Between Co-Intelligence and Cognitive Offloading: Educators’ Perspectives from Dutch Academic Medical Settings on Generative AI, Critical Thinking, and Adult Professional Learning

May combine two claims.

Towards the 15th interview, no substantively new themes relevant to the research questions were identified, and informational/thematic saturation was judged to have been reached within the scope of the study.

“Towards the 15th interview, no substantively new themes relevant to the research questions were identified, and the researchers therefore judged that informational/thematic saturation had been reached within the scope of the study (Saunders et al., 2018).”

• cites Saunders et al., 2018 sourcePartly supported

  The ‘What?’ section defines data saturation in terms of informational redundancy, and ‘Perspectives taken on saturation’ discusses judging whether interviews yield new codes or themes (page numbers unavailable). These passages support the rationale for a saturation judgment, but they do not establish that the citing researchers found no substantively new themes by their 15th interview. (LLM, conf=0.86)

  Experimental — accuracy not yet measured

Classified as probably not a substantial claim. May combine two claims.

In medicine, AI literacy is required: learners need sufficient technical, ethical, legal, and disciplinary knowledge to evaluate GenAI output rather than merely consume it.

“In medicine, this creates a strong requirement for AI literacy: learners need sufficient technical, ethical, legal, and disciplinary knowledge to evaluate output rather than merely consume it (Çalışkan et al., 2022; Gazquez-Garcia et al., 2025; Lintner, 2024).”

• cites Çalışkan et al., 2022Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites Gazquez-Garcia et al., 2025Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites Lintner, 2024 sourcePartly supported

  The Introduction defines AI literacy as the ability to critically evaluate AI outputs, and the Results (Overview of AI literacy scales) identify technical understanding and ethics as core competencies, including for medical professionals. The cited passages do not establish legal or discipline-specific knowledge as requirements, or make the full claim specifically about medical learners evaluating GenAI output. (LLM, conf=0.86)

  Experimental — accuracy not yet measured

• AssumptionRests on an assumption: GenAI systems produce plausible but inaccurate information and reproduce bias in their training data.
• AssumptionRests on an assumption: GenAI systems operate through processes that are difficult for users to inspect.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

TAM has been extensively applied in educational technology research and increasingly in medical education.

“TAM has been extensively applied in educational technology research and increasingly in medical education (Granić & Marangunić, 2019; Lee et al., 2025).”

• cites Granić & Marangunić, 2019 sourceNot checked — no full text of the source
• cites Lee et al., 2025 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Individual semi-structured interviews enabled consistent exploration of common topics while allowing follow-up questions when participants introduced unexpected examples or distinctions.

“Individual semi-structured interviews enabled consistent exploration of common topics while allowing follow-up questions when participants introduced unexpected examples or distinctions (DeJonckheere & Vaughn, 2019).”

• cites DeJonckheere & Vaughn, 2019 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

In medical education, students can use large language models to generate explanations and practice questions, summarise dense materials, improve academic prose, support literature searches, analyse data, and rehearse clinical scenarios.

“In medical education, students can use large language models (LLMs) to generate explanations and practice questions, summarise dense materials, improve academic prose, support literature searches, analyse data, and rehearse clinical scenarios (Pham et al., 2025; Cervantes et al., 2024; Izquierdo-Condoy et al., 2025; Dong et al., 2025).”

• cites Pham et al., 2025 sourceSupported
• cites Cervantes et al., 2024 sourceNot checked — no full text of the source
• cites Izquierdo-Condoy et al., 2025Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Classified as probably not a substantial claim. May combine two claims.

Adult learning scholarship in medical education emphasises the growing role of learner agency, experience, application to authentic problems, and the shared responsibility of learners, educators, and institutions for creating conditions in which meaningful learning can occur.

“Adult learning scholarship in medical education emphasises the growing role of learner agency, experience, application to authentic problems, and the shared responsibility of learners, educators, and institutions for creating conditions in which meaningful learning can occur (Taylor & Hamdy, 2013; Gupta et al., 2026).”

• cites Taylor & Hamdy, 2013 sourceSupported
• cites Gupta et al., 2026 sourcePartly supported

  Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  Experimental — accuracy not yet measured

Classified as probably not a substantial claim.

Recent literature associates frequent or uncritical AI use with the risk that students will rely on algorithmic answers in place of source evaluation, reasoning, and independent problem solving.

“Recent literature associates frequent or uncritical AI use with the risk that students will rely on algorithmic answers in place of source evaluation, reasoning, and independent problem solving (Gerlich, 2025; Izquierdo-Condoy et al., 2025; Zhai et al., 2024).”

• cites Gerlich, 2025 sourceSupported
• cites Izquierdo-Condoy et al., 2025Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites Zhai et al., 2024 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

The medical education literature describes substantial potential benefits from GenAI, including 24/7 tutoring, personalised feedback, test generation, writing assistance, and support for self-regulated learning.

“The medical education literature describes substantial potential benefits from GenAI, including 24/7 tutoring, personalised feedback, test generation, writing assistance, and support for self-regulated learning (Mulaudzi & Hamilton, 2025; Sardi et al., 2025).”

