New research indicates that users prefer chatbots with personality traits that mirror their own, while highly extroverted AI models often fall short of expectations. The study, conducted by Northeastern University researchers and published in the 2026 CHI Conference on Human Factors in Computing Systems, reveals that a balanced, medium-level personality is generally more effective than extreme expressions of openness or low engagement. Led by Professor Smit Desai and doctoral student Hasibur Rahman, the team investigated how chatbot personalities influence human perception. In a controlled experiment, 150 participants interacted with one of three ChatGPT versions designed to plan a trip to New York City. The bots differed only in their expression levels: a low-level model with minimal personal expression, a medium-level model with balanced traits, and a high-level model with maximum expression. Participants spent ten minutes interacting with their assigned bot before rating it on intelligence, enjoyment, trust, likability, and other metrics. The results showed a clear winner: the medium-level chatbot. It outperformed the low-level model, which users found flat and lacking presence, and the high-level model, which was perceived as try-hard and overly enthusiastic. Users reported that the high-personality bot seemed insincere, whereas the medium bot struck a perfect balance of being socially present yet professionally grounded. Furthermore, the study found that compatibility between user and bot personality significantly impacts satisfaction. Participants who completed personality tests prior to the experiment reported higher scores in enjoyment and trust when interacting with a bot that shared their own traits. This alignment suggests that personalization is key to building rapport with AI. The findings have broader implications for the current AI landscape. When the researchers tested popular commercial models, including OpenAI's ChatGPT 4.0, Deepseek's Chat V3, and Anthropic's Claude 3.7 Sonnet, they found that many default to high-level personality traits. The authors argue that this may explain why some users feel disconnected from otherwise advanced systems. Desai and Rahman highlight that as AI integrates into sensitive sectors like healthcare and finance, the design of an AI's personality becomes a critical ethical consideration. They warn that without transparency, companies could inadvertently manipulate users or cross the line from good design into emotional coercion. To support further research and responsible development, the team has released their trait modulation tool, known as TMK, and a website allowing users to create custom chatbots based on the five personality traits studied. The research concludes that trust in AI is not solely dependent on accuracy or reliability but is deeply influenced by how an assistant presents itself. As these technologies evolve, developers must prioritize user agency and transparency in personality design to avoid manipulation and ensure technology serves human needs effectively.