The DOI matches its Crossref record; 32 of the 32 references it cites were found in a registry or an index; 4 of the 12 sources its claims cite were checked for support.
Choosing not to use generative AI in higher education: a mixed methods study of student reasoning, assessment design, and AI literacy
Open accessAbstract
Generative artificial intelligence tools are rapidly becoming embedded in higher education, yet little is known about students who deliberately choose not to use such tools, even when permitted. This exploratory study examines how non-using students’ reason about GAI in relation to learning, assessment, and academic integrity. Using a mixed-methods design, descriptive data on GAI use were combined with in-depth qualitative interviews with biomedical laboratory science students in Sweden who refrained from using GAI during a take-home examination followed by a viva voce assessment. This assessment format provided an authentic opportunity for students to engage with AI tools if they chose to, reflecting real-world educational conditions where digital assistance is increasingly available. Among students completing the take-home examination ( n = 46), most reported using GAI, while a small group reported deliberate non-use; two non-users participated in semi-structured interviews. Thematic analysis identified ethical considerations, perceptions of learning depth, and assessment design, as key factors shaping non-use. In particular, the viva voce examination functioned as a regulatory mechanism that foregrounded responsibility for independent understanding and reduced the perceived value of GAI-generated outputs. Non-use was not accompanied by lower pass rates in the viva voce examination and was interpreted as a context-sensitive, reflective form of AI literacy rather than resistance to technology. These findings highlight the importance of assessment design in shaping meaningful and equitable engagement with GAI and suggest that institutional policies should support both users and non-users of AI tools in higher education.
Overview
CoChat Summary
TL;DR
This exploratory mixed-methods study examines why biomedical laboratory science students at Karolinska Institutet deliberately declined to use generative AI during a take-home examination where its use was permitted. Non-use was interpreted as a context-sensitive, reflective form of AI literacy shaped by ethical concerns, perceptions of learning depth, and the regulatory effect of the follow-up viva voce examination.
Key findings
- Among students completing the take-home examination (n = 46), 40 reported using GAI and 6 reported non-use. Of the users, 28 passed the viva voce (28/40; 70%), 7 failed, and 5 did not participate; among non-users, 5 passed (5/6; 83%), 1 did not participate, and none failed. The authors report these differences descriptively and did not analyze them inferentially.
- Thematic analysis of two interviews yielded six themes: (1) Motivations for (Non)Use, (2) Assessment Design as a Regulatory Mechanism, (3) Learning Depth vs. Efficiency, (4) Feedback and Academic Relationships, (5) Reliability and Source Criticism, and (6) Institutional Needs: Guiding, Training, and Ethics.
- The viva voce functioned as a regulatory mechanism: participants saw little value in AI-generated material when understanding had to be demonstrated orally, and one suggested GAI use would increase if the oral examination were removed.
- Non-use was framed as principled (ethical and environmental concerns, perceived threat to deep learning) or pragmatic (AI "felt unnecessary" given the oral exam), and the authors argue it should not be equated with low AI literacy.
Methods
- Exploratory mixed-methods design combining descriptive cohort-level data on self-reported GAI use with in-depth qualitative interviews; reported per STROBE and COREQ guidelines.
- Setting and sample: second-year biomedical laboratory science students at Karolinska Institutet, Stockholm, in 2025 (N = 47 eligible; 1 excluded for missing disclosure data; n = 46 analyzed). The 5-week, 5-ECTS microbiology course used a take-home examination with permitted, disclosed GAI use followed by a standardized online viva voce limited to 25 min per student. Cohort median age was 24 years (range 21–58) and 89% were female; the two interviewed non-users had a median age of 25.5 years and both were female.
- Analysis: categorical data summarized descriptively in Excel 2016 and Prism 9.5.0 with listwise exclusion of missing data; interview transcripts (approximately 50 min each, conducted onsite after course completion, audio-recorded and transcribed verbatim) analyzed using Braun and Clarke's six-phase coding thematic analysis, with two researchers coding and discussing codes to agreement.
Limitations
- Stated by the paper: the qualitative sample was small, comprising only two interviews with self-identified non-users, limiting transferability; findings are shaped by context-specific factors (Karolinska Institutet, the Biomedical Laboratory Science program, a single data-collection time point); the absence of qualitative data from AI users means differences in reasoning between users and non-users cannot be directly examined; and the study lacked triangulation across multiple data sources, methods, or participant groups.
- Evident in the text: the quantitative comparison of pass rates rests on very small cell counts (6 non-users, 5 of whom passed) and the authors themselves note it was not analyzed inferentially, so no performance conclusion can be drawn; the interviewer had previously taught the participants, a power dynamic the authors acknowledge and attempt to mitigate through voluntariness, post-course timing, and non-involvement in assessment.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
4 Valid · 0 Not valid · 8 Not checked · 78 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Prior studies also indicate that high levels of AI literacy may manifest differently among both users and non-users, ranging from selective and confident use to deliberate restraint or normative uncertainty regarding academic boundaries.”
What the sources say
- Partly supportedcites Adamakis and Rachiotis, 2025 sourceExperimental — accuracy not yet measured
Section 3.1 describes AI literacy as critical, ethical engagement and warns that unselective AI use may undermine learning. Section 7 links inconsistent academic AI policies to uncertainty. Neither section establishes that high literacy manifests differently among users and non-users, or that non-use reflects deliberate restraint or uncertainty about academic boundaries. Page numbers were not available in our copy. (LLM, conf=0.82)
hedgedstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Within this framework, non-use should not be equated with lack of understanding but rather interpreted as a context sensitive judgement informed by responsibility for long-term learning rather than technical incapability (Adamakis and Rachiotis, 2025).”
