How to stop AI's most telltale writing habit. Mini-guide / May 2026.
> 📥 The full mini-guide is available as a PDF (Danish). Download it at the top of the page.
Who is this for?
- You who write with AI every day and are tired of sounding like a bot.
- You who see the pattern everywhere and want to know exactly what it's called.
- You who want to call it out with colleagues without sounding pedantic.
01 / The problem — the habit that gives AI away
It's called negative parallelism. The structure looks like this:
> "It's not X. It's Y."
>
> "It's not about A, it's about B."
>
> "It's not a question of C, but of D."
Every major language model does it. Claude, ChatGPT, Gemini, Mistral. Constantly. Multiple times per answer. Why? Because it feels balanced and smart. It creates an impression of nuance by rejecting a false opposition. The models learned it from millions of blog posts where the structure has dominated for years.
The numbers confirm it
Barron's counted the pattern across Fortune 500 documents. In two years, occurrences have more than quadrupled. Microsoft, McKinsey, Cisco, Accenture. All of them use it. Every shareholder letter, every keynote, every LinkedIn post.
Why it's a problem
The reader feels it. The brain registers the rhythm, even when the reader doesn't do so consciously. And the brain links that rhythm to AI text. Result: your credibility drops. Even when your content is sharp.
02 / 5 quick fixes
1. Delete the sentence entirely. Write directly what the point IS. Drop the antithesis.
- BEFORE: "It's not technology. It's culture."
- AFTER: "The change requires culture."
2. Replace it with a mechanism. Instead of two abstract concepts, describe what concretely needs to happen.
- BEFORE: "It's not tools. It's mindset."
- AFTER: "It requires three new habits in the leadership team."
3. Use a number. Numbers beat abstraction. Every time.
- BEFORE: "It's not a small problem. It's a big one."
- AFTER: "We have 47 open tickets. It needs to drop below 10."
4. Use an example. Concrete beats abstract.
- BEFORE: "It's not theory. It's practice."
- AFTER: "We tried it on the Maersk project last Tuesday."
5. Read it out loud before you send. If the sentence sounds like a TED-talk quote, you caught the AI. Rewrite.
Prompt your AI out of the pattern
The easiest fix: tell your AI to stop. Add this instruction in your system prompt. Or directly in the prompt every time you ask for text.
Copy-paste prompt
> Write without negative parallelism. NEVER use sentences of the type "it's not X, it's Y" or "it's not about A, it's about B". Write directly what the point is. Use concrete examples and numbers instead of abstract oppositions. If you catch me trying to slide back into the pattern, stop and suggest an alternative phrasing.
It doesn't work 100%. LLMs slide back into the pattern after a few responses. But you catch most of it, and you notice it yourself every time it happens.
Pro tip
Ask your AI to flag every time it uses the pattern in its own answer. Then you see it live. The learning sticks faster when you catch it in the act.
And then: write as yourself again. Use AI to think fast. Write the last sentence yourself.
📥 Download the full mini-guide as a PDF (Danish) at the top of the page.
🎓 Want to train your team to write with AI without sounding like AI? See courses or book a workshop.
📬 Subscribe to the newsletter "AI, Built Human" on Substack — weekly insights on AI in practice.
Stefano Vincenti · AI consultant and external lecturer, IT University of Copenhagen · aitrainer.dk