DeepSeek, ChatGPT, or Gemini? A Multi-Method Investigation of Neural and Behavioral User Experience

Abstract

As artificial intelligence (AI) tools become increasingly integrated into daily workflows, understanding user interaction patterns with these systems is critical for optimizing interface design and user experience. This study investigates the usability and emotional responses across three prominent conversational AI chatbots: DeepSeek, ChatGPT and Google Gemini, combining traditional usability assessment with neurophysiological measurement using the Emotiv Insight Electroencephalogram (EEG) headset. The research aims to compare AI tools based on user-friendliness and emotional responses, contributing to the development of emotionally adaptive AI.The study included 12 participants ranging in age from 18 to 48, with 75% identifying as female. Prior to the interaction with the AI platforms, the participants completed a presurvey gauging their previous experience and frequency using these platforms. Subsequently, participants completed 5 different randomized task scenarios across all three AI platforms. These tasks consisted of factual Q&A, reasoning and math, code debugging, creative writing, planning and decisions. Simultaneously, EEG data captured real-time emotional markers including interest, excitement, engagement, stress, relaxation, and attention. Tasks were followed by corresponding questions measured on a Likert scale. These questions measured confidence, clarity, helpfulness, creativity, and trustworthiness. After completing all 5 tasks interacting with the tool, the participants completed the User Experience Questionnaire (UEQ) to assess perceived interface quality. UEQ ultimately measures six categories: attractiveness, perspicuity, efficiency, dependability, stimulation, and novelty.Preliminary analysis and results suggest that the AI tools differ in terms of their attractiveness, novelty and stimulation. Some participants mentioned they would switch to using a tool that they had tried for the first time during the experimental session.

Keywords: Artificial Intelligence, Emotional Responses, Emotiv Insight EEG

DOI: 10.54941/ahfe1006882