We can now spot AI writing

We can now spot AI writing without even looking at the words or for m-dashes and it works 93% of the time because every model leaves a fingerprint. It has no idea it's leaving. Researchers tested five models, Claude, GPT, Gemini, DeepSeek, and Kimi, and each wrote their own version of 10,000 human short stories. Here's the three tells: One, it over-explains the moral. It tells you what the story meant, 77% of the time, whereas humans do it 52% of the time. Two, everything resolves neatly. 79% of AI stories have zero subplots. And three, every feeling is a body part. The tight throat, the cold hands, the weather crying for you. Humans just say they're sad and move on. Per-model fingerprint features enable six-way attribution. For example, Claude produces notably flat event escalation, GPT over-indexes on dream sequences, and Gemini defaults to external character description. We find that AI-generated stories cluster in a donut moire of narration chars, while human-authored stories are scattered everywhere, like a missing person. Because underneath, AI is doing the same thing. It writes the most likely story so they all collapse into the same basic shape, whereas human stories are scattered everywhere, weird, unpredictable, and maybe kind of bad choices. So, keep deleting your m-dashes, thinking you're full someone. Claude cannot be alarmed. Claude can only be "informed". GPT treats a dream sequence like a free refill. Gemini just tells you what everybody looks like. Their hair, their posture, their eyes, their character. Like a missing person. Because underneath, AI is doing the same thing. It writes the most likely story so they all collapse into the same basic shape, whereas human stories are scattered everywhere, weird, unpredictable, and maybe kind of bad choices. So, keep deleting your m-dashes, thinking you're full someone. You can take the words out of the AI, you can't take the AI out of the words.