{"id":"91ea59a1-5d41-4711-b75b-867756c212cb","arxiv_id":"2501.08864","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Interviews show university students self-govern their GenAI use under unclear guidelines, developing unspoken rules and reliance strategies amid plagiarism-focused institutional messaging.","lead":"This paper reports interviews with 26 students and 11 educators at two UK universities about how and why they use generative AI tools in coursework. It describes unspoken rules, reliance strategies, and worries about plagiarism and skills, and proposes a framework for how university policies and communication shape these practices.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'unspoken rules' finding may be an artifact of direct prompting in the interview protocol; the paper does not show these norms emerged spontaneously, so the self-governance claim is less secure.","rationale":"I read the paper as a qualitative, contextual account of emerging GenAI practices, and I take its central claim to be that students develop tacit norms ('unspoken rules') and reliance strategies in response to unclear, plagiarism-focused institutional guidance. The most load-bearing assumption is not only the representativeness of the sample (which the authors acknowledge) but whether the named findings reflect students' own spontaneous frameworks. The interview protocol directly asks about 'unspoken rules' and 'cheating', so the study's signature finding may be partly co-produced by the research instrument. This is a concrete, checkable threat to construct validity. The paper is otherwise careful: it reports counts, provides participant IDs, acknowledges limitations, and uses a post-hoc theoretical framework transparently. A simple transcript re-analysis would settle whether the themes are grounded in unprompted accounts. If they are, the concern is resolved; if not, the paper's main contribution should be downgraded from 'unspoken rules' to 'reported normative beliefs'. I therefore recommend no change to the reader's CONDITIONAL verdict, but with a specific condition: the authors should provide the spontaneous-mention analysis or modify the claim.","tokens_in":40162,"tokens_out":7217,"duration_ms":73951,"concrete_test":"Re-analyze the cleaned interview transcripts and code, for each participant, whether the three boundary rules (reference vs. plagiarise; edit vs. copy-paste; assist vs. do the work) and any reliance strategy (e.g., double-checking) were mentioned spontaneously before the interviewer asked the protocol questions about 'unspoken rules', 'cheating', or 'double-checking'. Report the number of participants with at least one spontaneous mention. If fewer than half of the 26 students spontaneously articulate at least one rule before prompting, the 'unspoken rules' theme should be reframed as an elicited normative account rather than an emergent practice, and the self-governance claim in the abstract should be softened.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that students self-govern GenAI use through 'unspoken rules' and reliance strategies in the absence of clear institutional guidance. However, the semi-structured interview protocol (Appendix B) explicitly primes these constructs: participants are asked 'Are there any \"unspoken rules\" you follow when using these new AI tools?' and 'Do you think using these new AI tools is like cheating?' and 'Do you double-check the answers...?' The findings in §5.4.1 report counts such as 13/26 and 14/26 for the boundary rules, but the paper never reports whether any participant articulated a rule before the moderator used the term. If the rules appear mainly in response to direct normative prompts, the conclusion that students have developed tacit, self-generated norms is not established; the themes may reflect the researchers' categories rather than students' own frameworks. This matters because the contribution is precisely the discovery of these emergent, unspoken norms. The same concern applies to reliance strategies (e.g., double-checking) in §5.4.2, which is directly prompted. A related weakness is the reliance on self-reports for actual behavior, but the more acute problem is the risk of leading questions. This does not invalidate the study, but it weakens the central interpretive claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a qualitative interview study of 26 students and 11 educators from two UK universities, examining how students use generative AI in higher education, what norms and reliance strategies shape that use, and what impacts they perceive. The authors apply Strong Structuration Theory post hoc to organize themes into a model with actions, internal structures (motivations, self-governance), external structures (university guidelines, educator communication), and outcomes (confidence, skill concerns, relationships, plagiarism anxiety, future expectations). The central claim is that, in the absence of clear institutional guidance and amid a plagiarism-focused policy environment, students develop their own 'unspoken rules' (reference, do not plagiarize; edit, do not copy-and-paste; assist, not do-for) and reliance strategies, with implications for their skills, confidence, relationships with educators, and anxiety about plagiarism. The paper includes full interview and survey protocols in appendices and explicitly discusses limitations and positionality.","tokens_in":40414,"tokens_out":4086,"duration_ms":42252,"significance":"If the