Answers are grounded in this paper and cite their sources.
Generative artificial intelligence (GenAI) is increasingly used for emotional and mental health support (EMS), a process in which users often described GenAI as empathic and friend-like. These descriptions suggest that people may develop imaginations of GenAI chatbots that may reflect their interpersonal needs and desires. The objective of this study is to systematically explore and thematically organizethe imaginations of GenAI among users who repeatedly use GenAI for EMS. In April 2024, we recruited 270 adults from 29 countries who had repeatedly utilized GenAI for EMS. Participants provided qualitative data regarding their imagination of ChatGPT. The analysis followed a structured reflexive thematic analysis approach. We identified two meta- themes: a) Entity (the “what”), which described the kind of beings or things they imagine GenAI as, appearing in 86.3% ( n = 233) of responses, and b) Trait (the "how"), which described users’ descriptions of perceived capacities of these beings, appearing in 36% (n = 96) of responses. The “Entity” descriptions clustered along two dimensions : abstract–concrete and human–nonhuman, which intersect to form a four-quadrant conceptual space containing 11 subthemes ranging from unbounded energy and spiritual beings to humans and robots . The “Traits” meta-theme fell into five themes: physical traits and manifestations (“looking”), cognitive abilities (“knowing”), productivity (“doing”), emotional abilities (“feeling”), and rapport (“relating”). The findings provide a systematic categorization of imaginations of GenAI for EMS, showing that GenAI is not only used as an informational tool but also imagined sometimes as an all-knowing, powerful, and highly interpersonal being. These imaginations can be highly relevant for examining human–AI relationships and underscore the need to examine how such imaginations relate to benefits and risks of AI chatbot use for mental health.
No takes yet. Share an insight, caveat, or question.
Xiang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: