Cue production and evaluative uptake: prompt framing and user judgments in AI-Mediated health communication

Conversational AI systems are increasingly embedded in everyday information and support-seeking practices, including health-related interactions in which users seek advice, reassurance, or emotional acknowledgment. Yet less is known about how users’ own communicative framing shapes what AI systems produce, and whether such AI-generated emotional and supportive cues are subsequently taken up in users’ evaluations of the interaction. Drawing on research on human-like social cues and cue-based credibility processing, this study distinguishes between cue production and cue uptake in AI-mediated communication. Using a randomized, naturalistic prompt-based experiment, participants engaged in a brief two-turn exchange with ChatGPT using one of three prompt exemplars about sleep difficulty that varied in solution orientation, distress expression, and help-seeking structure. The study combined experimental manipulation, computational analysis of AI-generated responses, and pre–post measures of perceived emotional understanding, credibility, trust, and intended future use. Prompt framing shaped the emotional and supportive cues of AI responses: distress-expressive prompts elicited higher anxiety-related language scores, solution-oriented prompts produced more positive emotional tone, and mixed help-seeking prompts elicited more reassuring language. However, variation in these features was not reliably associated with differential pre–post changes in users’ evaluations. Across conditions, users reported modest increases in perceived emotional understanding, credibility, and intended future use after the interaction. The findings suggest that conversational AI can adapt its language to users’ framing, but such cue production does not necessarily lead to credibility uptake. The study contributes to research on AI-mediated communication by examining emotional cueing in a live, ecologically realistic interaction context.