{"$schema":"https://policywindow.org/critique/api/schema","critique_id":"CRIT-000021","slug":"postgraduate-students-perceptions-ai-research-plos","url":"https://policywindow.org/critique/c/postgraduate-students-perceptions-ai-research-plos","doi":null,"status":"published","critique_type":"editorially_approved_ai_native_critique","publication_date":"2026-06-28","current_version":"1.0","target_paper":{"title":"Postgraduate students' perceptions of artificial intelligence integration in research: A cross-sectional study","authors":["Ibrahim Naif Alenezi","Fathia Ahmed Mersal","Amal Ahmed Elbilgahy"],"journal":"PLOS One","doi":"10.1371/journal.pone.0345726","url":"https://doi.org/10.1371/journal.pone.0345726","publicationDate":"2026","paperType":"empirical","accessBasis":"open_access","fullTextUsed":true,"fictional":false,"doi_url":"https://doi.org/10.1371/journal.pone.0345726"},"source_journal":{"tier":"exception","rankingSources":["off-monitored: peer-reviewed gold-OA megajournal not in the monitored determination; disclosed off-list"],"rankingNote":"Off-monitored: PLOS One is a peer-reviewed, gold open-access journal (CC BY) not in the journal's monitored top-tier determination; disclosed off-list. Critiqued at full text."},"selection_provenance":{"id":"postgraduate-students-perceptions-ai-research-plos","venue":"PLOS One","inMonitoredSet":false,"determinedTier":null,"recordedTier":"exception","effectiveTier":"exception","kind":"off_list","disclosed":true,"offListPeerReviewed":true},"selection":{"aiAgiCentralityScore":2,"societalRelevanceScore":3,"aiAgiCategories":["human_AI_interaction"],"selectionReason":"Continuous-staged autonomous run: OA full-text critique engaging the coverage blind spots (identification, sample_data, reproducibility); span-grounded to the gold-OA full text."},"scores":{"aiAgiContribution":2,"evidentiarySupport":3,"methodologicalRisk":3,"overclaiming":3,"reproducibilityOrAuditability":2,"societalImpactRelevance":3,"severity":"moderate","confidence":"high"},"severity_cap_for_access_basis":"high","plain_language_summary":"This is a single-university survey of 267 nursing/health master's students in Saudi Arabia about how they perceive using AI in research. It is honestly framed: the authors say upfront it cannot prove cause and effect, admit their volunteer sample skews toward tech-savvy students, and limit their conclusions to this one institution. After cleaning up an over-fired draft critique that was full of details not actually in the paper, three real problems remain. First, the paper interprets a statistical finding — that people more worried about privacy also reported more intention to adopt AI — as evidence of sophisticated \"critical literacy,\" which reads more into a single correlation than it can bear. Second, there's a logical snag: having used AI before was supposedly required to take part, yet \"prior AI use: yes/no\" is used as a predictor and only 85% report prior use, so the entry rule and the data don't line up. Third, only summary tables are shared, so no one can independently re-run the analysis. The descriptive numbers are useful context; the bigger interpretive claims are overstated.","claims":[{"id":"C1","text":"The paper interprets a positive cross-sectional privacy-concerns coefficient as a disposition: \"Privacy concerns appear to reflect critical literacy rather than barriers to adoption.\"","type":"causal","evidenceOffered":"Privacy concerns appear to reflect critical literacy rather than barriers to adoption.","support":"strong","overclaiming":"major","assessment":"A positive cross-sectional association between privacy concerns and adoption intention is recast as a psychological disposition ('critical literacy rather than barriers to adoption'). This is a causal-interpretive leap from a single correlation in a one-time-point, self-report design; given the paper's own reported VIF up to 4.8, a collinearity/suppression artifact is at least as plausible as a substantive trait, yet no such alternative is considered and the framing is asserted in the conclusion as a population-level characteristic.","mainWeakness":"A positive cross-sectional association between privacy concerns and adoption intention is recast as a psychological disposition ('critical literacy rather than barriers to adoption'). This is a causal-interpretive leap from a single correla","confidence":"high"},{"id":"C2","text":"The abstract frames simultaneously-measured self-report subscales causally: perceived benefits are \"the strongest predictor of intention to adopt\" and privacy concerns \"suggest informed and critical engagement.\"","type":"causal","evidenceOffered":"Perceived benefits were the strongest predictor of intention to adopt AI for research purposes (β = 0.588, p < 0.001). Privacy concerns were positively associated with adoption intention (β = 0.230, p < 0.001), suggesting informed and critical engagement rather than resistance.","support":"moderate","overclaiming":"moderate","assessment":"The abstract uses predictor/strength language ('strongest predictor', 'positively associated ... suggesting informed and critical engagement') that imports a directional, mechanistic reading the design cannot support: benefits, privacy concerns and adoption intention are all measured simultaneously by the same respondent on the same instrument, so common-method variance and reverse causation (intending to adopt inflates perceived benefits) are confounded with the reported associations. The Methods section does correctly disclaim causality, so this is an inconsistency between hedged methods and causal-flavored framing rather than a uniform error.","mainWeakness":"The abstract uses predictor/strength language ('strongest predictor', 'positively associated ... suggesting informed and critical engagement') that imports a directional, mechanistic reading the design