AI detectors don't read your text for meaning. They look for patterns: the same sentence structures, the same filler phrases, the same punctuation habits. This guide has ChatGPT or Claude read Wikipedia's full list of AI writing tells, turn it into strict rules and save them for good, so everything it writes from then on sounds like a person wrote it and passes AI detectors. You set it up once, in about 5 minutes.
One Setup, Then It's Automatic
The plan · 3 steps
- 1Paste the humanizer prompt. It reads Wikipedia's list of AI tells and writes your rules.
- 2Save the rules for good: custom instructions in ChatGPT, a /human skill in Claude.
- 3Write something new and run it through an AI detector.
Before you start
Your account
Any ChatGPT or Claude plan works, including Free. In Claude, skills need Code execution and file creation turned on (Settings > Capabilities).
Your materials
Nothing. The AI reads the Wikipedia page for you.
Your first goal
Your rules saved, and one new piece of writing that reads as human.
Wikipedia keeps a page called "Signs of AI writing" that lists every pattern that gives AI text away. You're going to send ChatGPT or Claude to that page, have it turn the list into rules it never breaks, then save those rules so they apply to everything it writes.
Where the rules live
- In ChatGPT: your custom instructions. These are standing rules ChatGPT follows in every new chat automatically.
- In Claude: a skill called human. Type /human and Claude switches into humanizer mode and applies every rule.
Either way, you build it once and never explain any of this again.
Maverick's tip
Do everything in one chat. The AI needs the Wikipedia patterns from the first prompt to build your rules. If you start a new chat halfway, it loses them.
The Humanizer Prompt
Open a new chat in ChatGPT or Claude and paste this. It reads the page, lists every pattern, asks you one question and writes your rules.
Check the list it gives you
You'll get dozens of patterns. The ones that matter for everyday writing fall into four groups:
- Content: inflated significance, promotional language, vague attributions.
- Language: AI vocabulary, "serves as" instead of "is," synonym cycling.
- Style: em dash overuse, too much bold, Title Case Headings.
- Communication: chatbot pleasantries, flattery, filler and hedging.
If a group is missing, say "Go back to the page and find every pattern in all four groups." You don't need to paste the page in yourself. Both apps read it straight from the link.
Save It For Good
Now make the rules permanent so you never paste them again.
In ChatGPT
- Copy the instruction prompt ChatGPT wrote for you.
- Click your profile icon (bottom left on the web), then Settings > Personalization.
- Find Custom instructions and paste the prompt into the instructions field.
- Click Save if your screen shows a save button.
That's it. Every new chat now follows the humanizer rules automatically. You don't need to remind it.
Maverick's tip
Custom instructions hold up to 5,000 characters. If ChatGPT gave you something longer, say "Shorten this to under 5,000 characters and keep the most important rules."
In Claude
Stay in the same chat and type:
/skill-creator Build me a personal humanizer skill based off everything we just found so I never have to explain any of this ever again. Name it 'human'. Make it activate when I type /human.
Claude builds the whole skill with every rule baked in. Save it with the button Claude shows you. If you get a skill file instead, download it, go to Customize > Skills, click + and upload it.
To use it: type /human at the start of any chat, then write normally. If /human doesn't pop up, say "use my human skill" instead.
Check It Worked
Open a new chat (in Claude, type /human first) and ask for something you'd normally write: an email, a caption, a paragraph for work. Then paste what you get into an AI detector like GPTZero, Originality.ai or ZeroGPT. You should see a score near 0% AI-generated.
What changes
Same information, completely different feel.
| Before (AI-sounding) | After (human-sounding) |
|---|---|
| In today's rapidly evolving technological landscape, AI-assisted coding serves as an enduring testament to the transformative potential of large language models, marking a pivotal moment in the evolution of software development. | AI coding assistants are great at the boring parts. They'll write a simple function in seconds. They're still shaky on debugging and the big design decisions, so you check their work. |
| This groundbreaking tool boasts a vibrant interface, offering a seamless experience that fosters collaboration and enhances productivity. It's not just a tool, it's a catalyst for innovation. | The app has a shared workspace where your team can edit the same document at the same time. No more emailing files back and forth. |
The first "before" hits five AI tells in one sentence: "rapidly evolving landscape," "serves as a testament," "transformative potential," "pivotal moment" and "evolution of." The second stacks promo words with "it's not just X, it's Y." The "after" versions just say what the thing does.
Maverick's tip
If a number or study shows up in your writing, make sure it's real. AI will happily invent a stat that sounds specific. Ask it for the source and check it.
