From production to verification: generative AI, doctoral formation, and the leadership of digital education

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

Becoming a scholar involves competence, confidence, autonomy, and agency, shaped by relationships among students, supervisors, programs, disciplines, and institutions.

“Research on doctoral identity development emphasizes that becoming a scholar involves competence, confidence, autonomy, and agency, shaped by relationships among students, supervisors, programs, disciplines, and institutions (Choi et al., 2021).”

• cites Choi et al., 2021 sourceSupported

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

AI tools may support multilingual writers, novice coders, and students entering unfamiliar conceptual terrain; large language models could lower the barrier to academic writing in English and reduce inequalities faced by second-language scholars.

“On one hand, AI tools may support multilingual writers, novice coders, and students entering unfamiliar conceptual terrain; some scholars have argued that large language models could lower the barrier to academic writing in English and reduce inequalities faced by second-language scholars (Hwang et al., 2023).”

• cites Hwang et al., 2023 sourceSupported

May combine two claims.

AI tools may intensify inequalities through differential access to paid tools, uneven institutional guidance, and disciplinary variation in what counts as legitimate assistance; AI detection tools have been shown to be biased against non-native writers.

“On the other hand, they may intensify inequalities through differential access to paid tools, uneven institutional guidance, and disciplinary variation in what counts as legitimate assistance, and AI detection tools have been shown to be biased against non-native writers (Liang et al., 2023).”

• cites Liang et al., 2023 sourceSupported

May combine two claims.

Authorship in the age of generative AI cannot be settled by detecting tool use alone, but depends on who can take accountability for the content.

“This pattern is consistent with broader arguments that authorship in the age of generative AI cannot be settled by detecting tool use alone, but depends on who can take accountability for the content (Bozkurt, 2024).”

• cites Bozkurt, 2024 sourceSupported

May combine two claims.

Students whose writing already marks them as non-native speakers may face additional suspicion if AI helps them produce more polished prose, a risk compounded by detection tools that disproportionately flag nonnative writing.

“Students whose writing already marks them as non-native speakers may face additional suspicion if AI helps them produce more polished prose, a risk compounded by detection tools that disproportionately flag nonnative writing (Liang et al., 2023).”

• cites Liang et al., 2023 sourcePartly supported

  The section on GPT-detector bias reports that detectors falsely flagged more than half of the non-native TOEFL essays, and the Discussion warns that non-native students face false accusations of cheating. The paper does not establish that AI-assisted polishing brings additional suspicion; its section on linguistic diversity enhancement reports that using ChatGPT to improve word choice reduced detector misclassification. No page numbers are available in our copy. (LLM, conf=0.84)

  Experimental — accuracy not yet measured

• AssumptionRests on an assumption: Academic English is not a neutral medium; it carries histories of disciplinary power, institutional expectation, and unequal preparation.

May combine two claims.

For other students, AI appeared mainly as technical infrastructure, supporting coding, debugging, and workflow acceleration while raising questions about verification and expertise.

“For others, it appeared mainly as technical infrastructure, supporting coding, debugging, and workflow acceleration while raising questions about verification and expertise.”

May combine two claims.

For some students, AI appeared mainly as linguistic and interpretive support, raising questions about voice, authorship, and academic suspicion.

“For some, AI appeared mainly as linguistic and interpretive support, raising questions about voice, authorship, and academic suspicion.”

May combine two claims.

AI lowered the threshold for entering a technical practice that might otherwise have been intimidating; it did not by itself confer proficiency, but it made first engagement possible.

“AI lowered the threshold for entering a technical practice that might otherwise have been intimidating; it did not by itself confer proficiency, but it made first engagement possible.”

• AssumptionRests on an assumption: Participants' self-reports in interviews accurately reflect their actual AI use practices and experiences.

May combine two claims.

AI did not have one stable meaning across doctoral work; where scholarly work was language-intensive, it often appeared as linguistic, rhetorical, and interpretive support; where work was more computational, it more often appeared as technical support.

“Taken together, these accounts suggest that AI did not have one stable meaning across doctoral work. Where scholarly work was language-intensive, it often appeared as linguistic, rhetorical, and interpretive support: grammar, tone, translation, concept clarification, and preparation for discussion. Where work was more computational, it more often appeared as technical support: coding, debugging, w”

• AssumptionRests on an assumption: Participants' self-reports in interviews accurately reflect their actual AI use practices and experiences.