findings are robust, the paper makes a useful contribution to HCI and higher-education research by providing an ecological, context-sensitive account of early-stage GenAI adoption, moving beyond attitude surveys to describe student practices, norms, and the structural conditions that shape them. The comparative student-educator design is a strength, and the SST-based model offers a framework for future work on policy, assessment design, and AI literacy. The paper also explicitly publishes its instruments and acknowledges its limitations, which supports transparency. However, the exploratory sample is small, self-selected, and heavily skewed toward humanities students and junior educators, and the central 'unspoken rules' claim is partly elicited by the interview protocol's direct wording. These issues constrain the generalizability and the strength of the interpretive claim, but the underlying data and model are valuable enough to warrant revision.","major_comments":[{"comment":"The central claim that students follow 'unspoken rules' is not fully supported by the evidence as presented, because the semi-structured interview protocol explicitly asks participants 'Are there any \"unspoken rules\" you follow when using these new AI tools?' and 'Do you think using these new AI tools is like cheating?'. The paper does not report whether any participant articulated a norm or boundary rule before the moderator introduced the term, nor does it distinguish between spontaneous mention and prompted agreement. As the contribution depends on the emergence of these norms from students' own framing, the authors should either re-analyze the transcripts to show that rules were mentioned unprompted, or revise the interpretation to state that, when asked, students articulated such norms. Without this, the 'unspoken rules' theme risks being an artifact of the research instrument rather than a discovery about students' tacit self-governance.","section":"§5.4.1, Appendix B"},{"comment":"Several educator-frequency counts are reported with a denominator of 26 instead of 11, which is inconsistent with the educator sample size. Concretely, the 'Educators' perspective' paragraphs in Sections 5.6.2, 5.6.3, and 5.6.4 report (6/26), (5/26), (7/26), and (4/26) when the correct denominator is 11. Because the paper uses counts to indicate the prevalence of themes among educators, these errors undermine the trustworthiness of the quantitative summaries. The authors should correct these values and audit all other sections for similar mistakes, since the current misreporting weakens the evidentiary base for the educator-related claims.","section":"§5.6.2, §5.6.3, §5.6.4"},{"comment":"The limitations section is candid about the sample composition and the post hoc use of SST, but it does not address the more acute construct-validity threat posed by the interview protocol's direct prompting of 'unspoken rules' and 'cheating'. The authors should add a paragraph that acknowledges this possible framing effect, explains what evidence (if any) indicates that participants held these norms independently of the questions, and qualifies the claims accordingly. This is a load-bearing issue because the paper's key contribution is the discovery of these self-governance norms.","section":"§4.4"}],"minor_comments":[{"comment":"There is a minor spacing error in 'students(17/26)'; it should read 'students (17/26)'.","section":"§5.3.1"},{"comment":"The sentence 'It is okay to use GenAI within certain boundaries ,' has an extra space before the comma and should be polished.","section":"§5.4.1"},{"comment":"In the sentence beginning 'Students desired education about the capabilities...', there is a missing space before the count '(15/26)'.","section":"§5.6.2"},{"comment":"The phrase 'will need to built into educational practice' should read 'will need to be built into educational practice'.","section":"§6.2.1"},{"comment":"The word 'dint' in 'as a dint to their confidence' should be 'dent'.","section":"§5.5.2"}],"recommendation":"major_revision","confidential_remarks":"This is a solid qualitative study with full transparency about methods and a relevant contribution to HCI and higher-education research. The two major issues are fixable: the authors should either provide evidence of unprompted norm articulation or soften the 'unspoken rules' framing, and they must correct the educator-denominator errors in Section 5.6. The paper fits the journal's scope. I would not reject on the basis of sample size, as the authors already acknowledge that limitation, but the prompting issue needs to be addressed before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is a genuine, careful qualitative study and worth a proper review. The core contribution is a detailed interview dataset: 26 students and 11 educators across two UK universities, with full protocols in the appendix, a transparent positionality statement, and clearly acknowledged limitations. The SST-based model tying university guidelines, educator communication, student motivations, and self-governance into one map is useful, even if the framework is applied post hoc (which the authors admit). The paper also does a fair job of situating itself in the existing CS-heavy literature and gives a nuanced account of how students use GenAI as tutor, assistant, or ideation partner.