The Psychology of AI in the Classroom: What Recent Research Reveals

What AI tool use is doing to student stress, loneliness, and mental health

Since late 2022, with the introduction of ChatGPT, educators have been debating whether AI belongs in the classroom. This is certainly an important discussion, and as an educator with a heavy interest in AI in the classroom, it’s an area I’ve been looking at a lot lately. However, the psychologist in me is also interested in a potentially more important question concerning what AI does to students’ minds when they use it.

Stress and Loneliness in the Mix

One of the most striking findings comes from Wang and Xu (2026), who surveyed 624 university students and found that AI tool usage significantly and positively predicted academic stress (Wang & Xu, 2026). Their model also found that loneliness partially explained this relationship, and that students with stronger self-efficacy actually showed a stronger link between AI use and loneliness (Wang & Xu, 2026). This study provides cause for concern, as it suggests that students most confident in their own abilities are not immune to AI’s psychological costs; they may simply experience them differently. In short, AI use correlates with increased stress and loneliness.

A Longer History of the Same Concern

This is not the first time researchers have found AI use tangled up with loneliness. Crawford et al. (2024) found that AI chatbot use increased students’ short-term feelings of social support from AI, but the students who felt most supported by AI tended to have fewer human friends to begin with, and those using AI for support saw worse grades and a higher intention to leave university than students supported by people. The researchers frame this as a legitimate ethical concern: AI may provide the illusion of being a substitute for human connection without students even realizing the change is happening, and this could have serious implications.

Al-Zahrani (2025) also found that AI chatbots in higher education create a trade-off: students value them for practical benefits like personalized assistance and efficient access to information, yet increased chatbot experience and satisfaction correlate with increased concern about losing human connection and emotional support, suggesting that familiarity heightens the awareness of AI’s relational limits. The authors argue chatbots both empower and alienate students depending on which dimension, cognitive or socio-emotional, is being assessed, and they recommend four mitigation strategies: hybrid models where chatbots handle routine tasks while humans manage emotionally sensitive interactions, adding social-presence features to chatbot design, training students on appropriate AI use and limitations, and clear institutional policies ensuring AI supplements rather than replaces human support (which should always be emphasized, AI is a supplement, not a replacement). The implication for higher education is that chatbot integration should be deliberately balanced, promoting interpersonal contact as essential to student well-being even as AI tools expand access and convenience.

Where It Becomes Clinical

The discussion on AI use and mental health goes deeper, and just like social media, can correlate with an increase in anxiety and depression. Chávez Sosa and Huancahuire-Vega (2026) found that generative AI dependence was significantly associated with mental health symptoms, and that each 1-point increase in dependence linked to a 3%-5% rise in anxiety and depression among medical students, consistent with prior work tying AI reliance to escapism and social substitution. Proposed mechanisms include cognitive offloading (outsourcing thinking to AI erodes self-confidence, triggering performance anxiety) and a “pseudosocial” dynamic where AI substitutes for human support, deepening isolation and depressive symptoms. In this study, AI dependence showed no significant link to stress, suggesting an avoidance-based coping pattern that eases workload pressure while reinforcing anxiety, and fifth-year students showed higher anxiety than sixth-year students, suggesting clinical transition and licensing exam stress as a critical intervention window. Given the cross-sectional design, the authors call the relationship likely bidirectional and self-reinforcing, urging longitudinal research and university policies pairing AI literacy with accessible counseling.

Worryingly, Liu et al. (2026) found a reverse relationship: the presence of anxiety, suicidality, and depression was more likely to result in AI use, using AI as a support for their mental health. Why worryingly? Surely, seeking support is a positive, right? It is if this is accurate, but other research (Perlis et al, 2026) has found that daily AI use correlated with an increase in depression and anxiety. This is an area in desperate need of more research to determine whether there is an AI use > decreased mental health > AI use > decreased mental health… cycle that exists.

What This Means for Course Design

Taken together, these six studies point to a consistent theme: AI’s psychological effects are not fixed properties of the technology, but outcomes of how it is integrated into learning environments. Stress, loneliness, anxiety, and mental health all change depending on design choices, cultural context, and whether students are asked to think alongside AI or simply delegate to it. For faculty, this reinforces a point worth repeating: the psychological question is not whether AI belongs in the classroom, but what conditions make its presence there constructive rather than problematic.