May combine two claims.

The analysis identified that linguistic support and authorship anxiety were salient for students doing language-intensive work, and coding support and technical verification were salient for students doing computational work.

“This comparison helped identify both recurring patterns, such as the salience of linguistic support and authorship anxiety for students doing language-intensive work, and the salience of coding support and technical verification for students doing computational work.”

• AssumptionRests on an assumption: Participants' self-reports in interviews accurately reflect their actual AI use practices and experiences.

May combine two claims.

The distinction is not a matter of discipline as such, and it is not absolute: students doing language-intensive work also used AI for technical tasks, and students in computational fields also used it for writing.

“The distinction is not a matter of discipline as such, and it is not absolute: students doing language-intensive work also used AI for technical tasks, and students in computational fields also used it for writing.”

• AssumptionRests on an assumption: Participants' self-reports in interviews accurately reflect their actual AI use practices and experiences.

May combine two claims.

For these students, AI opened a door to a concept without replacing disciplinary reading; its value lay in orientation, not final judgment.

“For these students, AI opened a door to a concept without replacing disciplinary reading. Its value lay in orientation, not final judgment.”

• AssumptionRests on an assumption: Participants' self-reports in interviews accurately reflect their actual AI use practices and experiences.

May combine two claims.

Students doing language-intensive work often responded through disclosure, caution, and professor-by-professor negotiation, while students doing computational work more often responded through testing, debugging, checking, and task-specific judgment.

“Students doing language-intensive work often responded through disclosure, caution, and professor-by-professor negotiation, while students doing computational work more often responded through testing, debugging, checking, and taskspecific judgment.”

• AssumptionRests on an assumption: Participants' interview accounts accurately represent their actual AI use practices.

May combine two claims.

Across the interviews, doctoral students drew distinctions between grammar support and idea generation, sentence restructuring and argument construction, coding help and methodological understanding, and translation and scholarly voice.

“Across interviews, participants drew careful distinctions between grammar support and idea generation, sentence restructuring and argument construction, coding help and methodological understanding, translation and scholarly voice.”

• AssumptionRests on an assumption: Participants' interview accounts accurately represent their actual AI use practices.

Classified as probably not a substantial claim.

For some doctoral students, AI is experienced primarily as support for language, writing, tone, translation, and conceptual clarification.

“For some students, AI is experienced primarily as support for language, writing, tone, translation, and conceptual clarification”

• AssumptionRests on an assumption: Participants' self-reports in interviews accurately reflect their actual AI use practices and experiences.

May combine two claims.

The surface form of these practices varied with the kind of labor a student's work required: language-intensive work led to questions about own voice and idea, while computational work led to questions about correctness, validity, and defensibility.

“The surface form of these practices varied with the kind of labor a student's work required: those doing language-intensive work tended to ask whether the writing was still their own voice and idea, while those doing computational work tended to ask whether the output was correct, valid, and defensible.”

May combine two claims.

For many participants, grammar and sentence-level support were considered acceptable, while intellectual substitution remained troubling.

“For many participants, grammar and sentence-level support were considered acceptable, while intellectual substitution remained troubling.”

• AssumptionRests on an assumption: Participants' interview accounts accurately represent their actual AI use practices.

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

Many students hesitate to disclose AI use because disclosure can feel like admitting misconduct.

“Many students hesitate to disclose AI use because disclosure can feel like admitting misconduct.”

May combine two claims.

Students drew lines between grammar and thinking, restructuring and authorship, coding support and methodological understanding, conceptual orientation and intellectual substitution.

“Students drew lines between grammar and thinking, restructuring and authorship, coding support and methodological understanding, conceptual orientation and intellectual substitution.”

Classified as probably not a substantial claim.

Generative AI has entered the ordinary infrastructure of scholarly work.

“Generative AI has entered the ordinary infrastructure of scholarly work.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

Generative AI is better understood as a sociotechnical condition that reshapes the everyday practices through which doctoral students learn to become scholars.

“It is better understood as a sociotechnical condition that reshapes the everyday practices through which doctoral students learn to become scholars.”

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

When institutional policies are unclear or uneven, students must interpret expectations for themselves.

“When institutional policies are unclear or uneven, students must interpret expectations for themselves.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

The problem of generative AI in doctoral education is not simply that students may misuse tools; the more fundamental issue is that AI intervenes in the practices through which doctoral students have traditionally developed scholarly judgment.

