What Makes a Robot Feel Affective? Verbal Cue Types and the Role of Agreeableness in Older Adults’ Perception of Robot Personality

This study explores the potential of social robots with distinct personality profiles for older adult companionship, focusing on the legibility of affective verbal cues during initial encounters, an area presently underexplored. In a within-subjects experiment, 30 Mandarin-speaking older adults (Mage = 72.4) interacted with two LLM-driven robots exemplifying high-affective (ENFP) and low-affective (ENTP) personalities. Linguistic distinctions were validated through LIWC analysis of conversational turns. Results showed that 80% of participants accurately identified the high-affective robot’s personality. Exploratory factor analysis of perception ratings identified two affective cue subtypes: Feeling Type I (other-directed emotional resonance), which effectively conveyed affectiveness, and Feeling Type II (self-referential emotional volatility), which impaired it, aligning with thin-slice judgment theory’s differentiation between expressive external cues and internal states. Agreeableness emerged as the only Big Five predictor of preference for the high-affective robot (Cohen’s d = 0.86–1.16). These findings suggest that other-directed emotional expression is a legible and viable target for affective human–robot interaction in single-session settings with older adults, whereas self-referential cues may carry risks. Pending further validation with larger samples, these insights can inform design strategies for LLM-based companion robots within the CASA paradigm, emphasizing affective expressiveness that enhances clarity during initial exchanges.