\n\nThe stress-test concern about 'unspoken rules' is legitimate. Appendix B shows that participants were directly asked 'Are there any \\\"unspoken rules\\\" you follow when using these new AI tools?' So the finding that students have unspoken rules is at least partly an artifact of prompting. That doesn't make the norms unreal, but the paper's framing of them as naturally emergent self-governance is weaker than claimed. The same applies to double-checking and other reliance strategies, which are also explicitly prompted. I'd want the authors to either acknowledge this directly or soften the language.\n\nThe educator sections contain clear errors: several counts use 26 as the denominator when the educator sample is 11 (e.g., 6/26 in 5.6.2 and 5.6.3, 7/26 and 4/26 in 5.6.4). These are fixable, but they undercut confidence in the educator findings until corrected. The sample is also heavily humanities, only three educators had taught undergrads since ChatGPT, and self-selection is acknowledged — all fair limitations for a qualitative study, but they should stay in the abstract's framing.\n\nOverall, this is a real empirical contribution, not a simulation or a model-fitting exercise. The claims are mostly proportionate, and the authors are transparent about method. The two soft spots — the prompting issue and the count errors — are worth referee attention but not desk rejection. I'd send it to review and ask for targeted revisions.","headline":"Solid qualitative snapshot of GenAI use in higher education, but the 'unspoken rules' finding is partly prompted and the educator counts have denominator errors.","tokens_in":40904,"tokens_out":2285,"would_cite":true,"duration_ms":26612,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Students self-govern their use of generative AI when university rules stay vague.","keywords":["generative AI","higher education","student practices","self-governance","unspoken rules","plagiarism anxiety","qualitative interviews","structuration theory"],"falsifier":"A large-scale observational study that logs students' actual GenAI interactions (rather than interviewing them) and compares them with their institution's stated policies could settle the claim: if students in institutions with clear, well-communicated guidelines still develop the same unspoken rules and plagiarism anxiety, the proposed link between unclear guidelines and self-governance would be undermined.","tokens_in":39981,"feed_emoji":"🎓","tokens_out":4867,"duration_ms":45115,"temperature":0.7,"pith_summary":"This paper argues that university students are not waiting for institutions to define acceptable use of generative AI: in the absence of clear guidelines, they quietly develop their own rules about when and how to use tools like ChatGPT. Drawing on interviews with 26 students and 11 educators at two UK universities, it shows students' choices are governed by unspoken conventions—reference rather than plagiarise, edit rather than copy-paste, assist rather than do the work—and by practical reliance strategies for checking AI output. The study also finds that institutional fixation on plagiarism, inconsistent messages from educators, and fear of being wrongly accused create 'plagiarism anxiety' that can push students to hide their use. If this picture is right, universities' current compliance-focused response is incomplete, and what is needed is clearer communication, responsible-use training, and assessments redesigned for a world where generative AI is already part of students' working habits.","feed_headline":"Students write their own AI rules when universities stay vague","feed_subtitle":"Interviews with 37 students and educators reveal unspoken rules filling the AI policy vacuum.","key_machinery":"The mechanism that carries the argument is students' self-governance of GenAI use, organized into a structuration-based model. It consists of unspoken rules about appropriate use, reliance strategies for handling limitations (double-checking outputs, weighing task importance, and using domain expertise), and considerations of agency and skill development. The model links these internal structures to external structures—university guidelines, educator communication, peer communication—and to outcomes such as confidence shifts and plagiarism anxiety; the paper explicitly marks reciprocal influences it did not explore as dashed arrows.","core_discovery":"The central discovery is that students' generative-AI practices in higher education are shaped less by official policy than by self-governance under ambiguity. Students use GenAI in three roles—tutor, assistant, and ideation partner—and justify their choices through three unspoken rules: 'reference, do not plagiarise'; 'edit, do not copy and paste'; and 'use it to assist you, not to do work for you.' These rules, along with reliance strategies and concerns about skills and agency, emerge in a context of unclear university guidelines, institutional fixation on plagiarism, and inconsistent educator communication; they produce perceived impacts on confidence, skill development, relationships with educators, and plagiarism anxiety. The paper presents this as a guiding model that integrates micro, meso, and macro perspectives and expects both external structures (guidelines, training, assessments) and internal structures (students' motivations and self-governance) to change as GenAI becomes normalised.","pith_inferences":["The paper's sample is skewed toward humanities and self-selected users, so a plausible testable extension is a larger, more diverse survey to quantify how widespread these unspoken rules and plagiarism anxiety are across disciplines and