Waiting for GPT: Probing Interaction Patterns in High-Latency RAG Chatbots

As AI systems shift toward compute-intensive reasoning, traditional interface paradigms are being challenged. This work presents an exploratory case study investigating emerging interaction patterns for AI interfaces, using high-latency Retrieval-Augmented Generation (RAG) as a representative probe. We explore a subset of a broader design space focused on three critical dimensions of latency feedback, source transparency, and confidence communication. Through a pilot study, we probe how specific interaction patterns within this design space impact user experience. Findings suggest that in high-latency scenarios, users prioritize informative transparency over traditional metrics of perceived speed, which helps inform initial design considerations tailored to the emerging era of transparent, reasoning-heavy AI interfaces.