Do LLMs Care How You Ask? Prompt Tones and AI Accuracy, Trust, and Engagement

Abstract

This paper examines how prompt tone impacts accuracy, trust, and engagement with artificial intelligence. Drawing from sociomaterial theory, it argues that prompt tone, ranging from polite and social-oriented to brusque and commanding, actively shapes both AI performance and user perceptions. Specifically, polite tones are theorized to enhance AI accuracy in complex tasks and increase user trust, whereas task-oriented tones yield better results in urgent contexts. Social-oriented tones, despite fostering trust, may reduce critical engagement, potentially compromising decision-making quality. Conversely, commanding tones might reverse algorithm aversion by promoting heightened scrutiny. The conceptual model integrates these dynamics, suggesting that tone is a critical sociomaterial element influencing both AI behavior and human interpretation. The paper contributes to AI governance and design, advocating for context-sensitive prompting strategies to enhance system reliability, mitigate ethical risks, and optimize human-AI collaboration in organizational settings.

Ähnliche Arbeiten