institution types.","The finding that students double-check outputs and weigh task importance suggests that over-reliance may be less severe than feared in this population, but the design implication is to build tools that nudge verification rather than assume it.","If universities adopt clear, permissive-but-bounded guidelines, students' self-governance may become explicit and shared; observing whether plagiarism anxiety drops would test the causal claim that vagueness drives anxiety.","The dashed arrows in the model (outcomes feeding back into structures) could be examined longitudinally to see whether student practices reshape institutional policy, as participants expect."],"forward_implications":["If universities want to shape GenAI use, they need to replace vague, plagiarism-focused guidance with systematic, clearly communicated rules that reflect how students actually work.","Students' reliance on self-governance means many are developing bespoke, inconsistent norms; formal GenAI literacy training could reduce disparities in skill and confidence.","Assessments will need to change, e.g., by incorporating GenAI into assignments or asking students to document and reflect on their use, to preserve academic integrity without punishing legitimate help-seeking.","Open communication among educators about GenAI is currently suppressed by fear of peers' judgement; encouraging these conversations is a prerequisite for consistent teaching practice.","If current patterns persist, students' relationships with educators may weaken as they go to ChatGPT first, while their vocabulary for asking questions and admitting uncertainty may shrink."],"supporting_citations":[{"why":"Supplies the Strong Structuration Theory framework used to organize the analysis into external structures, internal structures, actions, and outcomes.","marker":"[40]"},{"why":"Provides prior evidence on GenAI practices and policy clarity in computing education, the baseline the paper extends to other disciplines.","marker":"[78]"},{"why":"Documents students' perceptions, benefits, and challenges of GenAI, giving the paper a comparative anchor for its practice-focused findings.","marker":"[17]"},{"why":"Offers a qualitative study of ChatGPT's impact that informs the paper's interpretation of plagiarism anxiety and self-governance.","marker":"[83]"},{"why":"Gives the student and teacher perspectives on GenAI in writing that the paper builds on when discussing unspoken rules and assessment changes.","marker":"[8]"},{"why":"Reports student confidence and writing-assistance attitudes that the paper connects to its own findings on skill development and plagiarism concerns.","marker":"[49]"},{"why":"Provides educator and student perspectives on GenAI's impact on assessments, supporting the paper's call for assessment redesign.","marker":"[97]"}],"fun_headline_variants":["Students set their own AI rules amid vague university policies","When universities dodge AI policy, students make their own rules","GenAI in class: students invent unspoken rules to fill policy gap","Students self-govern AI use as universities lag on guidelines"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The central claim rests on the assumption that the self-reported behaviours of 26 students and 11 educators at two UK universities accurately represent what a broader, more diverse student and educator population actually does with generative AI.","fun_headline_variants_meta":{"raw":{"variants":["Students set their own AI rules amid vague university policies","When universities dodge AI policy, students make their own rules","GenAI in class: students invent unspoken rules to fill policy gap","Students self-govern AI use as universities lag on guidelines"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000239,"raw_usage":{"total_tokens":1492,"prompt_tokens":902,"completion_tokens":590,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":518,"completion_tokens_details":{"reasoning_tokens":529}},"tokens_in":518,"tokens_out":590,"duration_ms":5810,"temperature":1.0,"reasoning_tokens":529,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T20:15:02.352728+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A large-scale observational study that logs students' actual GenAI interactions (rather than interviewing them) and compares them with their institution's stated policies could settle the claim: if students in institutions with clear, well-communicated guidelines still develop the same unspoken rules and plagiarism anxiety, the proposed link between unclear guidelines and self-governance would be undermined.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the Strong Structuration Theory framework used to organize the analysis into external structures, internal structures, actions, and outcomes."},{"cited_title":"Students' Voices on Generative AI: Perceptions, Benefits, and Challenges in Higher Education","cited_arxiv_id":"2305.00290","evidence_quote":"Documents students' perceptions, benefits, and challenges of GenAI, giving the paper a comparative anchor for its practice-focused findings."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Offers a qualitative study of ChatGPT's impact that informs the paper's interpretation of plagiarism anxiety and self-governance."},{"cited_title":"Kizilcec","cited_arxiv_id":null,"evidence_quote":"Provides educator and student perspectives on GenAI's impact on assessments, supporting the paper's call for assessment redesign."}],"review_version":1}