<p>How can voice assistants (VAs) with generative-AI-based storytelling and questioning support reminiscing practice for older adults? In a mixed-method, between-participants study (<i>N</i> = 41), we examined the effects of AI-based VAs on memory self-efficacy, engagement, and user experience. Results showed that the change in self-efficacy on memory quantity was significantly higher in the three AI conditions (AI-generated storytelling, AI-generated questioning, and the presence of both) compared to the control condition (general questioning without AI). Although neither AI-generated storytelling nor questioning alone had a significant main effect on change in memory self-efficacy, the interaction effect between the two AI functions was significant; the increase in self-efficacy on memory quantity was significantly higher in the conditions using AI-generated storytelling or questioning alone than in the control condition, but not when both AI-generated storytelling and questioning were present. An exploratory correlation analysis revealed that participants who demonstrated a greater increase in recalled words from pre- to post-intervention storytelling tended to report lower self-efficacy for memory accuracy, possibly due to the VA. Conversation analysis showed how the VA provided high or low engagement during reminiscence. Thematic analysis of interviews identified guidance for AI-based VAs to support reminiscence, including safety and privacy, non-judgmental, guidance on response, active listening/reciprocity, and reassurance/question prompting. These preliminary exploratory findings emphasize the importance of AI-generated storytelling and AI-generated questioning functionalities as well as the balance between them in improving memory self-efficacy.</p>