Shaping Responsible GenAI Use in Research Through AI Literacy-Oriented Guidelines: Insights From Postgraduate Students

Synthesis: Wei Dai and Cecilia K. Y. Chan (2026) examined how 28 postgraduate research (PGR) students across seven focus groups enacted AI Literacy in their use of generative AI for research, arguing that responsible-use guidance must extend beyond teaching and assessment into the ethically more complex research process. Analyzing practices through a four-dimensional AI literacy framework, the study found students' awareness of GenAI's capabilities, limitations, and risks reflected the "Know & Understand" dimension; their diverse, discipline-specific applications across the research workflow reflected "Use & Apply"; their self-defined boundaries between ethical and unethical use reflected "Evaluate & Create"; and concerns about output accuracy, originality, data privacy, and skill degradation reflected the "AI Ethics" dimension. Building on these insights, the authors proposed researcher-oriented GenAI guidelines that foreground each AI literacy dimension across research tasks, positioning the guidelines less as rule enforcement and more as a developmental scaffold for researchers' sustained AI literacy growth.

Key Findings

• Researchers are active, calibrated users, not passive adopters: 27 of 28 participants used GenAI across ideation, literature review, explanation, data processing, programming, academic writing, editing, and translation, but matched tools to tasks based on perceived stakes, intellectual demands, and disciplinary norms — heavier use in low-stakes procedural work and more cautious engagement in tasks central to scholarly contribution.
• Human oversight retained throughout: GenAI was positioned as a supportive, efficiency-enhancing research assistant rather than a substitute for core intellectual work; participants did not adopt AI-generated outputs uncritically, recognizing that ideas could be "misleading" or "totally wrong."
• Nuanced boundary judgments on writing: Students distinguished between acceptable drafting assistance and full generation in academic writing, and between AI helping to structure versus substitute their own reasoning.
• AI ethics concerns cluster around scholarly integrity: Worries centered on non-deterministic behavior, limited capacity for genuine originality, output accuracy, data privacy, and the degradation of researchers' own scholarly competence — concerns that map to the "AI Ethics" dimension.
• Policy gap: Existing institutional GenAI policies focus on teaching, learning, and assessment and remain abstract; research-specific guidance that is practical, task-sensitive, and grounded in researchers' lived experience is largely absent.

Connection to the broader knowledge base

The study contributes an empirical account of how AI Literacy is enacted as a situated capacity in research practice, rather than a static set of competencies, with direct implications for integrity and Ethics policy in graduate education. For institutions it argues that responsible GenAI-use guidelines should move beyond binary rules toward practice-oriented support that scaffolds each AI literacy dimension across the research workflow — a stance that treats AI Governance and researcher development as mutually reinforcing. For supervisors and research-training programs it signals a need to cultivate evaluative judgment and self-regulation of GenAI use, echoing calls to embed AI literacy across the entire scholarly lifecycle rather than only in teaching and assessment contexts.

What this means for practice

• Researchers. Define in writing which research tasks allow GenAI assistance and which do not before starting a project, mirroring how participants calibrated use by task stakes and intellectual centrality rather than by blanket permission.
• Supervisors. Teach AI literacy as a research competence rather than a compliance topic: work through concrete, discipline-sensitive cases across the workflow — ideation, literature review, data processing, writing, translation — so students practice boundary judgments on real decisions.
• Institutions. Write research-specific GenAI guidance, because existing policies stop at teaching, learning, and assessment and participants experienced them as abstract and misaligned with research practice.
• Supervisors. Normalize disclosure of GenAI use in supervision and progress reviews, since participants weighed the stigma of "cheating" and may have withheld part of their actual practice.
• Researchers. Keep GenAI in the drafting and efficiency role and draw the line where it would replace your own reasoning — participants themselves distinguished AI structuring their thinking from AI doing it for them.

Limitations

• A single institution (University of Hong Kong) and 28 PGR students across seven focus groups; participants were not purposively sampled across faculties, giving an uneven disciplinary distribution with a higher share from the social sciences and limited representation from Law.
• The design did not systematically examine departmental GenAI guidelines or supervisor perspectives, and supervisor references appeared only in passing.
• Variation by discipline, stage of study, and familiarity with GenAI was not systematically analyzed, limiting the basis for comparative claims.
• Reliance on self-reported interview data may introduce social desirability or response bias, and students may have underreported use out of integrity concerns.

Citation

Dai, W., & Chan, C. K. Y. (2026). Shaping responsible GenAI use in research through AI literacy-oriented guidelines: Insights from postgraduate students. International Journal of Educational Technology in Higher Education, 23, 33.