Work

Managing Hallucination Risk in LLM-Generated Outputs: The Roles of Prompting Literacy, Trust Calibration, and Verification Behavior

Year: 2026
Type: conference-paper
Study design: Observational Study
Abstract: Large language models (LLMs) are increasingly used by student users for academic learning and knowledge work, but their ability to generate fluent yet inaccurate outputs creates hallucination risk. Prior research has mainly examined generative AI adoption and ethical concerns, while less is known about how student users verify LLM-generated outputs before relying on them. Drawing on AI literacy, epistemic vigilance theory, trust calibration, and processing fluency theory, this study develops a dual-path model examining how prompting literacy is associated with hallucination-risk management among student users. Survey data were collected from 210 students who had used LLM tools for academic tasks. After removing 10 straight-lining responses, 200 valid responses were analyzed using PLS-SEM. The findings indicate that prompting literacy is positively associated with hallucination awareness, trust calibration, and perceived output fluency. Hallucination awareness and trust calibration are positively associated with scenario-based self-reported verification behavior, which is strongly associated with responsible LLM use. However, perceived output fluency is not significantly associated with verification behavior. The findings suggest that responsible LLM use among student users is associated mainly with protective mechanisms of awareness, calibrated trust, and verification rather than with fluency-based risk. This study contributes by shifting attention from LLM adoption to verification behavior under hallucination risk and by positioning verification behavior as an important pathway toward responsible LLM use in academic contexts.
Institution: FPT University
Language: English
Subfield: Education
Open Access status: gold
OpenAlex ID: W7164828862
FWCI: 30.22
Cites: 20
Cited by: 2
Related to: 10