“The problem of generative AI in doctoral education is not simply that students may misuse tools. The more fundamental issue is that AI intervenes in the practices through which doctoral students have traditionally developed scholarly judgment.”

• AssumptionThe authors state this as an assumption: Students become scholars by doing the work of scholarship.

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

AI tools can support academic writing and research while raising concerns about integrity, transparency, and the balance between machine assistance and human insight.

“Scholarship has noted that AI tools can support academic writing and research while raising concerns about integrity, transparency, and the balance between machine assistance and human insight (Khalifa and Albadawy, 2024).”

• cites Khalifa and Albadawy, 2024 sourceNot checked — no full text of the source

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

In traditional doctoral training, production and verification were often closely linked: students learned by producing drafts, analyses, and interpretations themselves.

“In traditional doctoral training, production and verification were often closely linked: students learned by producing drafts, analyses, and interpretations themselves.”

Classified as probably not a substantial claim.

Writing, coding, synthesizing literature, designing studies, and interpreting evidence are formative practices through which doctoral students learn what counts as knowledge, evidence, and responsibility.

“Writing, coding, synthesizing literature, designing studies, and interpreting evidence are therefore not merely academic tasks. They are formative practices through which doctoral students learn what counts as knowledge, evidence, and responsibility.”

• AssumptionThe authors state this as an assumption: Students become scholars by doing the work of scholarship.

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

Independence can no longer be inferred from polished products alone; it must be demonstrated through judgment, explanation, verification, and responsibility.

“Independence can no longer be inferred from polished products alone. It must be demonstrated through judgment, explanation, verification, and responsibility, which are precisely the capacities a defense is designed to test.”

• AssumptionRests on an assumption: In most doctoral systems, scholarly formation is assessed through a culminating oral defense in which candidates explain, justify, and stand behind their work before disciplinary experts.

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

A qualitative design is appropriate because the purpose of the study is not to measure the prevalence of AI use, but to analyze how doctoral students make meaning of AI-assisted writing, reading, coding, research, authorship, verification, and academic responsibility.

“A qualitative design is appropriate because the purpose of the study is not to measure the prevalence of AI use, but to analyze how doctoral students make meaning of AI-assisted writing, reading, coding, research, authorship, verification, and academic responsibility.”

May combine two claims.

The study does not treat any disciplinary category as a homogeneous population, and it does not attribute differences to field membership as such.

“The study does not treat any disciplinary category as a homogeneous population, and it does not attribute differences to field membership as such.”

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

Doctoral education is best understood as scholarly formation.

“This is why doctoral education is best understood as scholarly formation.”

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

Generative AI can support these tasks, but it can also obscure whether students understand the reasoning behind them.

“Generative AI can support these tasks, but it can also obscure whether students understand the reasoning behind them.”

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

Generative AI loosens the connection between production and verification; it can produce outputs before understanding has fully developed.

“Generative AI loosens that connection. It can produce outputs before understanding has fully developed.”

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

Generative AI is not simply a technology problem; it is a leadership problem because it requires institutions to clarify the purposes, expectations, and evaluative practices of doctoral education.

“In this sense, generative AI is not simply a technology problem. It is a leadership problem because it requires institutions to clarify the purposes, expectations, and evaluative practices of doctoral education.”

Classified as probably not a substantial claim.

Doctoral writing is part of 'writing the scholar' into being.

“Kamler and Thomson (2014) similarly describe doctoral writing as part of 'writing the scholar' into being.”

• cites Kamler and Thomson, 2014 sourceNot checked — no full text of the source

May combine two claims.

Generative AI does not enter doctoral education as a single, uniform tool; its meaning depends on the scholarly labor at stake.

“This design is appropriate because generative AI does not enter doctoral education as a single, uniform tool. Its meaning depends on the scholarly labor at stake.”

• AssumptionRests on an assumption: Participants' self-reports in interviews accurately reflect their actual AI use practices and experiences.

Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

Verification includes evaluating sources, testing code, identifying hallucinations, judging conceptual coherence, preserving scholarly voice, assessing methodological fit, and deciding whether one can ethically stand behind a claim.

“Verification includes evaluating sources, testing code, identifying hallucinations, judging conceptual coherence, preserving scholarly voice, assessing methodological fit, and deciding whether one can ethically stand behind a claim.”

