AI evidence extraction

Shaping responsible GenAI use in research through AI literacy-oriented guidelines: insights from postgraduate students

Wei Dai, Cecilia Ka Yuk Chan · International Journal of Educational Technology in Higher Education · 2026

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

9/10

Relevance

1/4

Quality (LMQS)

Evidence

Citations

0.00

FWCI

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.1186/s41239-026-00609-6

Methodology & findings

Study design

Seven semi-structured focus group interviews with 28 postgraduate research (PGR) students from various faculties at the University of Hong Kong.

Sample

N = 28, 7 groups

Primary method

Qualitative thematic analysis. Inductive coding guided by research questions. Two researchers conducted independent coding of first transcript to develop initial coding framework, compared coding, discussed interpretive differences, and refined codebook through consensus. Remaining six transcripts coded independently (three each) with iterative refinement of codebook. Each coded interview reviewed by second researcher for consistency and alignment. Interpretive disagreements resolved through discussion. Final thematic structure developed through collaborative review and iterative discussion among researchers. Adequacy of thematic structure assessed by considering whether final themes captured recurring patterns across interviews and provided sufficient coverage of four research questions.

Main result

The study found that "PGR students are not passive adopters of GenAI but active users who enact AI literacy across all four of its dimensions in their research practice. They demonstrated 'Know & Understand AI' through awareness of GenAI's task-specific capabilities and limitations, 'Use & Apply AI' through diverse, discipline-sensitive applications across the research workflow, 'Evaluate & Create AI' through nuanced boundary judgements between assistance and substitution, and 'AI Ethics' through reasoned concern about accuracy, authorship, originality, data integrity, and the development of scholarly competence."

Reports effect sizes.

Research paradigm

Interpretivist/constructivist

Author conclusions

"The rapid integration of GenAI into research has outpaced the development of guidance that is specific, practical, and grounded in researchers' lived experience, leaving postgraduate researchers, among the most active adopters of GenAI tools, to navigate complex ethical decisions without adequate institutional support." The authors conclude that "PGR students are not passive adopters of GenAI but active users who enact AI literacy across all four of its dimensions in their research practice" and propose "task-sensitive guidelines that engage all four AI literacy dimensions across diverse research tasks, designed to function not only as a practical resource for responsible GenAI use but also as a developmental scaffold for the sustained growth of researchers' AI literacy."

Risk of bias

Selection bias: Non-purposive sampling resulted in uneven disciplinary distribution with higher proportion from social sciences and limited perspectives from Law and other disciplines; Social desirability bias: Participants may have underreported certain GenAI uses due to concerns about academic integrity or fear of consequences; Attrition/documentation bias: Two focus groups documented through written notes rather than audio recordings, potentially losing nuance in participant wording and group interaction; Self-report bias: Study relied entirely on self-reported interview data without triangulation through observational data or system logs; Responder representativeness: Sample limited to one institution (University of Hong Kong) where students were recruited via invitation email; Selection bias: Participants self-selected through invitation emails; uneven disciplinary distribution; Social desirability bias: Participants may have underreported AI use due to concerns about academic integrity or cheating stigma; Response bias: Differential measurement quality between audio-recorded (5) and written-note (2) focus groups; Lack of supervisor perspectives: Study centered on student accounts without validating against supervisor views; Self-reported data without triangulation: No observational data, system logs, or anonymous surveys to validate claims; Selection bias: non-purposive sampling, uneven disciplinary representation; Social desirability bias: participants may underreport GenAI use due to concerns about academic integrity; Response bias: participants may fear consequences of disclosing certain practices; Documentation bias: two focus groups recorded as written notes rather than audio transcripts; Researcher interpretation bias: thematic analysis reliant on researchers' interpretive judgement despite consensus procedures

Limitations

"First, participants were not purposively sampled to ensure representation across faculties, resulting in an uneven disciplinary distribution
For example, the sample included a higher proportion of students from the social sciences and limited perspectives from disciplines such as Law." Additionally, "this study did not systematically examine departmental-level GenAI guidelines, nor did it directly investigate supervisor perspectives or supervisor-student dialogue around GenAI." Furthermore, "the reliance on self-reported interview data may introduce social desirability or response bias
Participants may have underreported certain uses of GenAI due to concerns about academic integrity, stigma associated with 'cheating', or fear of potential consequences." The authors also note that "two focus groups were documented through written notes rather than audio recordings at participants' request," which "provide less detail than verbatim transcripts and may not fully capture the nuance of participants' wording or group interaction."

Open questions raised

Future studies should involve participants from a broader and more balanced range of disciplines to examine disciplinary differences in GenAI use and ethical reasoning more systematically
Future research should examine departmental variation in GenAI guidance, incorporate supervisor perspectives, and investigate how supervisor-student conversations shape the development of AI literacy in research
Systematic examination of variation by discipline, stage of study, familiarity with GenAI, or type of research task through purposive sampling across disciplines, larger-scale surveys, or comparative qualitative designs
Complementing self-reported accounts with observational data, system logs, or anonymous surveys to triangulate findings and reduce biases
Continuous revisiting and updating of empirical understandings of GenAI use to keep pace with rapid technological change
Disciplinary variation in GenAI use and ethical reasoning - future studies should involve broader and more balanced sample across disciplines

Extracted from: pdfAgreement 73%

Explore related topics

AI Literacy Training AI Governance Qualitative