Affective computing aims to narrow the gap between artificial intelligence (AI) and emotional intelligence (EI) by enabling machines to detect and respond to human emotions during oral or text-based conversations. However, AI systems cannot fully and genuinely emulate empathy, which remains a core component of human interaction that enriches oral communication with non-verbal cues. This study investigated the capacity of emotional AI to detect and emulate emotions and offered a systematic assessment of how humans perceive and respond to AI-driven empathetic responses. The primary objective was to quantify the discrepancy between perceived human and AI empathy and its correlational relationship with trust and satisfaction, while secondary objectives were (a) to compare model-level emotion recognition accuracy; (b) to characterize distributional properties and cumulative behavior of trust; and (c) to profile emotion intensity patterns. The study employed an experimental design consisting of eight sessions, during which humans interacted with an AI system in a controlled environment. The data gathered included the empathy gap, user trust, emotion recognition accuracy, and user satisfaction, which were analyzed and the results presented in five distinct and non-redundant visualization graphs: histogram, empirical cumulative distribution function, error bar, polar bar, and step plot. The outcomes indicate a persistent and right-skewed empathy gap, suggesting challenges in expressing genuine EI, and trust accumulated in a non-linear fashion, and varying accuracy levels affecting user satisfaction during interaction with AI. The study concludes that a novel, multi-visual empirical framework for evaluating affective AI systems should integrate human factors into its design and frame its conclusions within an ethical paradigm aligned with UNESCO values that promote human dignity and oversight.