The 9 tells that scream AI wrote it

Confession time: I can usually smell AI writing from a mile away, and it drives me a little nuts. So when I came across a post from an AI professional who writes about AI for a living, and admitted he hates AI writing too, I had to stop and read the whole thing twice.

The author keeps a running file of the tics that give machine-written text away. Not the old em dash trick, that ship has sailed. The models got trained off em dashes, so now people get falsely accused of using AI just for writing clean sentences. The real tells are sneakier. And the fixes are so simple that I actually tested a couple while reading.

The 9 dead giveaways of AI writing

Here’s the checklist the original poster shared, plus the quick human fix for each one.

1. “That’s not X. That’s Y.” Like “That’s not compliance. That’s stalling.” Just say the second half. “They’re stalling.”
2. Two short bits stuck together. “Fast. Simple.” “No fluff. Just answers.” Pick one and write like a real person.
3. The fake metaphor combo. “Less a hammer, more a scalpel.” Two pictures, zero advice. Just say what to do instead.
4. Clapping for itself. “And that matters.” “That’s the part everyone misses.” “Which is exactly the point.” Delete them. You lose nothing.
5. The comparison you can’t follow. “It’s the Excel of AI agents.” Only lands if the reader knows both things. Usually they don’t.
6. Warming up before talking. “Here’s the thing.” “Let me be clear.” “The truth is.” Start one sentence later.
7. Always three things. “Faster, cheaper, smarter.” Real life gives you two reasons, or five. Never always three.
8. The soft range. “5 to 10 minutes.” A range means you never actually did it. Say 7 minutes.
9. The ending that repeats the whole post. “In short…” “At the end of the day…” Just stop typing. End on a high note.

Why this matters: once you see these nine patterns, you cannot unsee them. And that’s the point of the whole exercise, because catching them in your own drafts is what makes your writing sound human again.

How the author actually stops AI from doing all nine

Spotting the tells is half the battle. The genuinely clever part is how this AI professional forces the models to write cleaner from the start. Here’s his step-by-step system, and each step comes with a reason.

1. Type two words into your prompt. Open your chat and write: “Use ASD-STE100.” That’s the writing standard for airplane manuals. Short words, short sentences, because a confusing line in a plane manual can get someone killed. Claude already knows it, and so do ChatGPT, Gemini, and Grok. The author says those two words strip out about 80% of the bad writing on their own.
2. Save the standard so you stop retyping it. Instead of pasting the instruction into every chat, bundle the full rule set into a reusable file or custom skill you can trigger on demand. Set it up once, call it whenever you write.
3. Turn it off for creative stuff. The creator tried it on a poem and a LinkedIn post. The poem was unreadable and the post “sounded German.” His rule: use it for guides, steps, explanations, emails, and documents. Switch it off for anything with a heart.
4. Add four lines to your claude.md. Paste in: “follow Zinsser’s four principles of quality writing: 1. Simplicity 2. Brevity 3. Clarity 4. Humanity” The airplane standard nails the first three but forgets humanity. This line puts it back.
5. Make a forbidden.md file. Copy the 9 patterns above into a file and tell Claude to check every draft against it. Then add a new line each time you catch a fresh tic. The author says his file gets sharper every week, and I believe it, because that’s basically a personal style guide that compounds.
6. Start a new chat more often. Long chats make the writing worse. The model repeats itself and drifts back to the pretty, fluffy sentences. When your chat feels half full, open a fresh one and paste your files back in.

My take

What I love about this system is that it’s not one magic prompt. It’s a loop. Two words to set the tone, a forbidden file that learns your weak spots, and the discipline to reset the chat before it goes soft. The person who posted it clearly built this from real frustration, and that’s exactly why it works.

I started my own forbidden.md the same afternoon. Watching Claude flag its own “and that matters” habit was oddly satisfying.

There’s more nuance and a few extra examples in the original write-up. Go read the full LinkedIn post from the author for the complete breakdown, then start your own list of tells today.