What the sources say
- Partly supportedcites Adamakis and Rachiotis, 2025 sourceExperimental — accuracy not yet measured
Section 4.1 says AI tool selection requires pedagogical judgment rather than purely technical assessment, and section 3.1 warns that unselective use may undermine independent learning. The paper does not explicitly say that non-use should be interpreted as an informed choice rather than a lack of understanding or technical capability. Page numbers are unavailable in our copy. (LLM, conf=0.82)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: unclear — It is unclear whether it could be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Participants described GAI as simultaneously helpful and potentially counterproductive, highlighting a tension between the efficiency AI provides and the deeper learning processes required for long-term learning.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Participants perceived little value in relying on AIgenerated material when understanding would later need to be demonstrated orally.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The second participant framed non-use as a functional decision shaped mainly by assessment demands particularly the requirement to demonstrate independent reasoning during the viva voce examination: ' Not a statement… the oral exam means I must know it myself; AI felt unnecessary' (I2).”
plainstates a causeSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
36 more claims
Create a free account to see all 90 claims and every source verdict.
- Probably not a substantial claim. 41
- It may state two separate claims as one. 24
- The quote only partly states this claim. 1
Citations
Cites
Using thematic analysis in psychology
Braun, V., & Clarke, V. (2006) · Qualitative Research in Psychology · doi:10.1191/1478088706qp063oa · Cited by 194,414
Tong, A., Sainsbury, P., & Craig, J. (2007) · International Journal for Quality in Health Care · doi:10.1093/intqhc/mzm042 · Cited by 42,509
von Elm, E., Altman, D. G., Egger, M., Pocock, S. J., Gøtzsche, P. C., Vandenbroucke, J. P., & Initiative, F. T. S. (2007) · Annals of Internal Medicine · doi:10.7326/0003-4819-147-8-200710160-00010 · Cited by 17,796
What is AI Literacy? Competencies and Design Considerations
Long, D., & Magerko, B. (2020) · Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3313831.3376727 · Cited by 3,241
Students' voices on generative ai: perceptions, benefits, and challenges in higher education
Chan, C. K. Y., & Hu, W. (2023) · arXiv · doi:10.48550/arxiv.2305.00290 · Cited by 2,504
Ray, P. P. (2023) · Internet of Things and Cyber-Physical Systems · doi:10.1016/j.iotcps.2023.04.003 · Cited by 2,463
Students’ voices on generative AI: perceptions, benefits, and challenges in higher education
Chan, C. K. Y., & Hu, W. (2023) · International Journal of Educational Technology in Higher Education · doi:10.1186/s41239-023-00411-8 · Cited by 2,301
Zhai, C., Wibowo, S., & Li, L. D. (2024) · Smart Learning Environments · doi:10.1186/s40561-024-00316-7 · Cited by 1,567
ChatGPT and Open-AI Models: A Preliminary Review
Roumeliotis, K. I., & Tselikas, N. D. (2023) · Future Internet · doi:10.3390/fi15060192 · Cited by 706
GPT detectors are biased against non-native English writers
Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023) · Patterns · doi:10.1016/j.patter.2023.100779 · Cited by 673
Cited by
Saravanan, P., Meissner, C., & Glenn, J. (2026) · Frontiers in Education · doi:10.3389/feduc.2026.1880863 · Cited by 4
Generative AI Use Patterns and Ethical Dilemmas among Pre-Service Teachers
Andayani, M. S. L., & Damayanthi, L. P. E. (2026) · Jurnal Riset dan Inovasi Pembelajaran · doi:10.51574/jrip.v6i2.5619
Marín, F. J. S., & Legaz, A. J. C. (2026) · Frontiers in Education · doi:10.3389/feduc.2026.1947527
Daya, H. (2026) · International Journal of Learning Teaching and Educational Research · doi:10.26803/ijlter.25.9.14
Harahap, R. U., & Fahmi, M. (2026) · Ilomata International Journal of Tax and Accounting · doi:10.61194/ijtc.v7i4.2497
Metadata
- Authors
- Lindkvist, A., Curbo, S., & Hultgren, C.
- Affiliations
- Karolinska Institutet (SE)
- Corresponding author
- Hultgren, Catharina
- Venue
- Frontiers in Education
- Publisher
- Frontiers Media SA
- Published
- 2026-04-23
- Volume
- 11
- ISSN
- 2504-284X (ISSN-L)
- OpenAlex
- W7155389471
- Cited by
- 5
- References
- 29
- Citations by year
- 2026: 5
- FWCI
- 42.53
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- Frontiers in Education (publishedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Topics
- Social Sciences › Education › Artificial Intelligence in Education
- Social Sciences › Education › Academic integrity and plagiarism
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Keywords
- generative AI, AI literacy, assessment design, academic integrity, viva voce examination, student reasoning, ethical considerations, higher education, institutional AI policies
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Choosing not to use generative AI in higher education: a mixed methods study of student reasoning, assessment design, and AI literacy. https://app.cochat.ai/papers/10.3389/feduc.2026.1813306
- The paper
- Lindkvist, A., Curbo, S., & Hultgren, C. (2026). Choosing not to use generative AI in higher education: a mixed methods study of student reasoning, assessment design, and AI literacy. Frontiers in Education. https://doi.org/10.3389/feduc.2026.1813306
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.