When Generative AI Explain Itself: The Dual Pathways to Trust and Verification Intention in Reasoning-Driven Chatbots

Drawing on two complementary perspectives—the social actor pathway (perceived effort and anthropomorphism) and the explainable AI system (XAI) pathway (perceived transparency and explainability)—this study investigates how communicative self-related cues shape user responses to large reasoning models (LRMs) acting as self-explaining agents. Using OpenAI’s GPT-4o Mini API, we developed a reasoning-based chatbot framed as a mental health specialist and conducted a between-subjects online experiment, manipulating the presence of self-awareness and self-reference cues in its reasoning output. Results showed that self-awareness cues consistently enhanced social actor perceptions and trust, which in turn increased verification intentions. In contrast, the effect of self-reference cues depended on users’ need for cognition, even backfiring among those with low cognitive motivation. While the XAI pathway explained users’ responses to the existential format of reasoning, a transparency paradox emerged: while presence of reasoning section promoted verification intent via increased transparency perception, it simultaneously and directly reduced users’ motivation to verify. This study advances the theoretical understanding of layered secondary cues in human-AI communication and offers practical guidance for designing LRMs that foster more mindful, transparency-calibrated, and responsible human–AI interactions.