AI detectors don't read your text for meaning. They look for patterns. The same sentence structures repeated over and over. The same filler phrases. The same punctuation habits. Remove those patterns and the text reads as human.
Wikipedia maintains a page called "Signs of AI writing" that catalogues every pattern AI detectors flag. You're going to tell ChatGPT to read that page, turn it into a strict set of writing rules, and then save those rules to your custom instructions so they apply permanently.
Why custom instructions? Custom instructions are ChatGPT's permanent memory. Anything you put there gets applied to every new conversation automatically. You set this up once and never think about it again. Every piece of text ChatGPT writes from that point forward follows the humanizer rules.
The Setup
Step 1: Get the AI Writing Patterns
Open ChatGPT and start a new chat. Paste this prompt to make ChatGPT go read Wikipedia's full list of AI writing tells:
ChatGPT will go read the Wikipedia page and come back with a full list of patterns: em dashes, filler phrases, robotic tone, AI vocabulary words, promotional language, and more. Check the list to make sure it found everything.
Step 2: Turn Patterns Into Writing Rules
Stay in the same chat. Now tell ChatGPT to convert all those patterns into a set of strict rules and package them as a custom instruction prompt:
ChatGPT will generate a clean, formatted block of custom instructions. This is the prompt you're going to save permanently. Copy it.
Tip: Stay in the same chat for Steps 1 and 2. ChatGPT needs the context from the Wikipedia patterns to build the rules. If you start a new chat, it won't have that information.
Step 3: Save to Custom Instructions
Now you're going to save those rules permanently so they apply to every future conversation.
Where to find it: Profile icon (top right) > Settings > Personalization > Custom instructions. Paste the prompt there and save.
That's it. From this point on, every new conversation with ChatGPT will automatically follow the humanizer rules. You don't need to remind it. You don't need to paste the prompt again. It just works.
How to verify: Start a new chat and ask ChatGPT to write something. Then paste the output into an AI detector like GPTZero, Originality.ai, or ZeroGPT. If the setup worked, you should see a score near 0% AI-generated.