Do Chatbots Exhibit Personality Traits? A Comparison of ChatGPT and Gemini

The underlying design of Large Language Models (LLMs), trained on vast amounts of human texts, suggests that chatbots based on them might exhibit personality traits because of the human-like characteristics within the texts. The question raised here is whether chatbots are able to perceive their own personalities and distinguish them from humans’. Using the Big Five personality traits model, this study explores whether there are personality differences between several chatbot models (ChatGPT versions 3.5 and 4o, Gemini, and Gemini Advanced) and in languages (English and Polish). The results suggest subtle differences in personality traits between the chatbots. When questioned about their personalities, chatbots initially gave responses emulating a human, but when more specific and targeted inquiries were made they did disclose their own preferences as a LLM. The more advanced models (ChatGPT-4o and Gemini Advanced) displayed larger differences between their initial answers (as given by a supposed human) and the later answers (as given by the LLM itself) than did more basic models (ChatGPT-3.5 and Gemini). Compared to the average human subject, chatbots generally displayed higher Emotional stability and Intellect/Imagination but lower Agreeableness. Investigating chatbot personality traits is crucial for enhancing human–computer interaction by ensuring that LLM-based chatbots are perceived as relatable and trustworthy. It may also provide an insight into whether they might show self-awareness. Understanding how these traits are manifested by various types of chatbots might inform the development of more sophisticated and adaptive versions, perhaps leading to improved user satisfaction and more effective communication.