Reflecti-Mate adapts AI support to your thinking style, boosting reflection

Making high-stakes personal decisions involves a mix of cognitive, emotional, and intuitive processes. Most decision-support systems focus only on the cognitive side, ignoring how individuals differ in their thinking profiles. Researchers from TU Delft developed Reflecti-Mate, a conversational agent that dynamically adapts its support to the user's System 1 (fast, intuitive) and System 2 (slow, analytical) thinking patterns. In a between-subjects study with 128 participants, Reflecti-Mate fostered broader and more elaborated thinking, leading to personalized reflective trajectories. Users of the adaptive agent produced more integrative reflective language and perceived the agent as providing stronger support for holistic reflection. In contrast, a baseline non-adaptive agent produced homogenized profiles dominated by cognitive language across participants. The work, accepted at UMAP 2026, demonstrates that personalization to thinking styles can significantly improve the quality and perceived support of AI-driven decision-making tools.

Reflecti-Mate works by analyzing the user's language in real-time to detect whether they are leaning toward intuitive or analytical reasoning, then adjusting its prompts and responses accordingly. For example, if the user shows a strong analytical (System 2) pattern, the agent might encourage considering emotional aspects (System 1) to integrate both modes. The study measured reflective behavior through linguistic analysis and user surveys. Results showed that the adaptive agent not only made users reflect more deeply but also helped them connect different types of thoughts—a key benefit for complex decisions like career moves or medical choices. The baseline agent, which used a fixed set of reflective prompts, failed to elicit such diversity. This suggests that one-size-fits-all decision support may actually narrow thinking rather than expand it. The team notes that future work could integrate Reflecti-Mate into practical applications, such as therapy or strategic planning tools.