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

Self-governance is produced by a policy environment in which universities often provide broad statements while leaving the meaning of acceptable use to be negotiated in courses, research groups, and disciplinary communities.

“This self-governance is not simply an individual ethical choice. It is produced by a policy environment in which universities often provide broad statements while leaving the meaning of acceptable use to be negotiated in courses, research groups, and disciplinary communities.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

Coding, modeling, data analysis, literature synthesis, and methodological design are ways doctoral students learn how evidence is made, how methods carry assumptions, and how claims become defensible.

“Coding, modeling, data analysis, literature synthesis, and methodological design are not merely tasks to be completed. They are ways doctoral students learn how evidence is made, how methods carry assumptions, and how claims become defensible.”

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

Generative AI often enters doctoral life through writing: grammar correction, translation, sentence restructuring, tone adjustment, outlining, summarizing, and revision.

“These arguments matter because generative AI often enters doctoral life through writing: grammar correction, translation, sentence restructuring, tone adjustment, outlining, summarizing, and revision.”

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

The study was conducted under institutional review board approval.

“The study was conducted under institutional review board approval.”

May combine two claims.

This self-governance was not simply a matter of individual ethics; it was produced by institutional ambiguity.

“This self-governance was not simply a matter of individual ethics. It was produced by institutional ambiguity”

May combine two claims.

Doctoral students inhabit a governance gap: institutions issue broad statements, instructors make course-level decisions, disciplines develop informal norms, and students must assemble these signals into workable practice.

“These accounts suggest that doctoral students inhabit a governance gap. Institutions issue broad statements, instructors make course-level decisions, disciplines develop informal norms, and students must assemble these signals into workable practice.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

One of the risks of AI-mediated scholarship is that students may lose the formative value of struggle if they delegate too much too early.

“Indeed, one of the risks of AI-mediated scholarship is that students may lose the formative value of struggle if they delegate too much too early.”

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

Lovitts (2005) argues that doctoral education requires a transition from being a successful coursetaker to becoming an independent researcher, a transition that depends on learning to make judgments under conditions of uncertainty.

“Lovitts (2005) argues that doctoral education requires a transition from being a successful coursetaker to becoming an independent researcher, a transition that depends on learning to make judgments under conditions of uncertainty.”

• cites Lovitts (2005) sourceNot checked — no full text of the source

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

The distinctive responsibility of the doctoral student increasingly lies in the ability to interrogate, revise, verify, and stand behind machine-assisted work.

“The distinctive responsibility of the doctoral student increasingly lies in the ability to interrogate, revise, verify, and stand behind machine-assisted work.”

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

The salience of linguistic equity in the findings may partly reflect the researcher's own sensitivity to that issue, and there is a danger of reading participants' accounts through the lens of the researcher's experience.

“The salience of linguistic equity in the findings may partly reflect the researcher's own sensitivity to that issue, and there is a danger of reading participants' accounts through the lens of the researcher's experience.”

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

Production alone is no longer sufficient as evidence of scholarly formation.

“Production alone is no longer sufficient as evidence of scholarly formation.”

May combine two claims.

AI access is never merely instrumental; it is bound to legitimacy, learning, and responsibility.

“AI access, in this sense, is never merely instrumental. It is bound to legitimacy, learning, and responsibility.”

• AssumptionRests on an assumption: Participants' self-reports in interviews accurately reflect their actual AI use practices and experiences.

May combine two claims.

AI can make code appear, but it cannot by itself make the student methodologically responsible for that code.

“AI can make code appear, but it cannot by itself make the student methodologically responsible for that code.”

Classified as probably not a substantial claim.

The analysis compared cases across different language backgrounds, doctoral stages, and disciplinary contexts to avoid relying on a single dominant narrative.

“the analysis compared cases across different language backgrounds, doctoral stages, and disciplinary contexts to avoid relying on a single dominant narrative.”

May combine two claims.

AI creates new dependencies and new risks.

“At the same time, it creates new dependencies and new risks.”

• AssumptionRests on an assumption: Participants' self-reports in interviews accurately reflect their actual AI use practices and experiences.

May combine two claims.

AI appeared to shift the visible center of doctoral labor from production toward verification, from fluency toward judgment, and from speed toward responsibility.

“More precisely, AI appeared to shift the visible center of doctoral labor from production toward verification, from fluency toward judgment, and from speed toward responsibility.”

Classified as probably not a substantial claim.