cannot support: benefits, privacy conc","confidence":"high"},{"id":"C3","text":"Prior AI experience is stated as an inclusion criterion, yet only 85.0% report prior use and \"Prior AI Tool Use (Yes vs. No)\" is entered as a regression predictor.","type":"methodological","evidenceOffered":"who have prior experience using or evaluating AI tools, such as ChatGPT, DeepSeek, Zotero, QuillBot, Grammarly, or Mendeley, for research purposes","support":"strong","overclaiming":"moderate","assessment":"Internal contradiction between eligibility and analysis. Prior experience using or evaluating AI tools is stated as an explicit inclusion criterion, yet the paper reports only 85.0% prior use and enters 'Prior AI Tool Use (Yes vs. No)' as a regression predictor in Table 4. If the inclusion filter were applied, there could be no 'No' category and no ~15% non-users; the coexistence of the filter, the 85% figure, and the Yes/No predictor is logically incoherent and signals either inconsistent screening or mis-specification. (The draft's specific '40 participants (15.0%)' / 'Table 1' figures are not in the text and are not relied upon here.)","mainWeakness":"Internal contradiction between eligibility and analysis. Prior experience using or evaluating AI tools is stated as an explicit inclusion criterion, yet the paper reports only 85.0% prior use and enters 'Prior AI Tool Use (Yes vs. No)' as a","confidence":"high"},{"id":"C4","text":"The paper states \"All relevant data are fully presented within the manuscript itself\" — i.e. only aggregate tables, no analysable dataset.","type":"methodological","evidenceOffered":"All relevant data are fully presented within the manuscript itself, and no additional supporting files were generated.","support":"strong","overclaiming":"minor","assessment":"The analysis is not independently reproducible. The Data Availability Statement declares that all data are within the manuscript and no supporting files were generated, but the manuscript provides only aggregate tables — not the participant- or item-level data needed to reproduce the regression, the HC3 standard errors, the G*Power calculation, or the Cronbach's alpha values. The unresolved prior-AI-use contradiction also cannot be checked without underlying data that is declared not to exist.","mainWeakness":"The analysis is not independently reproducible. The Data Availability Statement declares that all data are within the manuscript and no supporting files were generated, but the manuscript provides only aggregate tables — not the participant","confidence":"high"}],"sections":[{"id":"what","title":"What the paper does","body":"A single-institution cross-sectional survey of 267 nursing/health-profession master's students in Saudi Arabia on perceptions of AI in research, using a TAM-based validated instrument; the authors explicitly disclaim causal inference and name their convenience-sampling bias."},{"id":"overclaim","title":"Interpretive overreach","body":"A positive cross-sectional association between privacy concerns and adoption intention is recast as a psychological disposition ('critical literacy rather than barriers to adoption'). This is a causal-interpretive leap from a single correlation in a one-time-point, self-report design; given the paper's own reported VIF up to 4.8, a collinearity/suppression artifact is at least as plausible as a substantive trait, yet no such alternative is considered and the framing is asserted in the conclusion as a population-level characteristic."},{"id":"identification","title":"Cross-sectional design vs causal-flavoured framing","body":"The abstract uses predictor/strength language ('strongest predictor', 'positively associated ... suggesting informed and critical engagement') that imports a directional, mechanistic reading the design cannot support: benefits, privacy concerns and adoption intention are all measured simultaneously by the same respondent on the same instrument, so common-method variance and reverse causation (intending to adopt inflates perceived benefits) are confounded with the reported associations. The Methods section does correctly disclaim causality, so this is an inconsistency between hedged methods and causal-flavored framing rather than a uniform error."},{"id":"sample","title":"Eligibility-vs-analysis contradiction","body":"Internal contradiction between eligibility and analysis. Prior experience using or evaluating AI tools is stated as an explicit inclusion criterion, yet the paper reports only 85.0% prior use and enters 'Prior AI Tool Use (Yes vs. No)' as a regression predictor in Table 4. If the inclusion filter were applied, there could be no 'No' category and no ~15% non-users; the coexistence of the filter, the 85% figure, and the Yes/No predictor is logically incoherent and signals either inconsistent screening or mis-specification. (The draft's specific '40 participants (15.0%)' / 'Table 1' figures are not in the text and are not relied upon here.)"},{"id":"strengths","title":"What the paper does well","body":"This is a competently executed, unusually self-aware descriptive survey for its genre. The authors explicitly disclaim causal inference (\"this design precludes causal inference and cannot establish temporal precedence\"), name the convenience-sampling bias and its likely direction, restrict generalization to a single regional case study rather than a population, and ground the work in an established framework (TAM) using a previously validated published instrument administered in its original English to an English-instruction health-sciences cohort. The methods exceed many comparable perception studies: an a priori G*Power sample-size calculation, HC3 robust standard errors, reported VIF/outlier diagnostics, recomputed subscale reliabilities, and — notably — a full Table 4 with B, SE, 95% confidence intervals, standardized betas and p-values (so the draft's \"no confidence intervals\" charge is simply wrong). As context-specific, exploratory evidence from an under-studied non-Western health-professional population, which the paper repeatedly and accurately frames it as, the descriptive findings are informative and modestly stated."