedIn post about using AI to save time."
[00:21] Here's what comes back.
[00:21] See that?
[00:22] In today's fast-paced world, three M dashes, a conclusion that starts with ultimately, polished, safe, completely fake-sounding.
[00:28] That's what Claude does without [music] a system.
[00:30] Typing humanize this after the fact doesn't fix it because Claude is still guessing what human [music] means to you.
[00:37] What you need is a filter, something that shows a Claude what AI writing looks like before it starts, so it [music] knows exactly what to avoid.
[00:43] Here's how to build it.
[00:45] Step one, research AI writing patterns.
[00:48] Open Chat GPT deep research.
[00:48] It pulls from multiple sources at once and goes deep fast.
[00:52] Ask it to find [music] everything about how AI writes.
[00:54] Common AI words, overused phrases, robotic sentence structures, big professional
[01:01] tone, LinkedIn AI language, weak introductions, empty transitions, generic conclusions.
[01:05] Here's the prompt.
[01:07] Do a deep research on common AI writing patterns.
[01:11] Find every word, phrase, and sentence pattern that keeps showing up in AI content, LinkedIn posts, emails, YouTube scripts, blog articles, all of it.
[01:18] Leave nothing out.
[01:20] Hit send and let it run.
[01:28] You want a serious document, 20 to 50 pages.
[01:30] This is the foundation of the whole system.
[01:31] Step two, build the word document.
[01:33] Organize everything into sections.
[01:35] Common AI words, AI sentence patterns, AI email habits, AI LinkedIn habits, AI YouTube script habits, weak introductions, weak conclusions, human alternatives, natural sentence rhythm, conversational examples.
[01:59] You're building a blacklist I reference Claude checks against before writing.
[02:03] anything.
[02:05] Step three, add outside references.
[02:07] Search online for articles about AI generated writing and what makes it feel fake.
[02:08] Wikipedia works too.
[02:11] Search AI writing or AI generated text.
[02:13] Take useful notes, save source links, keep the document clean.
[02:24] Step four, convert to markdown.
[02:26] Open the Word document, upload it into Google Docs.
[02:38] Click file, download, choose markdown, save it as anti-ai-writing.md.
[02:44] That file is what we upload into Claude.
[02:46] Step five, test Claude before and after.
[02:48] Open Claude, ask, "Write a LinkedIn post about using AI to save time."
[03:08] Keep that output on screen.
[03:10] Now, upload your anti-ai-writing.md file.
[03:12] Ask the same question again and add, "Write the response without using any words, tone, structure, or writing patterns listed in this file."
[03:35] Look at the difference.
[03:37] The first one, polished, safe, suspicious.
[03:39] Sounds impressive until you read it twice.
[03:41] The second one, sharper, less constructed.
[03:43] The tone relaxes.
[03:45] It actually sounds like a person wrote it.
[03:47] Most people stop here and think that's the system.
[03:48] It's not.
[03:51] This is just a test.
[03:53] The real setup is the next step.
[03:53] Step six, turn it into a Claude skill.
[03:55] Open Claude, go to customize, go to skills, turn on skill creator, create a new skill, name it anti-AI writing, upload your anti-AI writing.md file, write the skill instructions.
[04:04] Tell Claude to avoid all listed AI words,
[04:08] avoid robotic sentence reading,
[04:10] avoid fake professional tone,
[04:13] avoid overused internet phrases,
[04:15] keep natural conversational flow,
[04:16] preserve the original meaning,
[04:18] remove filler, check every response against the anti-AI writing file before finishing.
[04:22] Save it, enable it, test it.
[04:24] To call it, type dollar in the chat window and select anti-AI writing.
[04:26] Claude writes with the filter already active.
[04:28] No uploading, no repeating instructions, it just runs.
[04:32] If the output still feels off after setup, go back into the instructions and tighten them.
[04:37] Without the skill, you open Claude, hunt for the file, upload it, then start.
[04:41] Half the time you forget.
[04:43] With the skill active, you just write.
[04:44] Claude already has the filter loaded.
[04:46] Every chat, every session.
[04:48] Now, here's why this works when humanize this never did.
[04:50] Every time you type that, Claude had no reference point.
[04:52] It just rewrote the same content with slightly softer words.
[04:55] Nothing changed because nothing told it what change looked like.
[04:57] This system gives it that, a document full of patterns to avoid, a skill that loads those rules automatically.
[05:04] Claude stops guessing and starts filtering and it works everywhere.
[05:08] LinkedIn, emails, YouTube
[05:10] scripts, blog posts, the cold formal tone disappears.
[05:14] That stiff sign-off nobody actually says in real life, [music] gone.
[05:16] What's left sounds like you wrote it.
[05:18] Use the skill inside Claude chat, Claude core, Claude quote, every session.
[05:22] One setup, better [music] writing every time.
[05:23] Believe once, use it forever.
[05:26] Like and share this with someone still typing [music] humanize this into Claude.
[05:29] Subscribe if you want more of this.
[05:30] That's it from me.
[05:32] Go make Claude useful.