Less or More: Towards Glanceable Explanations for LLM Recommendations Using Ultra-Small Devices

Structured explanations improve efficiency but sacrifice satisfaction: The research found that organizing LLM explanations into defined components reduced the time users needed to read them and lowered cognitive load on smartwatch-sized screens, but users were less satisfied with these condensed explanations because they felt important details were missing.

Always-on structured explanations increase trust in recommendations: When structured explanations were continuously displayed (rather than shown adaptively), users were more likely to accept AI recommendations, suggesting that consistent access to organized information builds confidence in the system's suggestions.

Confidence-based adaptive explanations underperformed expectations: Contrary to what might be intuitive, showing explanations only when the AI system had high confidence in its recommendation was less effective at improving user perceptions than always displaying structured explanations, indicating that users may benefit from consistent transparency regardless of system confidence levels.

Ultra-small devices create a fundamental design tension: The paper highlights a core challenge in responsible AI deployment: the verbosity of LLM explanations conflicts with the physical constraints of smartwatches and similar devices, forcing designers to choose between readability and completeness when explaining AI decisions to users.

User feedback reveals the need for balanced design solutions: Interview data indicated users want both efficiency and detail, suggesting future approaches should explore ways to provide glanceable summaries while maintaining access to fuller explanations rather than forcing a choice between the two.