• cites Mulaudzi & Hamilton, 2025 sourcePartly supported

  The Results and Discussion section describes 24/7 support, tailored learning paths, feedback, and help with essay writing. But the paper addresses higher education generally, not specifically medical education; the passages do not substantiate test generation or support for self-regulated learning. (LLM, conf=0.87)

  Experimental — accuracy not yet measured
• cites Sardi et al., 2025 sourcePartly supported

  The Introduction and Discussion describe instant guidance, personalised feedback, and support for students’ self-regulated learning. The paper addresses education generally, not medical education, and the passages found do not establish 24/7 tutoring, test generation, or writing assistance as benefits discussed by this review. (LLM, conf=0.87)

  Experimental — accuracy not yet measured

May combine two claims.

Plausible AI-generated misinformation can reduce diagnostic accuracy in novice medical learners while leaving them poorly calibrated in their confidence.

“More recent evidence from novice medical learners similarly indicates that plausible AI-generated misinformation can reduce diagnostic accuracy while leaving learners poorly calibrated in their confidence (Teng et al., 2026).”

• cites Teng et al., 2026 sourceSupported

Classified as probably not a substantial claim.

The findings reinforce arguments that institutions should redesign assessment rather than assume that plagiarism-detection logic can simply be extended to GenAI.

“The findings therefore reinforce arguments that institutions should redesign assessment rather than assume that plagiarism-detection logic can simply be extended to GenAI (Ardito, 2025; Boscardin et al., 2024).”

• cites Ardito, 2025 sourceSupported
• cites Boscardin et al., 2024 sourceNot checked — no full text of the source

May combine two claims.

Educators associated low-friction GenAI use with cognitive offloading, weaker source appraisal, reduced information filtering, possible erosion of foundational knowledge, and difficulties verifying authentic student understanding.

“However, they associated low-friction use with cognitive offloading, weaker source appraisal, reduced information filtering, possible erosion of foundational knowledge, and difficulties verifying authentic student understanding.”

Classified as probably not a substantial claim.

Three interconnected themes captured educators' accounts.

“Three interconnected themes captured educators' accounts.”

May combine two claims.

Three themes were identified in the educators' accounts: educational value and efficiency; threats to independent critical thinking; and conditions for responsible integration.

“Three themes were identified: educational value and efficiency; threats to independent critical thinking; and conditions for responsible integration.”

• Also stated in: § 2 . 4 . Transcription and Analysis · p. 6 · plain

  “These were organised into three overarching themes: (1) perceived educational value in the adoption of GenAI; (2) threats to independent critical thinking; and (3) conditions for responsible use of GenAI in the participating medical-education settings.”

May combine two claims.

Educators reported that students use GenAI to rephrase text, improve academic register, correct language, and structure references.

“Educators reported that students use GenAI to rephrase text, improve academic register, correct language, and structure references.”

May combine two claims.

Educators valued GenAI for personalised learning, writing support, research assistance, accessibility, simulation, and administrative efficiency.

“Educators valued GenAI for personalised learning, writing support, research assistance, accessibility, simulation, and administrative efficiency.”

Classified as probably not a substantial claim.

Educators argued that curriculum design must ensure that the convenience of digital assistance does not displace experiences through which tacit judgement, interpersonal competence, and professional identity are developed.

“Educators did not reject GenAI because it could not reproduce every aspect of expert practice; instead, they argued that curriculum design must ensure that the convenience of digital assistance does not displace experiences through which tacit judgement, interpersonal competence, and professional identity are developed.”

Classified as probably not a substantial claim. May combine two claims.

Participants proposed AI literacy, transparent use, institutionally governed models, comparative human-AI tasks, simulation, staged access, and AI-free oral or in-class assessment.

“Participants proposed AI literacy, transparent use, institutionally governed models, comparative human-AI tasks, simulation, staged access, and AI-free oral or in-class assessment.”

May combine two claims.

Educators themselves reported uncertainty and a sense that curriculum development was lagging behind technological change.

“Educators themselves reported uncertainty and a sense that curriculum development was lagging behind technological change.”

Classified as probably not a substantial claim. May combine two claims.

Participants proposed oral examinations, presentations with questioning, face-to-face feedback, closed-laptop clinical cases, and in-class group problem solving as AI-free assessment formats.

“Participants proposed oral examinations, presentations with questioning, face-to-face feedback, closed-laptop clinical cases, and in-class group problem solving.”

Classified as probably not a substantial claim. May combine two claims.

Personalised explanations, alternative modalities for learners with dyslexia, language support, and always-available practice can reduce access barriers.

“Personalised explanations, alternative modalities for learners with dyslexia, language support, and always-available practice can reduce access barriers.”

May combine two claims.

Across these proposals, the unifying principle was not prohibition; participants wanted learners to become capable of working with GenAI while remaining able to reason, explain, and act without it when professional situations require.

“Across these proposals, the unifying principle was not prohibition. Participants wanted learners to become capable of working with GenAI while remaining able to reason, explain, and act without it when professional situations require.”

May combine two claims.

Participants linked foundational medical knowledge to safe AI use, arguing that learners need domain knowledge to identify when an output is implausible, incomplete, or contextually inappropriate.

“Participants also linked foundational medical knowledge to safe AI use. A future clinician may not have time to consult a chatbot during every decision, but more importantly, learners need domain knowledge in order to identify when an output is implausible, incomplete, or contextually inappropriate.”

• AssumptionRests on an assumption: A future clinician may not have time to consult a chatbot during every decision.

Classified as probably not a substantial claim.

A related institutional problem was the difficulty of determining how much of a take-home written submission represents the student's unaided work.

“A related institutional problem was the difficulty of determining how much of a takehome written submission represents the student's unaided work.”

Classified as probably not a substantial claim.

Participants stressed that AI literacy is not a replacement for medical knowledge.

“They also stressed that AI literacy is not a replacement for medical knowledge.”

Classified as probably not a substantial claim.

Participants argued for shared institutional rules and educator development rather than leaving each lecturer to improvise.

“Participants therefore argued for shared institutional rules and educator development rather than leaving each lecturer to improvise.”

May combine two claims.

Participants identified an efficiency paradox alongside concerns about factual inaccuracy.

“Participants identified an efficiency paradox alongside concerns about factual inaccuracy.”

• AssumptionRests on an assumption: GenAI lowers the time and effort needed to produce plausible academic work.

May combine two claims.

Participants described AI literacy needs for both students and staff, including understanding how LLMs generate output, why hallucinations and bias occur, how to verify references, how data may be processed, and how prompt quality shapes responses.

“Participants described AI literacy needs for both students and staff, including understanding how LLMs generate output, why hallucinations and bias occur, how to verify references, how data may be processed, and how prompt quality shapes responses.”

Probably not a substantial claim: may not pass S3 (load-bearing).

P13 cautioned that AI-assisted rephrasing can alter the substantive content of an argument without novice writers necessarily recognising the shift.

“P13 cautioned that AI-assisted rephrasing can alter the substantive content of an argument without novice writers necessarily recognising the shift.”

Classified as probably not a substantial claim.

Participants raised questions about whether conventional literature reviews, reflection reports, and other take-home written tasks remain valid as stand-alone measures of individual learning.

“This raised questions about whether conventional literature reviews, reflection reports, and other take-home written tasks remain valid as stand-alone measures of individual learning.”

Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing).

One position favoured staged or incremental access, with more AI-free zones early in the curriculum while foundational knowledge and reasoning are established, followed by broader use later as students prepare for AI-enabled professional practice.

“One position favoured staged or incremental access, with more AI-free zones early in the curriculum while foundational knowledge and reasoning are established, followed by broader use later as students prepare for AI-enabled professional practice.”

May combine two claims.

Participants did not uniformly accept a narrative of inevitable cognitive decline; a contrasting perspective was that the effect depends strongly on how GenAI is used, with controlled examinations continuing to provide evidence of baseline knowledge.

“However, they did not uniformly accept a narrative of inevitable cognitive decline. A contrasting perspective was that the effect depends strongly on how GenAI is used, with controlled examinations continuing to provide evidence of baseline knowledge.”

Probably not a substantial claim: may not pass S3 (load-bearing).

One participant described the premium-model ecosystem as 'pay-to-win', because students able to purchase better access may receive an academic advantage.

“One participant described the premium-model ecosystem as 'pay-to-win', because students able to purchase better access may receive an academic advantage.”

Classified as probably not a substantial claim.

The argument was that some forms of effort perform a learning function: selecting relevant evidence, constructing an explanation, testing a diagnostic hypothesis, or persisting through uncertainty.

“The argument was not that effort is valuable in itself, but that some forms of effort perform a learning function: selecting relevant evidence, constructing an explanation, testing a diagnostic hypothesis, or persisting through uncertainty.”

May combine two claims.

Core content identified by participants includes model limitations, hallucinations, source verification, prompt design, privacy and copyright, bias, equity, and the relationship between AI output and disciplinary evidence.

“Core content identified by participants includes model limitations, hallucinations, source verification, prompt design, privacy and copyright, bias, equity, and the relationship between AI output and disciplinary evidence.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Two other participants (P11 and P13) challenged the separation of writing proficiency from critical thinking.

“Two others (P11 and P13) challenged this separation.”

May combine two claims.

Participants identified inequities created by paid premium models and privacy risks when learners enter sensitive data into commercial systems.

“Participants also identified inequities created by paid premium models, alongside privacy risks when learners enter sensitive data into commercial systems.”

Probably not a substantial claim: may not pass S3 (load-bearing).

P11 emphasised that converting ideas into a logically coherent text is itself an important student capability.

“P11 emphasised that converting ideas into a logically coherent text is itself an important student capability”

Classified as probably not a substantial claim.

AI literacy emerged as a curriculum-level recommendation.

“AI literacy emerged as a curriculum-level recommendation.”

Classified as probably not a substantial claim. May combine two claims.

Examples of AI-augmented learning activities included comparing a student-written text with an AI-generated version, identifying errors and biases in both human and AI reports, using LLMs as Socratic tutors that reveal information step by step, and simulating patient consultations or multidisciplinary meetings.

“Examples included comparing a student-written text with an AI-generated version, identifying errors and biases in both human and AI reports, using LLMs as Socratic tutors that reveal information step by step, and simulating patient consultations or multidisciplinary meetings.”

Classified as probably not a substantial claim. May combine two claims.

Participants proposed two complementary categories of learning and assessment: one that deliberately uses GenAI to stimulate critical engagement, and one that deliberately removes GenAI to verify independent competence.

“Participants proposed two complementary categories of learning and assessment. The first deliberately uses GenAI to stimulate critical engagement.”

• Also stated in: § 3.3.3. A Dual-Mode Assessment Architecture · p. 14 · plain

  “The second category deliberately removes GenAI in order to verify independent competence.”

May combine two claims.

Participants' concerns centred on cognitive offloading, weaker source evaluation and information filtering, possible erosion of foundational knowledge and ideation, and the declining interpretability of take-home written work as evidence of individual competence.

“Concerns centred on cognitive offloading, weaker source evaluation and information filtering, possible erosion of foundational knowledge and ideation, and the declining interpretability of take-home written work as evidence of individual competence.”

