Everyone can spot AI writing now, and it is costing you trust. This guide gives you a copy-paste instruction set that strips the robotic tells out of ChatGPT and Claude, a way to feed it your actual voice, and a final check that catches the words that give it away.
You have felt it. You ask ChatGPT or Claude to write something, it comes back polished, and yet the second you read it you know a human did not write it. Your audience knows too. The em-dashes, the "in conclusion," the "it is important to note," the words no real person says out loud. That gap is the difference between content that builds trust and content that gets scrolled past.
The good news is that AI writing sounds robotic in predictable ways. It reaches for the same formal words, the same hedge phrases, the same filler, the same punctuation. Once you name those patterns, you can tell the model to stop using them. That is what this guide does. You are not going to rewrite everything by hand. You are going to build one instruction set that does the work every time.
This is the exact system: the rules that make AI write like a person, the step that makes it write like you, and the 30-second check that catches anything that slipped through. It works in both ChatGPT and Claude, and you save it once so you never set it up again.
You cannot remove what you cannot see. AI writing gives itself away with a handful of habits, and once you know them you will never unsee them. The big ones: em-dashes everywhere, hedge words like "perhaps" and "it seems," filler like "in conclusion" and "it is important to note," inflated verbs like "leverage" and "utilize" where "use" would do, and thesaurus words like "robust," "seamless," and "delve."
Real people write with contractions, short sentences, and plain words. They start sentences with "and" or "but." They do not hedge every claim. The whole fix is teaching the model to write the way you already talk.
This is the core of the system. Drop this at the top of a chat, or better, save it into your custom instructions or a project so it applies to everything. It combines every "do this" and "never do that" rule into one block.
From now on, write everything the way a sharp, warm human would actually write it. Follow these rules for every response.
DO: Use clear, simple language. Use short, punchy sentences and vary their length. Use active voice. Write to the reader with "you" and "your." Use contractions the way people speak, like "you're" and "it's" and "don't." Start sentences with "and" or "but" when it flows. Use the occasional idiom or conversational aside. Back up claims with a specific example instead of a vague statement.
DON'T: Never use em-dashes. Use commas, periods, or a rewritten sentence instead. Do not use hedge words like "perhaps," "maybe," "it seems," or "I think." Do not use filler like "in conclusion," "it is important to note," "that being said," "basically," "actually," or "very." Do not use inflated verbs like "leverage," "utilize," "facilitate," "implement," "harness," "streamline," or "optimize." Do not use thesaurus words like "robust," "seamless," "delve," "elevate," "unlock," "transformative," "game-changer," or "cutting-edge." No hashtags, no exclamation-point spam, no emoji unless I ask. Do not over-explain simple things or repeat the same point in new words.
When you finish, reread your own draft and rewrite any sentence that sounds like AI wrote it.
You do not have to memorize any of this. Paste it once and the model does the rest.
The rules above remove the robot. They do not add you. A guide can sound human and still sound like a stranger. To close that gap, give the model real samples of how you write and tell it to match your rhythm.
Here are 2 to 3 samples of my own writing. Study them for my sentence length, my rhythm, the words I reach for, and how formal or casual I am. From now on, match this voice, not a generic one. If a sentence does not sound like something I would actually say, rewrite it. [paste your samples]
The rules plus your samples together is what makes it sound like you and not just less robotic. Save both into the same project or custom instructions so they always travel together.
Even with good instructions, a giveaway word slips through now and then. Before anything goes out, run this quick pass. It takes the model 10 seconds and catches what you would otherwise miss.
Scan this draft one more time. Remove every em-dash and rewrite the sentence around it. Flag and replace any of these if they appear: leverage, utilize, delve, robust, seamless, in conclusion, it is important to note, that being said, unlock, elevate, transformative, game-changer. Then read it out loud in your head. If any line sounds like AI, rewrite it in my voice. Show me the clean version.
Do not skip this on anything public. The em-dash and a couple of giveaway words are what most people notice first, and they are the fastest thing to fix.
A system you have to rebuild every time is not a system. Store the instruction set once and it runs on autopilot.
You just built a voice system that turns generic AI output into writing that sounds like a real person, and specifically like you. No more retyping the same corrections, no more content that gets flagged as robotic before anyone reads the point.
This is one piece of the bigger picture. Your voice is infrastructure, and once it lives in a system your AI can run on, everything you publish starts sounding like it came from you. That is exactly what we build inside Her AI Systems™.
AI models lean on a predictable set of habits: long formal words, hedge phrases like maybe and perhaps, filler like in conclusion and it is important to note, and em-dashes everywhere. Those patterns are the tell. Strip them and the writing reads like a person.
Yes. The instruction set is written for both. Paste it into ChatGPT custom instructions or a Claude project, or drop it at the top of any single chat, and it applies the same rules to whatever the model writes next.
Rules remove the robot voice, but they do not add yours. Paste in 2 or 3 samples of your own writing and tell the model to match your rhythm, sentence length, and word choice. The rules plus your samples together is what makes it sound like you.
Ban em-dashes and the giveaway words: leverage, utilize, delve, robust, seamless, in conclusion, it is important to note. Those alone account for most of the AI tell. Adding contractions and shorter sentences does the rest.
Free Download
Keep the instruction set handy for the next time you open a blank chat.
Follow along for the next guide
New free guide every Tuesday. Find them first on Instagram, TikTok, or by subscribing to the email list.
Want more?
One new free guide. Every Tuesday. No spam, no fluff, just the systems and prompts I'm building for clients, simplified for you.
Subscribe