AI personas built with OpenAI's GPT beat human bias in UX research

A new position paper by Ozgur Taylan Celik, accepted for the ACM CHI workshop on Responsible AI Personas, presents a striking paradox in UX research: real human participants can deliver less authentic insights than AI personas due to context-induced biases. The study recounts a design thinking workshop with high-net-worth banking clients held at a luxury hotel, where the presence of portfolio managers and formal hospitality dynamics subtly skewed participant feedback. Researchers built AI personas using OpenAI's custom GPT builder to simulate more impartial user perspectives, highlighting how environmental factors—like status cues and service expectations—can distort human responses in ways that thoughtful AI personas can avoid.

The paper proposes that AI personas are an underexplored tool to overcome specific human limitations in user research, particularly when the research context itself introduces biases. The authors call for new frameworks that help UX teams recognize when traditional human-only testing may be less reliable than AI-driven alternatives. While cautioning against over-reliance on AI, the work suggests that in high-pressure or status-sensitive settings, carefully constructed AI personas can provide more honest feedback than real participants influenced by their surroundings. This challenges long-held assumptions about the primacy of human input in user experience studies.