Classified as probably not a substantial claim.

Participants reported seeing more uniform phrasing and argumentation in student work, raising concerns about reduced ideational diversity.

“Participants also reported seeing more uniform phrasing and argumentation in student work, raising concerns about reduced ideational diversity.”

May combine two claims.

Participants described stylistic cues but acknowledged that they could rarely prove GenAI use with sufficient certainty for disciplinary action.

“Participants described stylistic cues but acknowledged that they could rarely prove GenAI use with sufficient certainty for disciplinary action.”

Classified as probably not a substantial claim.

The perceived solution was not more aggressive detection, but redesign of assessment around demonstrable learning outcomes.

“The perceived solution was not more aggressive detection, but redesign of assessment around demonstrable learning outcomes.”

May combine two claims.

GenAI was valued by participants for writing support, personalised study, accessibility, research assistance, simulation, and educator efficiency.

“GenAI was valued for writing support, personalised study, accessibility, research assistance, simulation, and educator efficiency.”

Classified as probably not a substantial claim. May combine two claims.

The perceived consequences of cognitive offloading included weaker ability to distinguish central from peripheral information, poorer source literacy, lower retention, and reduced readiness to reason without external assistance.

“The perceived consequences included weaker ability to distinguish central from peripheral information, poorer source literacy, lower retention, and reduced readiness to reason without external assistance.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

University-owned or institutionally governed LLM environments were proposed as a way to improve privacy, control data use, reduce reliance on commercial accounts, and provide a common platform for pedagogical experimentation.

“University-owned or institutionally governed LLM environments were proposed as a way to improve privacy, control data use, reduce reliance on commercial accounts, and provide a common platform for pedagogical experimentation.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Participants highlighted privacy, copyright, algorithmic bias, and unequal access to paid models as concerns.

“They also highlighted privacy, copyright, algorithmic bias, and unequal access to paid models.”

Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing). May combine two claims.

A contrasting position emphasised that early-stage learners can also benefit from personalised GenAI tutoring, suggesting that the appropriate sequence should depend not on academic year alone but on the learning objective, the cognitive activity being delegated, and the learner's demonstrated capacity to evaluate the resulting output.

“A contrasting position emphasised that early-stage learners can also benefit from personalised GenAI tutoring, suggesting that the appropriate sequence should depend not on academic year alone but on the learning objective, the cognitive activity being delegated, and the learner's demonstrated capacity to evaluate the resulting output.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Participant accounts differed regarding the sequencing of GenAI access.

“Participant accounts differed regarding the sequencing of GenAI access.”

May combine two claims.

Participants described fragmented or unclear policies, including uncertainty among educators about what uses should be permitted in particular assignments.

“Participants also described fragmented or unclear policies, including uncertainty among educators about what uses should be permitted in particular assignments.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Two participants (P12 and P6) treated writing proficiency as at least partly separable from critical thinking, arguing that GenAI-assisted rephrasing could be acceptable when the underlying ideas, source selection, and conclusions remained the student's own.

“Two participants (P12 and P6) treated writing proficiency as at least partly separable from critical thinking, arguing that GenAI-assisted rephrasing could be acceptable when the underlying ideas, source selection, and conclusions remained the student's own.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The TAM and Paul-Elder frameworks were used as sensitising rather than exhaustive analytical structures.

“The frameworks were therefore used as sensitising rather than exhaustive analytical structures.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The interview guide was explicitly informed by TAM and the Paul-Elder framework while retaining open questions about practices, concerns, and possible curriculum responses.

“The interview guide (Supplementary Materials S1) was explicitly informed by TAM and the Paul-Elder framework while retaining open questions about practices, concerns, and possible curriculum responses.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Recent work on ethical GenAI integration in higher education identifies academic integrity, privacy and data security, transparency and accountability, equity and access, and AI literacy as interdependent governance challenges rather than isolated technical problems.

“Recent work on ethical GenAI integration in higher education similarly identifies academic integrity, privacy and data security, transparency and accountability, equity and access, and AI literacy as interdependent governance challenges rather than isolated technical problems (Yarboi et al., 2026).”

• cites Yarboi et al., 2026 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

The second researcher conducted an independent analytic review of the coding framework, coded excerpts, and emerging thematic structure, referring to source transcripts where clarification was required.

“The second researcher then conducted an independent analytic review of the coding framework, coded excerpts, and emerging thematic structure, with reference to the source transcripts where clarification was required.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Neither researcher approached the analysis from a position of advocacy for or opposition to GenAI; both treated its educational value as contingent on the learning objective, context of use, and extent of learner cognitive engagement.

“Neither researcher approached the analysis from a position of either advocacy for or opposition to GenAI; both treated its educational value as contingent on the learning objective, context of use, and extent of learner cognitive engagement.”

Probably not a substantial claim: may not pass S4 (based).

Medical education is an especially informative professional-learning context because the consequences of weak knowledge calibration, poor source evaluation, or overconfidence extend beyond academic performance to future clinical reasoning and patient-facing practice.

“Medical education provides an especially informative professional-learning context because the consequences of weak knowledge calibration, poor source evaluation, or overconfidence extend beyond academic performance to future clinical reasoning and patient-facing practice.”

May combine two claims.

The same low-friction characteristics that made GenAI easy and efficient to use could enhance task completion while reducing educational value when they displaced cognitive activity that participants regarded as integral to the learning objective.

