Researchers in Germany had 70 people and seven chatbots take personality tests and write self-descriptions, then asked humans and bots to guess each other's authorship; humans spotted AI writing about 83% of the time, the bots only 67%.

Researchers at Johannes Gutenberg University Mainz had seven chatbots, including ChatGPT and Gemini, and 70 people fill out the same personality questionnaires, then read each other's texts and guess who was human.
Humans correctly identified AI writing about 83% of the time; the bots managed only 67%, and were not even good at recognising their own output. When the texts were matched for length, the humans got noticeably worse.
What stays with me is that even when you are told a text was written by AI, you still end up sensing a temperament in it. That kind of trust, placed in a model that still makes things up, feels risky.



Curated from high-quality sources, with concise summaries and key takeaways.