The sample is not intended to be representative of all doctoral students.

“The resulting sample is not intended to be representative of all doctoral students.”

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

The same researcher conducted all interviews using a single protocol to maintain consistency in probing and follow-up.

“A single interview protocol was used throughout, and the same researcher conducted all interviews to maintain consistency in probing and follow-up.”

May combine two claims.

AI-assisted writing can be both empowering and precarious for multilingual doctoral students: it gives access to a professional register that may otherwise take years to acquire, but also exposes them to suspicion.

“This cluster of cases is analytically important because it shows that AI-assisted writing can be both empowering and precarious. It gives multilingual doctoral students access to a professional register that may otherwise take years to acquire. But it also exposes them to suspicion.”

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

AI-assisted writing must be understood as an intervention into the formation of scholarly voice, authorship, and judgment, not only as textual improvement.

“then AI-assisted writing cannot be treated only as a matter of textual improvement. It must be understood as an intervention into the formation of scholarly voice, authorship, and judgment.”

May combine two claims.

Later interviews largely confirmed already identified patterns and did not introduce substantially new themes, indicating sufficient thematic saturation.

“Later interviews largely confirmed already identified patterns and did not introduce substantially new themes relevant to the study's research questions, indicating sufficient thematic saturation for the purposes of this study.”

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

The researcher's insider position likely facilitated rapport and candor, particularly around language-related anxiety, and may have made participants more willing to disclose cautious or concealed AI practices.

“This insider position likely facilitated rapport and candor, particularly around language-related anxiety, and it may have made participants more willing to disclose cautious or concealed AI practices.”

Classified as probably not a substantial claim.

The analysis deliberately retained contradictory and negative cases so that the themes would account for variation rather than confirm a preferred interpretation.

“the analysis deliberately retained contradictory and negative cases, including participants who used AI frequently, participants who used it cautiously, and participants who distrusted AI-generated content, so that the themes would account for variation rather than confirm a preferred interpretation.”

Classified as probably not a substantial claim.

Gieryn (1983) used boundary work to analyze how scientific authority is constructed through demarcation.

“Gieryn (1983) used boundary work to analyze how scientific authority is constructed through demarcation.”

• cites Gieryn (1983) sourceNot checked — no full text of the source

May combine two claims.

The study was conducted by a single researcher, who collected and analyzed all data, which carries a risk of analyst bias and limits procedural safeguards.

“This study was conducted by a single researcher, who collected and analyzed all data. This design has the advantage of interpretive consistency but carries a clear risk of analyst bias, and it limits the procedural safeguards available to multi-researcher teams.”

Classified as probably not a substantial claim.

To partially offset the absence of these procedures, the analysis relied on a documented audit trail of analytic memos, systematic constant comparison, deliberate attention to disconfirming cases, and retention of verbatim participant language.

“To partially offset the absence of these procedures, the analysis relied on a documented audit trail of analytic memos, systematic constant comparison within and across groups, deliberate attention to disconfirming cases, and the retention of verbatim participant language so that readers can assess the fit between evidence and interpretation.”

Classified as probably not a substantial claim.

Kamler and Thomson (2014) argue that doctoral writing is part of 'writing the scholar' into being.

“if doctoral writing is part of 'writing the scholar' into being, as Kamler and Thomson (2014) argue”

• cites Kamler and Thomson (2014) sourceNot checked — no full text of the source

May combine two claims.

The study did not employ a second independent coder, formal intercoder agreement, or external auditing, and it did not conduct member checking.

“The study did not employ a second independent coder, formal intercoder agreement, or external auditing, and it did not conduct member checking in which participants reviewed transcripts, codes, or findings.”

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

Boundary work, scholarly formation, and verification were not used as a pre-specified deductive codebook imposed on the transcripts; rather, they functioned as sensitizing concepts.

“Boundary work, scholarly formation, and verification were not used as a pre-specified deductive codebook imposed on the transcripts; rather, they functioned as sensitizing concepts”

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

AI offers answers quickly but does not remove the need to decide whether the answer is conceptually sound, methodologically appropriate, ethically defensible, and intellectually one's own.

“It offers answers quickly, but it does not remove the need to decide whether the answer is conceptually sound, methodologically appropriate, ethically defensible, and intellectually one's own.”

Classified as probably not a substantial claim.

The themes reported were developed from the data and then refined in dialogue with theory, not predetermined.

