Same tool, different work: patterns of generative AI use and academic outcomes

Same tool, different work: patterns of generative AI use and academic outcomes — a survey study of 157 undergraduates showing that how students use GenAI matters more for academic outcomes than how often they use it. Stamatoulis et al. (2026) identify two distinct patterns of use — evaluative integration (EI), using GenAI to support understanding, and low-verification uptake (LVU), using it as a shortcut — with opposite associations with performance.

The study argues that research on students' GenAI use has relied too heavily on frequency, which captures exposure but not how students engage. Using exploratory factor analysis of items developed from pilot data, it isolates two qualitatively different use patterns and shows that their associations with performance diverge sharply — while frequency predicts neither performance nor academic self-efficacy.

Method

• Design: Survey; exploratory factor analysis of GenAI-use items developed inductively from pilot data; path modeling.
• Sample: 157 undergraduates.
• Key variables: Evaluative integration (EI, use to support understanding), low-verification uptake (LVU, shortcut use), academic self-efficacy (ASE), academic performance; frequency included for comparison.

Key Findings

• Evaluative integration (EI) was positively associated with performance; low-verification uptake (LVU) was negatively associated with it.
• Mediation via academic self-efficacy: EI's association with performance was fully indirect through ASE (no remaining direct effect); LVU showed both a significant indirect association and a remaining direct negative association.
• Frequency predicted neither performance nor ASE — a striking null result supporting the study's core claim that patterns of use, not amount, matter.
• In a separate model, frequency strengthened the EI→ASE positive association, though the interaction was not significant when EI and LVU interactions were estimated jointly; no moderation was found for LVU.

Implications

• For higher education and AI Literacy: measuring how students use GenAI (evaluative integration vs. shortcut uptake) is more informative than frequency — supporting measures of student–GenAI interaction and pedagogies that promote evaluative integration.
• For over-reliance and the performance–learning gap: LVU (uncritical shortcut use) mirrors the learning-harm pattern, while EI (understanding-oriented use) aligns with productive learning — connecting to Reducing AI Misuse.
• For Self Efficacy: EI's full mediation through academic self-efficacy suggests understanding-oriented use builds confidence, whereas shortcut use predicts poorer outcomes partly independent of it.

Connected Concepts

• Generative AI
• Higher Ed
• AI Literacy
• Academic Integrity
• Cognitive Offloading
• Self Efficacy
• Reducing AI Misuse
• AI Misuse Learning Harm
• Student Experience

Connected Articles

• Walton Bearman Assessment Judgement 2025 — Judgement in students' work with GenAI on assessment
• Jiang GenAI Activity Theory Disciplines 2026 — Disciplinary differences in GenAI use and disclosure
• Learning To Learn In The Age Of Generative AI A Scoping Review And Conceptual Fr — Scoping review of GenAI and learning
• GenAI Thoughtless Use Self Directed Learning 2026 — Thoughtless GenAI use and self-directed learning

Citation

Stamatoulis, C., Pyrovetsi, L., Mourikis, C., Ponnam, A., & Karayianni, I. (2026). Same tool, different work: patterns of generative AI use and academic outcomes. Manuscript submitted for publication.