Personal informatics (PI) systems aim to enhance self-understanding and support behavioral change, yet raw numerical data often lacks emotional resonance and fails to sustain user engagement. This work presents a novel PI system that transforms daily physiological data into AI-generated images and poems, offering personalized, emotionally evocative feedback. A seven-day user study indicated that the aesthetic appeal and novelty of the images, complemented by poetic language, stimulated curiosity and supported self-reflection. However, repeated exposure to similar content reduced engagement over time, revealing the need for flexible content design that adapts to users’ contexts and psychological states. Our findings demonstrate both the potential and limitations of AI-driven multimodal feedback and offer insights for developing more adaptive and enduring reflection-support within PI systems.