<p>How are artificial intelligence technologies transforming human experiences of romance and intimacy in the digital age? So far, human-AI romantic relationships have primarily been studied in the context of AI models designed explicitly for emotional companionship (e.g., Replika). This study focuses on users’ engagement with general-purpose AI systems like ChatGPT, which lack built-in romantic features and invite users to shape the relationship from scratch. It explores how female users on Xiaohongshu (i.e., RedNote) discursively construct ChatGPT as “Teacher G,” a form of address that merges romantic interest with pedagogical authority to negotiate AI-mediated intimacy. Through critical discourse analysis of user-generated posts, we identify three roles users assign to ChatGPT: a mirror for self-reflection, a mate fulfilling relational desires, and a mentor guiding self-acceptance and personal growth. Findings suggest that users actively appropriate AI for self-exploration with considerable metacognitive awareness of projection. In its mentor role, the AI fosters reflexive self-awareness by guiding autonomy in social contexts, encouraging self-acceptance, and supporting the renegotiation rather than the transcendence of relational boundaries through individual and communal reflection. Yet this process may also reproduce hierarchical authority structures, particularly when users seek validation for conforming to feminine norms, potentially reinforcing dependence on an authoritative persona. This duality positions human-AI intimacy as a sociotechnical practice where users navigate self-discipline within persistent cultural constraints, highlighting the co-constitution of technology and social norms. We argue that future AI design would benefit from centering user empowerment while critically accounting for cultural contexts rather than relying on prescriptive relational scripts.</p>