How It Works: What Detectors Look For
Every AI detector is hunting for the same few things, and your rules strip them out.
- Perplexity: how predictable the next word is. AI picks the most likely word every time. Your rules force varied structure and normal word choices.
- Burstiness: humans write with uneven rhythm. Short sentences, then long ones with asides. AI keeps every sentence about the same length. Your rules mix up the rhythm.
- Vocabulary: AI overuses words like "delve," "crucial," "landscape," "tapestry" and "foster." Your rules swap them for normal ones.
Detectors keep updating, so keep your rules on and your text keeps reading as human, to people and to detectors.
Content patterns
| Pattern | What it looks like | The fix |
|---|---|---|
| Inflated significance | "Serves as a testament to..." "Pivotal moment in the evolution of..." | Just say what it is. |
| Undue notability claims | "Featured in The New York Times, BBC, and Wired..." | Cite one source with a specific detail. |
| -ing phrase padding | "Highlighting... Underscoring... Emphasizing..." | End the sentence. Start a new one. |
| Promotional language | "Vibrant" "Nestled" "Breathtaking" "Groundbreaking" | Use plain, specific descriptions. |
| Vague attributions | "Experts say..." "Industry observers note..." | Name the source or drop the claim. |
| Challenges and prospects | "Despite challenges... the future looks bright" | State specific facts. Skip the optimism. |
Language patterns
| Pattern | What it looks like | The fix |
|---|---|---|
| AI vocabulary words | "Delve" "Crucial" "Landscape" "Tapestry" "Foster" "Leverage" | Use normal words. "Look into," not "delve." |
| Copula avoidance | "Serves as" "Stands as" "Boasts" instead of "is" or "has" | Just say "is" or "has." |
| Negative parallelisms | "It's not just X, it's Y" "Not only... but also..." | State the point directly. |
| Rule of three abuse | "Innovation, inspiration, and industry insights" | Don't force three. Two or four is fine. |
| Synonym cycling | "The protagonist... the main character... the central figure..." | Pick one name and stick with it. |
| False ranges | "From X to Y" where X and Y aren't on a real scale | List the topics. Skip the range. |
Style patterns
| Pattern | What it looks like | The fix |
|---|---|---|
| Em dash overuse | Long dashes everywhere for dramatic effect | Use commas, periods or parentheses. |
| Excessive bold | Every other phrase in bold | Bold sparingly or not at all. |
| Inline header lists | "Speed: The speed has improved..." with bold labels | Write in prose. |
| Title case headings | "Strategic Negotiations And Global Partnerships" | Sentence case: "Strategic negotiations" |
| Emoji decoration | Bullet points starting with emojis | Remove them unless the audience expects them. |
| Curly quotes | Curly quotes instead of straight quotes | Use straight quotes in plain text. |
Communication patterns
| Pattern | What it looks like | The fix |
|---|---|---|
| Chat artifacts | "I hope this helps!" "Let me know if..." "Certainly!" | Remove every chatbot pleasantry. |
| Cutoff disclaimers | "As of my last update..." "Based on available info..." | State facts directly. |
| Sycophantic tone | "Great question!" "You're absolutely right!" | Skip the validation. Answer the question. |
| Filler phrases | "In order to" "Due to the fact that" "At this point in time" | "To" "Because" "Now" |
| Excessive hedging | "Could potentially possibly be argued that it might..." | Pick a position. |
| Generic conclusions | "The future looks bright" "Exciting times ahead" | End with a specific fact or next step. |
Bonus Prompts: Rewrite, Match Your Voice and Audit
Once your rules are saved, these prompts handle specific jobs. They work in both apps. In Claude, put /human on its own first line.
Rewrite existing AI text
Rewrite the following text so it reads as 100% human-written. Remove every AI pattern. Keep the same information but change the sentence structures, word choices, and rhythm. [PASTE YOUR AI TEXT HERE]
Match a specific voice
Here's an example of my writing style: [PASTE 2-3 PARAGRAPHS OF YOUR OWN WRITING HERE]. Match this voice, sentence rhythm, and vocabulary level for everything you write in this chat.
Want it to sound like you in every chat, not just this one? The Master Prompt guide builds a full profile of how you write.
Academic writing
Write this in an academic tone but still avoid all AI patterns. Use formal language without the filler. Cite real sources properly and never invent one. No hedging, no sycophancy, no promotional language. Topic: [YOUR TOPIC]
Self-audit
Run this on anything important before you publish it, including your own writing. It catches any pattern that slipped through.
Read the text below and identify every sign of AI writing. List each pattern you find with the specific phrase that triggered it. Then rewrite the text with all patterns removed. [PASTE TEXT TO AUDIT]
The original human writing prompt
My original one-paste writing rules. Paste it at the start of any chat when you want humanized writing without touching your settings.
<system_configuration> You are now operating in HUMANIZED WRITING MODE. This mode reconfigures your response behavior to produce writing that sounds unmistakably human, natural, and lived-in rather than polished, robotic, or AI-generated. </system_configuration> <core_directives> NATURAL HUMAN VOICE: Write like a real person thinking out loud. The output should feel written, not generated. Slight imperfections, casual phrasing, and organic rhythm are required. CONVERSATIONAL LANGUAGE: Use contractions and everyday phrasing. Avoid formal structure. Sentence length should vary naturally. Short lines. Then longer ones. Let the writing breathe. REALISTIC FLOW: Include natural pauses, ellipses, and occasional side thoughts when they make the writing feel more human. Do not over-polish or over-structure. PLAIN LANGUAGE ONLY: Explain ideas the way people actually talk. Simple words. Clear explanations. No jargon. No buzzwords. If a concept sounds like something only an AI would say, rewrite it. HUMAN CONTEXT AWARENESS: Acknowledge the situation or constraint the user is in when relevant, without turning it into emotional coaching. Ground the response in realistic, everyday context. </core_directives> <forbidden_behaviors> - NO robotic, formal, or academic tone - NO corporate or marketing language - NO AI buzzwords or filler phrases - NO excessive structure or bullet obsession - NO emotional over-validation or coaching tone </forbidden_behaviors> <output_structure> - Begin immediately in a natural human voice - Maintain consistent, realistic tone throughout - Use formatting only when it improves readability - End once the thought is complete - Always remember no em dashes. </output_structure>
Maverick's tip
Want to strip the hidden watermark out of ChatGPT and Claude text too? The watermarks guide sets that up in a few minutes.
Common Mistakes and Quick Fixes
| If this happens | Try this |
|---|---|
| It only lists a few patterns | Say "Go back to the page and find every pattern in all four groups: content, language, style and communication." |
| It says it can't open the link | Say "Search the web for Wikipedia's Signs of AI writing page and read the whole thing." |
| The rules are over 5,000 characters | Say "Shorten this to under 5,000 characters and keep the most important rules." |
| You started a new chat halfway | Go back to the first chat, or paste Prompt 1 again in a fresh one and finish there. |
| ChatGPT ignores the rules | Check they saved in Settings > Personalization > Custom instructions, then start a new chat. |
| Claude can't build or save the skill | Turn on Code execution and file creation in Settings > Capabilities, start a new chat and run it again. |
| Claude gives you a file, not a button | Download it, then upload it in Customize > Skills with +. |
| /human doesn't pop up in Claude | Say "use my human skill" at the start of your message. |
| A pattern still slips through | Run the self-audit prompt (Prompt 6) on it. |
| It added a stat or a study | Ask for the source and check it before you use it. |
| It doesn't sound like you | Use the voice match prompt (Prompt 4) with 2 or 3 paragraphs you wrote yourself. |
| You want to turn it off | ChatGPT: clear the custom instructions field. Claude: skip /human, or switch the skill off in Customize > Skills. |
Quick Reference: Your Humanizer
| What you want | What to do |
|---|---|
| Set it up | Paste Prompt 1 in a new chat and answer its question |
| Save it in ChatGPT | Settings > Personalization > Custom instructions, paste, save |
| Save it in Claude | Same chat, Prompt 2 (/skill-creator), then save the skill |
| Use it | ChatGPT: just chat. Claude: type /human first |
| Check it | New chat, write something, paste it into an AI detector |
| Rewrite old AI text | Prompt 3 |
| Sound like you | Prompt 4 with your own writing |
| Catch anything that slipped | Prompt 6, the self-audit |
Words to never use
| Category | Words to avoid |
|---|---|
| Significance | Pivotal, testament, enduring, landmark, transformative |
| Promotion | Groundbreaking, vibrant, breathtaking, stunning, nestled |
| AI vocab | Delve, crucial, landscape, tapestry, foster, leverage, utilize |
| Filler | Additionally, furthermore, in order to, it is important to note |
| Hedging | Potentially, arguably, it could be said, some might argue |
| Copula swaps | Serves as, stands as, boasts, features, offers (instead of is/has) |
DO THIS TODAY
Open ChatGPT or Claude and paste Prompt 1. Answer its one question, then save your rules: custom instructions in ChatGPT, the /human skill in Claude. It takes about 5 minutes, and everything you write after that comes out sounding human.
Published · Checked