AI ‘personality’ isn’t fixed: Chatbots answer differently across languages
What Happened
Researchers tested five AI systems, including ChatGPT-4o, Gemini 2.0 Flash and DeepSeek-V3, using the 60-question Big Five Inventory-2 in English and Mandarin. The assessments were repeated 30 times for each language. Changing language produced measurable differences. For example, ChatGPT-4o averaged 4.74 out of 5 for open-mindedness in English versus 4.03 in Mandarin. DeepSeek-V3 showed the opposite pattern, scoring 4.38 in English and 4.81 in Mandarin.
Key Takeaways
The findings do not mean chatbots literally have personalities or that their underlying abilities change with language. They show that the language used can systematically affect how an AI answers personality questionnaires, something researchers should consider when using LLMs to simulate or study human responses across cultures.