We asked 5 AIs to write 100 LinkedIn posts. Here's how you can tell.

ChatGPT, Claude, Gemini, Qwen and GPT-OSS each got the same 20 everyday requests, the kind people really type. Promotions, layoffs, hiring, lessons learned. Then we checked every post for the habits that give AI writing away.

The most AI-sounding post we got

Hover or tap a highlight to see why it reads as AI.

The habits they share

Ranked by how many of the 100 posts had each one. Any single habit is fine. People use hashtags and say thank you. It's the pile-up that makes a post read like it came from a machine.

Which AI was the worst offender

How we did it

• Each AI got the same 20 requests, like "Write a LinkedIn post announcing that I got promoted to Senior Product Manager. Just give me the post." No instructions to sound human, and none to sound robotic.
• We used each AI's default settings, one try per request, in October 2026.
• We counted 12 habits in every post: the words, emoji and formatting that kept coming up. A post counts once per habit, however many times it repeats it.
• A habit is not proof. People write "thrilled to share" too. What gives a post away is how many of them pile up in one place. Desloppy's check for social posts looks for the same pile-up.
• All 100 posts and the counting script are kept, so the numbers can be checked.