“The same low-friction characteristics that made GenAI easy and efficient to use could enhance task completion while reducing educational value when they displaced cognitive activity that participants regarded as integral to the learning objective.”

Classified as probably not a substantial claim. May combine two claims.

Recent scholarship on GenAI in higher and adult education frames the challenge as one of learning design rather than simple adoption, requiring institutions to combine human agency, AI literacy, assessment validity, transparency, privacy, and equity while preserving the developmental work learners are expected to perform.

“Recent scholarship on GenAI in higher and adult education increasingly frames the challenge as one of learning design rather than simple adoption: institutions must combine human agency, AI literacy, assessment validity, transparency, privacy, and equity while preserving the developmental work learners are expected to perform (Yarboi et al., 2026).”

• cites Yarboi et al., 2026 sourceNot checked — no full text of the source

May combine two claims.

The same GenAI system can function as a cognitive partner when it prompts learners to articulate, compare, verify, and revise, or as a cognitive substitute when it supplies a finished output before the learner has engaged with the task.

“In other words, the same system can function as a cognitive partner when it prompts learners to articulate, compare, verify, and revise, or as a cognitive substitute when it supplies a finished output before the learner has engaged with the task.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The researchers recognised that qualitative analysis is interpretive and that their disciplinary backgrounds, prior experiences, and perspectives on Generative AI could influence both engagement with the data and the development of themes.

“The researchers recognised that qualitative analysis is interpretive and that their disciplinary backgrounds, prior experiences, and perspectives on Generative AI could influence both engagement with the data and the development of themes.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The study did not employ independent double-coding of the complete dataset, and no formal interrater coefficient was calculated.

“The study did not employ independent double-coding of the complete dataset, and no formal interrater coefficient was calculated.”

May combine two claims.

An exploratory qualitative design was used because the study addressed an emerging and context-dependent educational phenomenon and sought to understand the reasoning underlying educators' decisions rather than estimate prevalence.

“An exploratory qualitative design was used because the study addressed an emerging and context-dependent educational phenomenon and sought to understand the reasoning underlying educators' decisions rather than estimate prevalence (Hunter et al., 2019).”

• cites Hunter et al., 2019Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The second researcher had an intermediate level of proficiency in Dutch (B1) and was not treated as an independent validator of Dutch-English semantic equivalence.

“The second researcher, with an intermediate level of proficiency in Dutch (B1), contributed to review of the English-language presentation but was not treated as an independent validator of Dutch-English semantic equivalence.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Translation was undertaken only at the stage of preparing quotations for English-language publication, while coding and primary interpretation were conducted using the original-language transcripts.

“Translation was undertaken only at the stage of preparing quotations for English-language publication; coding and primary interpretation were conducted using the original-language transcripts.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

GenAI is notable because both TAM constructs are unusually salient: the technology is useful across many tasks and is exceptionally easy to access.

“GenAI is notable because both constructs are unusually salient: the technology is useful across many tasks and is exceptionally easy to access.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Analytic discussions considered discrepant and contrasting accounts rather than seeking only evidence that reinforced the dominant thematic interpretation.

“Analytic discussions also considered discrepant and contrasting accounts rather than seeking only evidence that reinforced the dominant thematic interpretation.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

For professional education, institutions may need to move from a predominantly detection-based response towards assessment architectures that both teach responsible AI use and preserve opportunities to verify human competence.

“For professional education, the implication is that institutions may need to move from a predominantly detection-based response towards assessment architectures that both teach responsible AI use and preserve opportunities to verify human competence (Sikos, 2025).”

• cites Sikos, 2025 sourceNot checked — no full text of the source

Classified as probably not a substantial claim.

Paul and Elder's critical thinking framework conceptualises critical thinking through intellectual standards and traits, including clarity, relevance, significance, accuracy, depth, perseverance, and intellectual autonomy.

“Second, Paul and Elder's critical thinking framework conceptualises critical thinking through intellectual standards and traits, including clarity, relevance, significance, accuracy, depth, perseverance, and intellectual autonomy (Paul & Elder, 1999; Elder, 2016).”

• cites Paul & Elder, 1999Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites Elder, 2016 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Inductive codes were retained where participants introduced experiences, concerns, or interpretations not adequately represented by the predefined theoretical concepts, and the developing themes were repeatedly checked against the underlying coded data.

“Inductive codes were retained where participants introduced experiences, concerns, or interpretations that were not adequately represented by the predefined theoretical concepts, and the developing themes were repeatedly checked against the underlying coded data.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

Adult professional education should design GenAI as co-intelligence rather than cognitive substitution, preserving productive friction while verifying unassisted competence.

“The findings suggest that adult professional education should design GenAI as co-intelligence rather than cognitive substitution, preserving productive friction while verifying unassisted competence.”

Classified as probably not a substantial claim.

The Technology Acceptance Model explains technology adoption through a set of relationships involving perceived usefulness, perceived ease of use, attitudes or behavioural intention towards use, and subsequent technology use.

“First, the Technology Acceptance Model (TAM) explains technology adoption through a set of relationships involving perceived usefulness (PU), perceived ease of use (PEOU), attitudes or behavioural intention towards use, and subsequent technology use (Davis, 1989).”

• cites Davis, 1989 sourceNot checked — no full text of the source

May combine two claims.

Ease of use may increase adoption of GenAI while simultaneously reducing its pedagogical usefulness if it encourages students to bypass the learning process.

“Ease of use may increase adoption while simultaneously reducing pedagogical usefulness if it encourages students to bypass the learning process.”

