This study examines how users interpret identity-disclosed social bots on social media, how perceived human–machine differences shape psychological response orientations and how these orientations are enacted through visible engagement behaviours.
Drawing on a prior online experiment in a simulated Weibo comment environment, the study selected 30 cases through stratified random sampling and reconstructed them as simulation-based qualitative accounts using GPT-4. These accounts were analysed using a large language model (LLM)-enhanced grounded theory procedure involving open, axial and selective coding. The study is framed as a mechanism-building reconstruction rather than a substitute for direct participant testimony.
Analysis of the reconstructed accounts identified two broad types of perceived human–machine differences: content features and behaviour patterns. These reconstructed perceptions were associated with emotion- and trust-based response orientations, which were organised into four bot-oriented patterns: emotion connection avoidance, strategic-functional interaction, cautiously shallow interaction and exploratory evaluation. A four-stage model was developed comprising encounter, human–machine difference perception, behaviour responses and behaviour manifestations. The model further suggests that the same visible acts, such as liking or replying, may carry different meanings in human–human and human–bot interaction.
The study shifts social bot research from detection to post-recognition interpretation. It distinguishes response orientations from platform-visible engagement and proposes emotion and trust as linked mechanisms. Methodologically, it shows how empirically anchored LLM-simulated interviews can support LLM-enhanced grounded theory when behavioural records exist, but direct follow-up is unavailable.
