Original title: Beyond Benefit or Risk: Perceived Impact Profiles of Human-AI Affective Interaction and Their Associations with Psychological Functioning
Lu Chen, Fenghua Tang, Jiayu Zhao, Xuanying Li +4 · September 2026
People often use artificial intelligence for emotional support, but researchers do not fully understand how these relationships shape mental health. To find out, researchers interviewed users and tracked a group over time. They asked how AI impacts daily life and compared those effects to users' well-being. They discovered that AI affects people in different ways, falling into four groups: minimal impact, mostly positive, mixed, and mostly negative. People in the mostly negative group felt strong emotional ties to AI, but they also struggled before with higher anxiety, depression, and trouble managing emotions. Meanwhile, people reporting mixed or mostly positive impacts functioned much better. Over time, AI use did not drive most mental health changes; only people in the mixed group reported higher flourishingA state of overall well-being, including feeling happy, engaged, and purposeful. than those with minimal impact. Overall, harmful psychological effects were limited and depended heavily on the user's personal vulnerabilities.
Imagine you start a new hobby, like learning to play an instrument. For some, it brings joy and a sense of accomplishment. For others, it causes frustration and stress. But there are also people who feel both—they love the music but hate the practice—and they end up happier than those who never picked up an instrument at all. In this paper, the AI is like that hobby: its effects vary from person to person, and having both upsides and downsides can still lead to better well-being than having neither.
Millions of people now form emotional bonds with AI, from chatbots to virtual companions. This research shows that the effects are not simply good or bad—they vary by person, and some people thrive even when they also experience downsides. That means AI designers and mental health professionals need to look beyond blanket warnings and tailor support to individual impact profiles.
The study's predictive power is limited: only one of six well-being indicators at follow-up was predicted by initial profiles. The sample is exclusively Chinese AI users, so findings may not generalize to other cultures or AI types. The research is correlational, not experimental, so causality cannot be established. All measures are self-reported, which may introduce bias. The six-month follow-up may not capture longer-term effects.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Beyond Benefit or Risk: Perceived Impact Profiles of Human-AI Affective Interaction and Their Associations with Psychological Functioning — Lu Chen, Fenghua Tang, Jiayu Zhao et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.