May combine two claims.

GenAI affordances are relevant to adult learners because they can support flexible, just-in-time learning and reduce barriers associated with language, disability, or limited access to individual tutoring.

“These affordances are relevant to adult learners because they can support flexible, just-in-time learning and reduce barriers associated with language, disability, or limited access to individual tutoring.”

Classified as probably not a substantial claim. May combine two claims.

In the original TAM, perceived ease of use contributes to perceived usefulness, and these perceptions inform attitudes and behavioural intentions towards technology use.

“In the original TAM, PEOU also contributes to PU, while these perceptions inform attitudes and behavioural intentions towards technology use.”

• cites Davis, 1989 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The first researcher, a native Dutch speaker who conducted the interviews and primary analysis, rechecked all Dutch quotations selected for publication against the original verified transcripts and revised the English translations where necessary to preserve their semantic and contextual meaning.

“The first researcher, a native Dutch speaker who conducted the interviews and primary analysis, rechecked all Dutch quotations selected for publication against the original verified transcripts and revised the English translations where necessary to preserve their semantic and contextual meaning.”

Probably not a substantial claim: may not pass S3 (load-bearing).

For the revised manuscript, all Dutch quotations retained in the results were reviewed by an independent bilingual, native Dutch speaker with fluent English, who compared each English quotation directly with the corresponding Dutch source text.

“For the revised manuscript, all Dutch quotations retained in the results were additionally reviewed by an independent bilingual, native Dutch speaker with fluent English, who compared each English quotation directly with the corresponding Dutch source text.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Educators can use GenAI tools to develop teaching materials, translate or adapt cases, and support feedback workflows.

“Educators can likewise use these tools to develop teaching materials, translate or adapt cases, and support feedback workflows (Abd-alrazaq et al., 2023; Boscardin et al., 2024; Hale et al., 2024; Preiksaitis & Rose, 2023).”

• cites Abd-alrazaq et al., 2023Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites Boscardin et al., 2024 sourceNot checked — no full text of the source
• cites Hale et al., 2024 sourceNot checked — no full text of the source

May combine two claims.

GenAI may be educationally valuable when it removes incidental barriers or provides additional perspectives and feedback, but problematic when it removes the epistemic work that learners are expected to develop.

“GenAI may therefore be educationally valuable when it removes incidental barriers or provides additional perspectives and feedback, but problematic when it removes the epistemic work that learners are expected to develop.”

Classified as probably not a substantial claim.

AI-augmented assessment can evaluate the learner's capacity to prompt, critique, compare, verify, and improve AI output.

“AI-augmented assessment can evaluate the learner's capacity to prompt, critique, compare, verify, and improve AI output.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

As expertise develops, scaffolding can progressively decrease and learners can be expected to identify missing evidence, triangulate claims against authoritative sources, justify why an output should or should not be trusted, and document how verification affected their judgement.

“As expertise develops, scaffolding can progressively decrease and learners can be expected to identify missing evidence, triangulate claims against authoritative sources, justify why an output should or should not be trusted, and document how verification affected their judgement.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

A second researcher conducted an independent analytic review of the coding framework, coded excerpts, and developed thematic structure, and participated in resolving interpretive differences.

“A second researcher subsequently conducted an independent analytic review of the coding framework, coded excerpts, and developed the thematic structure and participated in resolving interpretive differences.”

Probably not a substantial claim: may not pass S2 (checkable).

The proposed principles (human first-pass work, visible verification, dialogic use, unassisted competence checks, and institutional governance) can be treated as hypotheses for design across adult learning ecosystems rather than medicine-specific prescriptions.

“The proposed principles, including human first-pass work, visible verification, dialogic use, unassisted competence checks, and institutional governance, can therefore be treated as hypotheses for design across adult learning ecosystems rather than medicine-specific prescriptions.”

Probably not a substantial claim: may not pass S3 (load-bearing).

These activities make the quality of human-AI interaction itself an object of learning.

“These activities make the quality of human-AI interaction itself an object of learning.”

May combine two claims.

The study did not employ independent double-coding of the full dataset or formal interrater-reliability testing.

“Although this process provided an additional perspective through which to challenge and refine the analysis, the study did not employ independent double-coding of the full dataset or formal interrater-reliability testing.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Following verification of the interview transcripts and completion of the analysis, the original audio recordings were permanently deleted to minimise the continued retention of identifiable participant data.

“Following verification of the interview transcripts and completion of the analysis, the original audio recordings were permanently deleted to minimise the continued retention of identifiable participant data.”

Classified as probably not a substantial claim.

If the learning objective is rapid clinical reasoning or oral explanation, live or closed-device assessment may provide stronger validity evidence.

“If the learning objective is rapid clinical reasoning or oral explanation, live or closed-device assessment may provide stronger validity evidence.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The original research document, which predates the linguistic intervention by ChatGPT, was submitted to Erasmus University's TMS document repository on 19 June 2026.

“For transparency, the original research document, which predates the linguistic intervention by ChatGPT (GPT-5.6 Sol), was submitted to Erasmus University's TMS document repository on 19 June 2026.”

Probably not a substantial claim: may not pass S2 (checkable).

GenAI-supported learning activities should be sequenced according to learners' demonstrated capacity to critically evaluate the particular forms of output they are expected to use.

“This suggests that GenAI-supported learning activities should be sequenced according to learners' demonstrated capacity to critically evaluate the particular forms of output they are expected to use.”

Probably not a substantial claim: may not pass S2 (checkable).

