Is using artificial intelligence tools for academic work cheating? Student perceptions, ethics, and the impact on academic performance and critical thinking

Synthesis: This PRISMA 2020 systematic review synthesizes recent research on whether students regard AI-assisted academic work as cheating, finding that student perceptions sit on a spectrum—ranging from viewing AI as a legitimate learning resource to seeing it as misconduct—depending on task type, transparency, and assessment design. A key theme is the persistent gap between student and faculty interpretations of acceptable AI use, alongside evidence that over-dependence on AI can drive Cognitive Offloading, reduced independent reasoning, and altered academic behavior. The review argues that the future of Higher Education lies in concrete ethical standards, promoted AI Literacy, and reshaped Assessment design that fosters Generative AI collaboration without compromising originality, equity, or intellectual growth.

Key Findings

• Mixed student perceptions: Student attitudes toward AI use are neutral and context-dependent—ranging from viewing it as a valid learning assistance resource to considering it a cheating method—shaped by task type, transparency, and how the assessment is designed. Students commonly frame AI tools as productivity aids comparable to search engines or grammar checkers.
• Student–faculty perception gap: Students tend to perceive AI as acceptable help (idea generation, language improvement), while faculty weight originality and independent thinking and are more likely to treat certain uses as Academic Integrity misconduct. Perceptions also vary by discipline: STEM students show stronger agreement that AI improves performance (+42% net agreement), whereas humanities students show the highest net agreement that AI usage constitutes cheating (+44%) and raises ethical concerns (+58%).
• Correlation structure: A pairwise analysis (N = 320) found AI usage negatively correlated with perceived cheating (r ≈ −0.51) and positively correlated with academic performance (r ≈ +0.44), while ethical concern tracked closely with perceived cheating.
• Cognitive offloading risk: Evidence indicates over-dependence on AI and adaptive systems can drive Cognitive Offloading, decreased independent reasoning, and altered academic behavior, threatening Critical Thinking—though guided, transparent AI use can support higher-order thinking.
• Detection limits and AI literacy: Conventional AI Detection struggles with generative text (false positives/negatives), pushing institutions toward process-based assessment. Students with higher AI Literacy use AI more responsibly and show lower tendency to commit academic misconduct, suggesting education rather than prohibition is the more effective lever.

Study Design & Method

A systematic literature review following the PRISMA 2020 framework. Four databases (Scopus, Web of Science, IEEE Xplore, PubMed) were searched from January 2019 to December 2025, spanning pre- and post-ChatGPT literature on AI ethics in education, academic misconduct, and educational technology. From 2,847 initial records, 624 duplicates were removed, 1,891 were excluded at abstract screening, and 332 full-text articles were read; 243 were excluded on documented criteria, leaving 89 studies for thematic synthesis. The review also reports secondary quantitative analyses (e.g., a pairwise scatter matrix across five core constructs and a grouped diverging bar chart on AI-ethics statements by discipline).

What this means for practice

• Instructors. Shift graded work from end-product evaluation to process-based Assessment that captures revision history, interaction patterns, and time-on-task, paired with transparent AI-use disclosure.
• Instructors. Require students to explain their reasoning when AI assists a task: supportive applications such as tutoring and adaptive feedback sustain Critical Thinking on that condition and turn into cognitive offloading when used as a shortcut.
• Instructors. Publish task-specific rules for acceptable AI help rather than assuming students and faculty interpret the same use identically, since students judge idea generation and language improvement as legitimate help while faculty weight originality and independent thinking.
• Instructors. Apply clear, adaptable AI Literacy and Ethics standards in your own course and feed what you learn back into institutional review, because the ethical reading of AI use is situational and technology outpaces policy.
• Instructors. Use education and guided human-AI collaboration rather than detection and punitive enforcement in your classroom, since automated AI Detection produces false positives and false negatives and erodes student trust.

Limitations

• The review synthesizes 89 studies screened from 2,847 records across Scopus, Web of Science, IEEE Xplore, and PubMed, and presents a thematic synthesis rather than a meta-analysis, so it reports correlations (e.g., r ≈ −0.51 between AI usage and perceived cheating) but no pooled effect of AI use on learning.
• The correlational evidence comes from a secondary pairwise analysis of N = 320 rather than a purpose-designed study, and the review covers literature published from January 2019 to December 2025.
• Included studies are heterogeneous in design, and most measure perceptions through self-report instruments rather than observed AI use or demonstrated learning, so the student–faculty gap and the cognitive-offloading risk are the most reported, and least measured, findings.

Citation

Padhy, A. (2026). Is using artificial intelligence tools for academic work cheating? Student perceptions, ethics, and the impact on academic performance and critical thinking. International Journal of Applied Resilience and Sustainability.