Proactive Artificial Intelligence: Evaluating Prompt Timing in Autonomous Driving Contexts

Featured Application: This work supports the design of timing-aware proactive Artificial Intelligence in intelligent cockpits, helping systems deliver prompts at moments that improve trust, usefulness, and user experience while limiting cognitive load. The proposed framework can guide in-vehicle assistants in providing anticipatory, real-time, and reflective support for tasks such as navigation, safety, and infotainment, and can be extended to other domains requiring well-timed human–machine interaction. Proactive Artificial Intelligence systems in intelligent cockpits can initiate prompts without explicit user commands, yet when such prompts should be delivered remains underexplored. This study examines how prompt timing affects user experience in autonomous driving contexts. Using a Virtual prototype developed in Unity and deployed on Meta Quest, 28 participants experienced a tourism-oriented autonomous driving scenario in a within-subjects design, encountering proactive prompts at three temporal positions relative to driving events: Before, During, and After. User experience was assessed across four dimensions using validated scales. Repeated-measures ANOVA revealed significant effects of prompt timing on all measures (p < 0.001, η2p = 0.35–0.59). Prompts delivered before and during events were consistently rated higher than those delivered after, particularly for trust, usefulness, and satisfaction. Differences between Before and During conditions were limited to overall experience satisfaction, while During prompts were associated with higher cognitive load. These findings suggest that temporal alignment between system behavior and user cognitive processes plays a key role in shaping interaction quality. A layered timing framework is proposed that assigns anticipatory, real-time, and reflective functions to prompts before, during, and after, respectively. Further studies in real-world contexts are needed to validate these results.