Earlier-stage learners may require bounded verification activities using curated cases, trusted disciplinary sources, worked examples of hallucinations or misleading reasoning, and educator-led debriefing.

“Earlier-stage learners may require bounded verification activities using curated cases, trusted disciplinary sources, worked examples of hallucinations or misleading reasoning, and educator-led debriefing.”

Classified as probably not a substantial claim. May combine two claims.

From this perspective, turning bullet points into polished prose with GenAI may conceal whether a student can independently communicate a coherent argument and may subtly alter the ideas being expressed.

“From this perspective, turning bullet points into polished prose with GenAI may conceal whether a student can independently communicate a coherent argument and may subtly alter the ideas being expressed.”

May combine two claims.

This establishes a circular challenge: GenAI can help learners acquire knowledge, but over-reliance on it may undermine the knowledge needed to use GenAI critically.

“This establishes a circular challenge: GenAI can help learners acquire knowledge, but over-reliance on it may undermine the knowledge needed to use GenAI critically.”

Probably not a substantial claim: may not pass S2 (checkable).

GenAI should be positioned as co-intelligence rather than cognitive substitution.

“GenAI should be positioned as co-intelligence rather than cognitive substitution.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

ChatGPT was not used as a source of evidentiary validation, and the authors remain responsible for all final translations, interpretations, and manuscript content.

“ChatGPT was not used as a source of evidentiary validation, and the authors remain responsible for all final translations, interpretations, and manuscript content.”

May combine two claims.

High perceived ease of use may increase behavioural uptake and operational usefulness while becoming educationally counterproductive when it substitutes rather than supports intended cognitive work (an 'ease-of-use inversion').

“This pattern can be described as an 'ease-of-use inversion', because high perceived ease of use may increase behavioural uptake and operational usefulness while becoming educationally counterproductive when it substitutes rather than supports intended cognitive work.”

May combine two claims.

Institutional AI infrastructure does not solve the pedagogical problem by itself, but gives institutions greater control over the conditions under which human-AI interaction occurs.

“Such infrastructure does not solve the pedagogical problem by itself, but it gives institutions greater control over the conditions under which human-AI interaction occurs.”

Classified as probably not a substantial claim. May combine two claims.

Taken together, these findings reveal a sustained ambivalence that is more analytically useful than a simple pro/anti-AI division and point towards curriculum and assessment redesign rather than either unrestricted adoption or prohibition.

“Taken together, these findings reveal a sustained ambivalence that is more analytically useful than a simple pro/anti-AI division and point towards curriculum and assessment redesign rather than either unrestricted adoption or prohibition.”

Classified as probably not a substantial claim.

The present study captures educator perceptions rather than direct measures of student cognition.

“Medical education is unusually high stakes, has distinctive clinical and ethical responsibilities, and the present study captures educator perceptions rather than direct measures of student cognition.”

Probably not a substantial claim: may not pass S3 (load-bearing).

AI-generated translation suggestions were treated as provisional editorial aids rather than as validated representations of the qualitative data.

“AI-generated translation suggestions were treated as provisional editorial aids rather than as validated representations of the qualitative data.”

Probably not a substantial claim: may not pass S2 (checkable).

AI literacy, educator development, privacy-preserving infrastructure, equitable access, and clear governance are not peripheral support to the co-intelligence model but the institutional conditions that make it possible.

“AI literacy, educator development, privacy-preserving infrastructure, equitable access, and clear governance are not peripheral support to this model, but the institutional conditions that make it possible.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

Adult professional education must prepare learners both to use AI effectively and to recognise situations in which it should not be relied upon.

“Adult professional education must therefore prepare learners both to use AI effectively and to recognise situations in which it should not be relied upon.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The researchers remained fully responsible for coding decisions, theme development, interpretation, and conclusions.

“The researchers remained fully responsible for coding decisions, theme development, interpretation, and conclusions.”

Classified as probably not a substantial claim.

The relevant design distinction is between friction that can appropriately be reduced (for example, linguistic barriers) and friction that should be preserved (such as selecting and weighing evidence in clinical reasoning).

“The relevant design distinction is consequently between friction that can appropriately be reduced-for example, linguistic barriers-and friction that should be preserved, such as selecting and weighing evidence in clinical reasoning.”

May combine two claims.

Continuing to rely on AI detection creates both practical and ethical problems: false suspicion can harm students, while undetected substitution can undermine validity.

“Continuing to rely on detection creates both practical and ethical problems: false suspicion can harm students, while undetected substitution can undermine validity.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

An inclusive design requires institutional infrastructure and governance as well as pedagogical creativity.

“An inclusive design therefore requires institutional infrastructure and governance as well as pedagogical creativity.”

Probably not a substantial claim: may not pass S2 (checkable).

Transfer of the findings to other adult professional programmes should be cautious.

“At the same time, transfer should be cautious.”

• AssumptionRests on an assumption: Medical education is unusually high stakes and has distinctive clinical and ethical responsibilities.

May combine two claims.

Students require knowledge in order to evaluate AI critically, yet premature cognitive offloading may weaken the development of the knowledge required for reliable evaluation.

“This sharpens the circular challenge evident in the participants' accounts: students require knowledge in order to evaluate AI critically, yet premature cognitive offloading may weaken the development of the knowledge required for reliable evaluation.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

The value of the findings lies in identifying mechanisms and design propositions that can be tested in other adult-learning contexts, not in claiming universal effects.

