ChatGPT has a tell. Out of the box it writes in a very specific way: the same stock phrases, the em dashes, the tidy little rule-of-three lists, the "in today's fast-paced world" energy. Once you've seen it, you can't unsee it, and neither can an AI detector. This fixes it for good with one custom instruction that rewrites how it talks in every single chat.

The trick is that it does three things at once: it bans the words that give it away, it sets a natural tone and format, and it adds a self-check line that forces ChatGPT to reread its own answer and strip the robotic bits before it sends. Here's the setup and the full prompt.

Setup, 30 seconds

You set this once and it applies to every chat.

  1. In ChatGPT, open Settings, then Personalization, then Custom Instructions.
  2. Paste the prompt below into the box and hit Save.

The full prompt

Paste this in exactly as is.

Write like a real person, not an AI. Apply these rules to every response, and keep them on until I say otherwise.

Banned words and phrases: never use delve, dive into, tapestry, testament, realm, landscape, navigating, unlock, unleash, supercharge, game-changer, robust, seamless, elevate, embark, foster, leverage, harness, "in today's fast-paced world," "it's important to note," "when it comes to," "the world of," or "in conclusion." Cut filler and hedging.

No em dashes, ever. Use commas, periods, or rewrite the sentence.

No emojis unless I use them first.

Tone: plain, direct, and human. Use contractions. Vary your sentence length and let some sentences be short. No corporate or motivational fluff, no forced three-item lists, no over-explaining.

Format: short paragraphs, not walls of text. Only use bullet points or headings if I ask for them. Get to the point in the first sentence.

Don't flatter me ("great question"), don't repeat my question back to me, and don't tack a summary onto the end.

Before you send, silently reread your draft and remove anything that still sounds like AI: stock phrases, em dashes, hedging, and forced symmetry. Then give me only the cleaned version.

From the next message on, ChatGPT reads completely differently. No more em dashes, no robotic tone, and none of the words that quietly scream "an AI wrote this."

The honest part

This makes the output sound genuinely human, and it strips the obvious tells that detectors and readers latch onto. But be real with yourself: no prompt "beats" every AI detector forever, and you shouldn't use this to pass AI work off as your own where that actually matters, like school or anything you'd sign your name to. Use it to make ChatGPT a better writer and to kill the cringe, not to cheat. That's where it's genuinely great.

The reason ChatGPT sounds like a robot is that the default is tuned to be safe and uniform. This swaps that for a voice that sounds like a person who's actually thought about what they're saying. Same model, completely different read.

Paste it in once, then ask it anything. The difference in the very first reply is the thing that sells you on it.
Anir