“the themes reported below were developed from the data and then refined in dialogue with theory, not predetermined.”

Classified as probably not a substantial claim.

The analysis does not attribute any pattern to a specific department, program, or college.

“the analysis does not attribute any pattern to a specific department, program, or college.”

May combine two claims.

Generative AI has not made doctoral education less necessary; it has made the purposes of doctoral education more visible and more urgent.

“Generative AI has not made doctoral education less necessary. It has made its purposes more visible and more urgent.”

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

The study does not constitute a formal comparison between fields.

“it does not constitute a formal comparison between fields”

May combine two claims.

Students with paid subscriptions, stronger prompt literacy, clearer faculty guidance, or more AI-tolerant disciplinary cultures may benefit more from AI.

“Students with paid subscriptions, stronger prompt literacy, clearer faculty guidance, or more AI-tolerant disciplinary cultures may benefit more.”

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

The study captures a single moment in a fast-moving technological landscape.

“the study captures a single moment in a fast-moving technological landscape”

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

The defense is where that independence is ultimately certified.

“the defense is where that independence is ultimately certified.”

• AssumptionRests on an assumption: The oral defense is the established setting in which doctoral candidates account for their work.

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

Faculty cannot infer scholarly judgment from polished writing or functional code alone.

“Faculty cannot infer scholarly judgment from polished writing or functional code alone.”

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

The study's claims are grounded in a single-site interpretive study and are offered as propositions for wider testing rather than as settled conclusions.

“These claims are grounded in a single-site interpretive study and are offered as propositions for wider testing rather than as settled conclusions.”

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

The study was conducted by a single researcher.

“the study was conducted by a single researcher”

May combine two claims.

The study did not employ a second coder, formal intercoder agreement, member checking, or external auditing.

“did not employ a second coder, formal intercoder agreement, member checking, or external auditing.”

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

Verbal informed consent was obtained from all participants before the interviews began.

“Verbal informed consent was obtained from all participants before the interviews began.”

May combine two claims.

A permissive policy may encourage experimentation but may also intensify inequalities between students who can verify AI output and those who cannot.

“A permissive policy may encourage experimentation, but it may also intensify inequalities between students who can verify AI output and those who cannot.”

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

Speed does not equal understanding.

“Yet speed does not equal understanding, fluency does not equal judgment, and output does not equal scholarship.”

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

The studies were conducted in accordance with local legislation and institutional requirements.

“The studies were conducted in accordance with local legislation and institutional requirements.”

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

The credibility safeguards available were an audit trail of memos, constant comparison, attention to disconfirming cases, and retention of verbatim language.

“The credibility safeguards available were an audit trail of memos, constant comparison, attention to disconfirming cases, and retention of verbatim language.”

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

Expertise in the AI era resides less in the mere appearance of scholarly fluency and more in the capacity to explain process, verify output, preserve intellectual agency, and assume responsibility for claims.

“It resides less in the mere appearance of scholarly fluency and more in the capacity to explain process, verify output, preserve intellectual agency, and assume responsibility for claims.”

May combine two claims.

Expertise in the AI era is not less important, but more difficult to recognize.

“The implication is that expertise in the AI era is not less important, but more difficult to recognize.”

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

Long and Magerko (2020) define AI literacy as the ability to understand, evaluate, and interact with AI systems.

“Long and Magerko (2020) define AI literacy as the ability to understand, evaluate, and interact with AI systems.”

• cites Long and Magerko (2020) sourceNot checked — no full text of the source

May combine two claims.

AI-mediated access is unevenly distributed and institutionally fragile.

“They show that AI-mediated access is unevenly distributed and institutionally fragile.”

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

Output does not equal scholarship.

“Yet speed does not equal understanding, fluency does not equal judgment, and output does not equal scholarship.”

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

The patterns reported in this study may not transfer to other institutions, national systems, or policy environments.

“the patterns reported here may not transfer to other institutions, national systems, or policy environments.”

• AssumptionRests on an assumption: Institutional culture, local AI policies, and program-specific norms may strongly shape how doctoral students encounter and reason about AI.

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

A policy that treats sentence polishing, literature synthesis, code generation, and argument production as the same kind of AI use will be too blunt to guide doctoral education.

“A policy that treats sentence polishing, literature synthesis, code generation, and argument production as the same kind of AI use will be too blunt to guide doctoral education.”

May combine two claims.

