Updated August 2026: Iāve refreshed this guide after another year of working with ChatGPT, Claude and Gemini every day.
Itās getting very easy to disappear into a sea of AI writing. Give ChatGPT or Claude a prompt about hiring, pricing or fundraising and itāll give you something perfectly usable.
The problem starts when you stop putting yourself back into the draft. You could swap your name for someone elseās and very little would change. Keep publishing drafts like that and people running completely different businesses start writing in the same polished voice.
Iāve spent more than twenty years writing and editing, and I now use AI almost every day. Iāve stopped caring much about catching one fashionable AI word because those giveaways change. I care far more about whether the finished draft sounds like something I would actually say.
If I wouldnāt have written the sentence myself, it isnāt finished.šš»
New here? Millennial Masters is where founders learn how to build better businesses in the age of AI. Every week I publish a practical feature like this, with interviews and ideas from entrepreneurs building modern companies.
The dead giveaways
The Max Planck Institute analysed over 740,000 hours of academic talks, YouTube videos and podcasts. They found that since ChatGPTās release, words like delve, meticulous, adept, realm, and swift have surged by as much as 51%.
Very few people will stop halfway through and decide your writing was generated by AI. Theyāll leave with a much simpler feeling that something didnāt quite sound like you. By then youāve already lost them.
These are the patterns I notice most, and Iāve caught myself leaving plenty of them in early AI draftsšš»
1ļøā£ The parallel/contrast/mirror structure problem
If your sentences keep falling into āItās not just X, itās Yā, youāre probably leaving the AI draft untouched. Real people donāt naturally write like that all the time.
2ļøā£ The em-dash obsession
Yes, Iāve been guilty of this too. ChatGPT leans heavily on em-dashes (thatās this long line ā) to join ideas together. One or two can work, but a wall of them screams AI.
3ļøā£ The vocabulary drift
Words like prowess, tapestry, delve, realm, and adept arenāt bad words. They become a giveaway when they start appearing in writing from people whoād never used them before ChatGPT arrived.
4ļøā£ Explaining everything
Phrases like āletās unpackā, āno fluffā and āat the end of the dayā have become AI defaults. They fill space without saying very much, which is usually reason enough to delete them.
5ļøā£ Perfect rhythm
AI likes paragraphs that all feel the same length, with neatly balanced sentences. It likes explaining every idea until thereās nothing left for the reader to work out. Real writing isnāt that tidy.
Put yourself back into the draft
Three things have helped me more than anything else:
Treat it as a first draft. ChatGPT is a good place to start. It shouldnāt be where you stop. I go through every draft looking for the patterns we just covered before I change anything else. AI slop usually starts when the first draft gets mistaken for the finished one.
Use voice dictation. Speaking forces me to use words Iād naturally choose in conversation. I still edit every draft, but it starts much closer to my own voice than a carefully written prompt.
Let your opinion survive the edit. You donāt need every sentence to be perfect. The occasional rough edge reminds people thereās a person behind the words.
How I edit an AI draft
I donāt throw an AI draft away and start again. I edit it the same way Iād edit a guest article or one of my own early drafts.
The first pass is for the obvious giveaways and taking myself back out of autopilot. Thatās where I usually find the patterns, the repeated sentence shapes and the phrases Iād never naturally use.
Take a founder writing about pricing. Give ten people the same broad prompt and youāll get ten drafts saying much the same thing. The difference only starts to show when someone puts their own view back into it, along with the words theyād actually use.
The second is where I start rewriting sentences until they sound like something Iād naturally say. By the end, whole paragraphs have usually disappeared and new ones have taken their place.
Thatās also why Iām not interested in prompts that promise āhumanā writing. The edit is where your voice comes back. AI can help with the first draft. It still canāt tell you whether the finished version sounds like you.
The edit is where the draft becomes yours.
Words that give you away
You canāt prompt your way out of this. You can tell ChatGPT ādonāt use delveā or āavoid em-dashes,ā but the fingerprints slip back in. The patterns run much deeper than individual words.
The fix comes later, when you read the draft as yourself instead of the person who wrote the prompt.
Delve ā look into, explore, break down
Realm ā space, field, world, industry
Prowess ā skill, ability, strength
Tapestry ā mix, web, mess, backdrop
Adept ā good at, skilled, sharp
At the end of the day ā in reality, bottom line, the truth is
Letās unpack ā letās break this down, hereās what matters
No fluff ā straightforward, cut to it, hereās the point
Just scratching the surface ā barely started, only the basics
A game-changer ā a big shift, a real move, it changed things
Paving the way ā opening doors, making room, setting things up
Harnessing the power of ā using, applying, putting to work
Unleashing potential ā developing, building, making the most of
Flaws are proof of life
AI will keep getting better, and the obvious giveaways will keep changing.
That makes your own judgement more important, because the final draft still needs to sound like something you would actually say rather than something the model happened to produce.
Iāve found the simplest test is still the best one: if I wouldnāt have written the sentence myself, it isnāt finished.
The short version
Can you make ChatGPT sound like you?
Yes, but it usually happens during editing rather than prompting. AI can produce a strong first draft, but your voice comes from the decisions you make afterwards.
Why does AI writing sound the same?
Because language models tend to produce similar sentence patterns, vocabulary and structure when given similar prompts.
Should you stop using ChatGPT or Claude for writing?
No. Use it as a first draft, then edit until the final version sounds like something you would naturally say.










I love this! I published a similar article in my Medium account, great minds!
āDigital landscape