Co-Constructing Meaning with Large Language Models: A Longitudinal Analysis of Human–AI Dialogues in Emotional Support Contexts

This study investigates how Large Language Models (LLMs), specifically Baidu's Ernie Bot, shape personal narratives when users seek emotional support over repeated sessions. Sixteen participants from China engaged in weekly chatbot interactions for four weeks, supplemented by reflective diaries and pre-/post-study interviews. Conversation analysis and quantitative measures (e.g., mood ratings, meaning-making scales) revealed incremental shifts in user language, including increased lexical alignment with AI-generated phrases and more positive emotion words. In-depth interviews highlighted the complex process by which participants alternately embraced or resisted the AI's framing, with many reporting newfound perspectives and a sense of empathic resonance. However, some voiced skepticism regarding the AI's genuine capacity for emotional understanding, underscoring ethical dilemmas related to anthropomorphism and data privacy. Overall, the findings suggest that iterative dialogues with an empathic-seeming LLM can facilitate meaningful narrative reframing, albeit with notable variations in user experience and potential risks of over-reliance.