}],"strongest_critique":"The flagship interpretive claim overreaches the design. A positive cross-sectional coefficient for privacy concerns is recast as a psychological disposition (\"Privacy concerns appear to reflect critical literacy rather than barriers to adoption\"), and benefits are labeled \"the strongest predictor of intention to adopt,\" even though every subscale and the outcome are measured at one time point by the same respondent on the same instrument — so common-method variance and reverse causation are fully confounded with the reported associations, and a counterintuitive positive coefficient is at least as plausibly a collinearity artifact (the paper itself reports VIF up to 4.8) as a substantive trait. This is compounded by an unresolved data-integrity contradiction: prior AI experience is an explicit inclusion criterion, yet \"prior AI use\" is entered as a Yes-vs-No predictor and only 85% report prior use — logically incoherent if the filter was applied — and it cannot be checked because the declared dataset is only the manuscript's aggregate tables.","strongest_fair_defence":"This is a competently executed, unusually self-aware descriptive survey for its genre. The authors explicitly disclaim causal inference (\"this design precludes causal inference and cannot establish temporal precedence\"), name the convenience-sampling bias and its likely direction, restrict generalization to a single regional case study rather than a population, and ground the work in an established framework (TAM) using a previously validated published instrument administered in its original English to an English-instruction health-sciences cohort. The methods exceed many comparable perception studies: an a priori G*Power sample-size calculation, HC3 robust standard errors, reported VIF/outlier diagnostics, recomputed subscale reliabilities, and — notably — a full Table 4 with B, SE, 95% confidence intervals, standardized betas and p-values (so the draft's \"no confidence intervals\" charge is simply wrong). As context-specific, exploratory evidence from an under-studied non-Western health-professional population, which the paper repeatedly and accurately frames it as, the descriptive findings are informative and modestly stated.","final_judgment":"REFUTE (with credit), but on a much narrower and better-grounded basis than the draft. The paper is a competent, self-aware single-institution descriptive survey that explicitly disclaims causal inference, names its convenience-sampling bias and likely direction, and restricts its claims to a context-specific case study. Three genuine, exactly span-grounded flaws nonetheless weaken its inferential and evidentiary core: (1) the abstract/conclusion recast a positive cross-sectional privacy-concerns coefficient as a psychological disposition (\"critical literacy rather than barriers to adoption\") and label simultaneously-measured self-report subscales as \"the strongest predictor,\" language stronger than a single-time-point common-method design supports; (2) an unresolved internal contradiction — prior AI experience is an explicit eligibility requirement, yet prior AI use appears as a Yes-vs-No regression predictor and only 85% report prior use, so either ineligible respondents were included or the filter was not applied; and (3) the analysis is not independently reproducible because the Data Availability Statement declares only aggregate manuscript tables exist. These are real but bounded; treat the descriptive prevalence findings as suggestive context-specific evidence and the interpretive \"critical literacy\"/\"pragmatic optimism\" framings as overclaimed. IMPORTANT: the draft critique was heavily contaminated with material absent from the full text (a Model 1/Model 2 split, adjusted R²=0.993, \"methodological circularity,\" literal \"40 participants (15.0%)\"/\"Table 1,\" subscale r up to .71, a 97.4%/227 ChatGPT denominator) and with at least one false claim (\"no confidence intervals reported anywhere\" — Table 4 includes a 95% CI column). Those points were dropped as ungrounded.","review_process":{"aiAgentsUsed":["claim_extraction","methods","statistics","reproducibility","overclaiming","adversarial","author_defence","plain_language","meta_review"],"reviewRounds":2,"humanEditor":{"name":"","role":"","approvalDate":"2026-06-28","declaredConflict":"none"},"expertCertification":{"used":false}},"author_response":{"notified":false,"status":"not_yet_invited","editorialActionAfterResponse":"Authors may reply at any time; this critique addresses claims, methods and inference only, never the authors."},"versions":[{"version":"1.0","date":"2026-06-28","note":"Initial publication (promoted from the continuous-staged queue; G92).","changeType":"initial"}],"transparency":{"modelCardUrl":"/critique/model-card","publicAuditSummary":"Full-text critique of a gold-OA paper (CC BY). Over-fired 8-dimension draft sharpened to 4 defensible span-exact flaws; over-reach holds, faithful, all spans exact, independently re-verified against the full text; metadata Crossref-verified. STAGED, not published.","privateAuditRecordExists":true,"citationVerification":{"status":"complete","checkedSources":[{"label":"DOI 10.1371/journal.pone.0345726 (Crossref: title+author+year matched)","url":"https://doi.org/10.1371/journal.pone.0345726","verified":true}],"fabricatedCitations":0},"riskReview":{"copyright":"completed","defamation":"completed","note":"Gold OA (CC BY) quoted sparingly under criticism/review; targets claims/methods/inference only."}}}