From Enhancement to Over-Reliance: A Mixed-Method Study of Generative AI and Sustainable Learning Performance

Synthesis: Gao, Sun, and Khan (2026) developed a dual-pathway model examining both the positive and negative effects of generative AI use on sustainable learning performance, integrating AI literacy, self-regulated learning, cognitive offloading, and individual differences. Using a mixed-method design with three-wave time-lagged survey data from 623 Chinese university students plus educator interviews, they found that AI literacy significantly enhances critical AI evaluation, which — along with self-regulated learning — promotes effective AI use. Effective AI use positively influences sustainable learning performance but also increases AI over-reliance, which negatively affects learning outcomes, with polychronicity moderating key relationships.

Key Findings

• AI literacy significantly enhances critical AI evaluation, which, along with self-regulated learning, promotes effective AI use.
• Effective AI use positively influences sustainable learning performance but also increases AI over-reliance, which negatively affects learning outcomes — the dual, or two-edged, nature of AI use.
• Polychronicity (multitasking tendency) moderates key relationships, shaping both AI dependency and learning effectiveness: students high in polychronicity are at greater risk of over-reliance.
• The qualitative findings from educator interviews support and explain the quantitative results, highlighting the behavioral mechanisms underlying AI-supported learning.
• The study's central contribution is recognizing two parallel processes coexisting in AI use: creating sustainable learning opportunities while also fostering cognitive over-dependency on AI.

Study Design & Method

The study employed a mixed-method design combining three-wave time-lagged survey data from 623 university students in China with qualitative interviews with educators. The quantitative strand used PLS-SEM to test the dual-pathway model linking AI literacy, critical AI evaluation, self-regulated learning, effective AI use, AI over-reliance, and sustainable learning performance, with polychronicity as a moderator. The qualitative strand used thematic analysis of educator interviews to deepen and explain the observed relationships. The theoretical framework integrated the AI literacy framework, Self-Regulated Learning Theory, and Cognitive Offloading Theory.

What this means for practice

• Instructors. Teach critical AI evaluation as a skill in its own right: AI literacy was strongly associated with critical AI evaluation (β = 0.893, f² = 3.916), which in turn predicted effective AI use (β = 0.567), so verification is learned, not assumed.
• Instructors. Pair AI integration with explicit self-regulated learning instruction — goal setting, monitoring, reflection — because self-regulated learning predicted effective AI use (β = 0.399).
• Instructors. Do not read effective use as a safety signal: in the same model, effective AI use predicted sustainable learning performance (β = 0.871) and AI over-reliance (β = 0.612), and over-reliance reduced performance (β = −0.205), so enhancement and dependency travel together.
• Instructors. Add assessment and reflection tasks that require independent reasoning and documented verification of AI output, rather than only AI-assisted products, to protect the internal effort that deep learning depends on.
• Administrators. Attend to multitasking learners specifically: polychronicity strengthened the effective-use-to-over-reliance path (β = 0.261) and weakened the effective-use-to-performance path (β = 0.113), so structured AI protocols and focused-attention routines matter most for students who habitually split attention.

Limitations

• The cross-sectional, self-report survey design limits causal inference despite the time-lagged structure.
• The sample is drawn from Chinese university students, bounding generalizability.
• Polychronicity and the AI-literacy/AI-evaluation constructs rely on self-report operationalizations.
• The qualitative strand, while informative, is limited to educator perspectives and does not directly capture students' behavioral mechanisms.

Citation

Gao, L., Sun, Y., & Khan, S. U. (2026). From enhancement to over-reliance: A mixed-method study of generative AI and sustainable learning performance.