Social desirability bias is a real possibility: participants may have understated concealed use or overstated their verification practices.

“social desirability bias is a real possibility: participants may have understated concealed use or overstated their verification practices.”

• AssumptionRests on an assumption: AI use is ethically sensitive.

May combine two claims.

No observational data, writing samples, AI interaction logs, assignment artifacts, or institutional policy documents were analyzed.

“No observational data, writing samples, AI interaction logs, assignment artifacts, or institutional policy documents were analyzed.”

May combine two claims.

These lines were not merely defensive; they were formative.

“These lines were not merely defensive. They were formative.”

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

As tools and norms change, the specific practices reported here may date quickly even if the underlying questions about judgment and responsibility persist.

“as tools and norms change, the specific practices reported here may date quickly even if the underlying questions about judgment and responsibility persist.”

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

The study's observations about disciplinary variation should not be read as field-level claims.

“its observations about disciplinary variation should not be read as field-level claims.”

May combine two claims.

AI may reduce one barrier while creating another.

“AI may reduce one barrier while creating another.”

May combine two claims.

Generative AI has an equity paradox: it can reduce barriers to doctoral participation but can also produce new forms of dependency, suspicion, and stratification.

“The findings also reveal an equity paradox. Generative AI can reduce barriers to doctoral participation, but it can also produce new forms of dependency, suspicion, and stratification.”

• AssumptionRests on an assumption: Academic English is not a neutral medium; it carries histories of disciplinary power, institutional expectation, and unequal preparation.
• AssumptionRests on an assumption: Technical work in coding, modeling, and data analysis often privileges students who enter doctoral programs with prior exposure to computational tools.

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

Verification is now central to the independence of doctoral scholars.

“verification is now central to that independence”

• AssumptionRests on an assumption: Doctoral education is the formation of independent scholars.

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

Students who use AI for technical work may become more efficient while also becoming dependent on outputs they cannot fully explain.

“Students who use AI for technical work may become more efficient while also becoming dependent on outputs they cannot fully explain.”

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

The future of doctoral education will depend less on whether students use AI than on whether doctoral programs teach them what cannot be delegated to it.

“The future of doctoral education will depend less on whether students use AI than on whether doctoral programs teach them what cannot be delegated to it.”

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

The IRB approved a waiver of written documentation of informed consent.

“The IRB approved a waiver of written documentation of informed consent.”

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

Without attention to equity and judgment in AI literacy, AI may become another mechanism through which advantage is quietly reproduced.

“Without such attention, AI may become another mechanism through which advantage is quietly reproduced.”

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

The defense is well placed to assess precisely the verification capacities this study foregrounds.

“the defense is well placed to assess precisely the verification capacities this study foregrounds”

• AssumptionRests on an assumption: The oral defense is the established setting in which doctoral candidates account for their work.

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

Fluency does not equal judgment.

“Yet speed does not equal understanding, fluency does not equal judgment, and output does not equal scholarship.”

Classified as probably not a substantial claim.

The study's sample was assembled through purposive and snowball recruitment.

“was assembled through purposive and snowball recruitment”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

This creates a culture of concealment rather than accountability.

“This creates a culture of concealment rather than accountability.”

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

For the leadership of digital education, broad AI policies are necessary but not sufficient.

“For the leadership of digital education, the implication is that broad AI policies are necessary but not sufficient.”

May combine two claims.

A blanket prohibition may protect some forms of learning but may also deny linguistic and technical support to students who need it.

“A blanket prohibition may protect some forms of learning, but it may also deny linguistic and technical support to students who need it.”

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

The connection drawn in this article between verification and the defense is an interpretive extension of the data rather than a directly evidenced finding.

“The connection drawn in this article between verification and the defense is therefore an interpretive extension of the data rather than a directly evidenced finding.”

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

The studies involving humans were reviewed and approved by the Institutional Review Board at The Pennsylvania State University on 28/01/26 under study number STUDY00028580.

“The studies involving humans were reviewed and approved by the Institutional Review Board at The Pennsylvania State University on 28/01/26 under study number STUDY00028580.”

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

In an AI-mediated environment, the oral defense's role as a test of whether candidates can verify and defend their own work becomes more, not less, important.

“in an AI-mediated environment, its role as a test of whether candidates can verify and defend their own work becomes more, not less, important.”

• AssumptionRests on an assumption: The oral defense is the established setting in which doctoral candidates account for their work.