Modeling the Double-Edged Effects of AI Companions on Chinese Generation Z: A Dual-Pathway Analysis of Risk and Protection

Abstract

We study Xingye AI (星野·AI皆你所见), a generative emotional companion used by Chinese Gen Z. This research took advantage of a quantitative research approach by employing survey as the data collection method and descriptive statistics as the data analysis method. With survey data from N = 402 active users, we tested a dual-path model: two risk factors (bedtime procrastination and problematic use) and two protection factors (anthropomorphism and perceived empathy). We also used hierarchical regression models. Risks align with higher depression, anxiety, loneliness, and sleep disturbance; protections show the opposite pattern. Adding protections improves model fit across outcomes (ΔR² ≈ 0.06–0.09). Beyond these behavioral results, we turn them into design. The coefficients set parameters for an adaptive pacing mechanism and a lightweight bandit/RL updater that tunes empathy and timing per user. The loop keeps a supportive tone by day and discourages over-engagement after 23:00. This links measurement to implementation and points to well-being-oriented affective computing.

Received: 21 September 2025 | Revised: 12 December 2025 | Accepted: 21 January 2026

Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.

Data Availability Statement

The data that support the findings of this study are openly available in Zenodo at https://doi.org/10.5281/zenodo.18279792.

Author Contribution Statement

Haiyang Li: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Data curation, Writing – original draft, Writing – review & editing, Visualization. Jing Liu: Conceptualization, Writing – review & editing, Supervision.