“The value of the findings lies in identifying mechanisms and design propositions that can be tested in other adult-learning contexts, not in claiming universal effects.”

Classified as probably not a substantial claim. May combine two claims.

A process-oriented model of responsible use is supported, in which a learner generates an initial response, uses GenAI to challenge or extend it, verifies the resulting claims against authoritative sources, and documents what changed.

“A learner can be required to generate an initial response, use GenAI to challenge or extend it, verify the resulting claims against authoritative sources, and document what changed.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The saturation judgement should not be interpreted as evidence of institutional or national representativeness, particularly given the concentration of participants at Erasmus MC.

“This judgement should not, however, be interpreted as evidence of institutional or national representativeness, particularly given the concentration of participants at Erasmus MC.”

Probably not a substantial claim: may not pass S2 (checkable).

If the learning objective is literature synthesis with responsible use of advanced tools, a transparent AI-enabled task may be authentic.

“If the learning objective is literature synthesis with responsible use of advanced tools, a transparent AI-enabled task may be authentic.”

Classified as probably not a substantial claim. May combine two claims.

These formats enable educators to probe a student's reasoning in real time and reduce ambiguity about whether a polished product represents actual understanding.

“These formats enable educators to probe a student's reasoning in real time and reduce ambiguity about whether a polished product represents actual understanding.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

AI literacy programmes should include staff as well as students, address discipline-specific risks, and be updated iteratively.

“AI literacy programmes should therefore include staff as well as students, address discipline-specific risks, and be updated iteratively.”

Classified as probably not a substantial claim.

Critical evaluation of a generated answer presupposes enough disciplinary knowledge to recognise whether the answer makes sense.

“Critical evaluation of a generated answer presupposes enough disciplinary knowledge to recognise whether the answer makes sense.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Participation was voluntary and written consent was obtained prior to data collection, in accordance with RSM guidance.

“Participation was voluntary and written consent was obtained prior to data collection, in accordance with RSM guidance requiring prospective participants to be free to decide whether to participate and whether data collected from them may be included in the analysis.”

Classified as probably not a substantial claim. May combine two claims.

These complementary proposals are synthesised here as a 'dual-mode assessment architecture': AI-augmented learning for comparison, simulation, feedback, and practice, coupled with strategically placed AI-free assessment for verification of core human competence.

“These complementary proposals are synthesised here as a 'dual-mode assessment architecture': AI-augmented learning for comparison, simulation, feedback, and practice, coupled with strategically placed AI-free assessment for verification of core human competence.”

Classified as probably not a substantial claim.

Although this study is situated in medical education, several mechanisms identified are relevant to other adult professional programmes.

“Although this study is situated in medical education, several mechanisms are relevant to other adult professional programmes.”

• AssumptionRests on an assumption: Practice-oriented professional fields other than medicine (law, business, teacher education, engineering) similarly require learners to build domain knowledge, communicate reasoning, judge evidence, and act under conditions where AI may be unavailable or unreliable.

Probably not a substantial claim: may not pass S3 (load-bearing).

All interview transcripts were checked against the recordings and corrected by the researchers.

“All transcripts were subsequently checked against the recordings and corrected by the researchers.”

Probably not a substantial claim: may not pass S2 (checkable).

Verification should be treated as a competence that must itself be taught, scaffolded, practised, and assessed rather than delegated to learners as an unsupported responsibility.

“Verification should therefore be treated as a competence that must itself be taught, scaffolded, practised, and assessed rather than delegated to learners as an unsupported responsibility.”

Classified as probably not a substantial claim.

Such sequences transform GenAI from an invisible production aid into a visible object of critical inquiry.

“Such sequences transform GenAI from an invisible production aid into a visible object of critical inquiry.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The study was conducted in accordance with the research ethics requirements of Rotterdam School of Management (RSM), Erasmus University Rotterdam, and the principles of the Netherlands Code of Conduct for Research Integrity.

“The study was conducted in accordance with the research ethics requirements of Rotterdam School of Management (RSM), Erasmus University Rotterdam, and the principles of the Netherlands Code of Conduct for Research Integrity.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Institutionally governed models may be valuable where privacy, standardisation, or equitable access are priorities.

“Institutionally governed models may be valuable where privacy, standardisation, or equitable access are priorities.”

May combine two claims.

Informational/thematic saturation had been reached for the principal patterns reported in this study, because later interviews predominantly reiterated previously identified issues and did not generate substantively new themes relevant to the research questions.

“Nevertheless, within the recruited sample, later interviews predominantly reiterated previously identified issues and did not generate substantively new themes relevant to the research questions, supporting the judgement that informational/thematic saturation had been reached for the principal patterns reported in this study.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Every Dutch quotation reported in the Results was rechecked against the original Dutch transcript by the first researcher, who is a native Dutch speaker, and independently reviewed against the source text by a bilingual Dutch-English reviewer.

“For the revised manuscript, every Dutch quotation reported in the Results was rechecked against the original Dutch transcript by the first researcher, who is a native Dutch speaker, and independently reviewed against the source text by a bilingual DutchEnglish reviewer.”

Probably not a substantial claim: may not pass S2 (checkable).

The balance between AI-enabled and AI-free assessment should be outcome-specific.

“The balance should be outcome-specific.”

Classified as probably not a substantial claim. May combine two claims.

AI-free assessment can verify foundational knowledge, clarity of explanation, and ad hoc reasoning.

“AI-free assessment can then verify foundational knowledge, clarity of explanation, and ad hoc reasoning.”