Decision-Making of Everyday Users When Interacting with Potentially Unreliable AI Chatbot Responses

Sammanfattning

AI chatbots such as ChatGPT, Copilot, and Gemini have become part of everyday digital life and are commonly used for information seeking, planning, learning, or routine decisionmaking. These tools can produce outputs which are fluent and confident, even though these responses might be inaccurate, incomplete, or misleading. Although many of the existing studies examine AI trust and decision-making, these were conducted in controlled or organisational environments. Consequently, little is known about how everyday users evaluate and act upon potentially unreliable chatbot outputs in routine digital practices. This study aimed to examine how everyday users make decisions when interacting with potentially unreliable AI chatbot responses. The study adopted an interpretivist philosophy and a qualitative, inductive research approach and used semi-structured interviews with fifteen everyday AI chatbot users from Sri Lankan and Indian backgrounds. The analysis was conducted using reflexive thematic analysis following Braun and Clarke (2006). Four themes were developed. These findings show that decision-making in human-chatbot interaction is shaped by trust cues, contextual judgement, perceived risk, prior experiences, and awareness of chatbot limitations. Users actively use different strategies such as verification through external sources, prompt refinement, comparing outputs with multiple chatbots, practical testing, and limiting chatbot use in high-risk situations, as they consider chatbots as supporting tools requiring active evaluation rather than as authoritative decisionmakers. The findings highlight that decision-making in human-chatbot interaction is an active, continuous evaluation and adaptive process rather than a simple acceptance or rejection of chatbot outputs. The study contributes to understanding of human-AI decision-making, AI reliability, and trust formation in everyday settings and shows the value and necessity of AI literacy development and transparent chatbot design.