A Study on the Role of Midjourney, A Generative Artificial Intelligence Tool, in Enhancing Student Creativity in Illustration Design Classes

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

In visual art education, AI-generated images have improved classroom engagement and self-efficacy without increasing cognitive load, although outcomes remain centred on engagement, self-belief, cognitive load, and painting-related performance.

“In visual art education, AI-generated images have improved classroom engagement and self-efficacy without increasing cognitive load, although the measured outcomes remain centred on engagement, self-belief, cognitive load, and painting-related performance (Bian et al. 2025).”

• cites Bian et al. 2025 sourceSupported

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

In graphic design education, text-to-image workshops have supported historical design poster production and shifted attention toward AI visual literacy, keyword formulation, design research, trend analysis, and content-creator competence.

“In graphic design education, text-to-image workshops have supported historical design poster production and shifted attention toward AI visual literacy, keyword formulation, design research, trend analysis, and content-creator competence (Hwang and Wu 2025).”

• cites Hwang and Wu 2025 sourceSupported

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

GenAI can improve efficiency and prompt literacy while increasing overreliance and outsourcing of cognitive effort.

“Higher-education reviews similarly caution that GenAI can improve efficiency and prompt literacy while also increasing overreliance and outsourcing of cognitive or metacognitive effort (Qian 2025).”

• cites Qian 2025 sourceSupported

May combine two claims.

GenAI is most commonly integrated into early conceptualisation and curation, while human judgement remains necessary in convergent stages.

“Studies of professional creative practice further suggest that GenAI is most commonly integrated into early conceptualisation and curation, while human judgement remains necessary in convergent stages of selection, refinement, and final authorship (Tsao et al. 2025).”

• cites Tsao et al. 2025 sourceSupported

May combine two claims.

Exposure to AI-generated images can increase fixation on initial examples and reduce idea variety and originality when learners rely heavily on generated references.

“Experimental evidence on visual ideation shows that exposure to AI-generated images can increase fixation on initial examples and reduce idea variety and originality when learners rely heavily on generated references (Wadinambiarachchi et al. 2024).”

• cites Wadinambiarachchi et al. 2024 sourceSupported

May combine two claims.

AI-generated images can strengthen classroom engagement and self-efficacy without increasing cognitive load, but this evidence is only partially transferable to undergraduate illustration settings.

“Evidence from visual art education further suggests that Artificial Intelligence (AI)-generated images can strengthen classroom engagement and self-efficacy without increasing cognitive load, yet this evidence remains only partially transferable to undergraduate illustration settings where originality, authorship, and final artefact quality carry greater assessment weight (Bian et al. 2025).”

• cites Bian et al. 2025 sourcePartly supported

  The Introduction reports higher engagement and self-efficacy, with no increased cognitive load, in a study of 78 fifth-grade art students. The limitations section calls for research across other grades and settings, supporting caution about transferability. It does not establish how transferable the findings are to undergraduate illustration or whether originality, authorship, and final artefact quality carry greater assessment weight there. (LLM, conf=0.85)

  Experimental — accuracy not yet measured

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

A seven-week course case reports high engagement and rapid ideation, but relies on perception measures and a small single context sample.

“A seven week course case using surveys, reflections, and course metrics reports high engagement and rapid ideation, but relies on perception measures and a small single context sample (Garvey 2025).”

• cites Garvey 2025 sourcePartly supported

  The Introduction reports full enrollment in both 2024 semesters, and Student Course Survey Responses reports student engagement while noting low response counts (Fall n=7; Spring n=13). The course is described as an accelerated seven-week online course, so the small, single-course context and reliance on student perceptions are fair characterizations. However, the paper does not report rapid ideation as a course finding; it discusses quick image creation more generally. Written responses are described as coursework, not clearly as a measure used to establish that finding. (LLM, conf=0.84)

  Experimental — accuracy not yet measured

May combine two claims.

A controlled comparison in digital textile design found that GenAI reduced perceived task difficulty and workload, yet weakened editability, personal-style articulation, creativity ratings, aesthetic value, satisfaction, and goal fulfilment.

“A controlled comparison in digital textile design found that GenAI reduced perceived task difficulty and workload, yet weakened editability, personal-style articulation, creativity ratings, aesthetic value, satisfaction, and goal fulfilment, indicating that easier production can coincide with weaker expressive control (Jang 2026).”

• cites Jang 2026 sourceSupported

May combine two claims.

Students require AI visual literacy, keyword precision, research competence, and content-creation judgement to use text-to-image tools productively.

“Graphic design education research on text-to-image generation shows that students require AI visual literacy, keyword precision, research competence, and content-creation judgement to use tools such as Midjourney and DALL-E productively in design courses (Hwang and Wu 2025).”

• cites Hwang and Wu 2025 sourceSupported

May combine two claims.

GenAI can support creativity, critical thinking, learning autonomy, prompt literacy, iterative feedback, and rapid variation when embedded within structured pedagogy, but its value declines when tool use is treated as unsupervised production rather than guided learning.

“Recent higher-education evidence shows that GenAI can support creativity, critical thinking, learning autonomy, prompt literacy, iterative feedback, and rapid variation when embedded within structured pedagogy, but its value declines when tool use is treated as unsupervised production rather than guided learning (Qian 2025).”

• cites Qian 2025 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

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

A prior power analysis specified a minimum target of 120 participants for a two-group comparison with baseline adjustment, with planned over-recruitment to accommodate attrition and incomplete documentation, using conventional small to medium effect size expectations.

“A prior power analysis conducted in G Power at alpha 0.05 specified a minimum target of 120 participants for a two-group comparison with baseline adjustment, with planned over-recruitment to accommodate attrition and incomplete documentation, using conventional small to medium effect size expectations for educational interventions (Cohen 1992;Faul et al. 2009Faul et al. , 2007)).”

• cites Cohen 1992 sourceNot checked — no full text of the source
• cites Faul et al. 2009 sourcePartly supported

  The paper’s Abstract (page unavailable) describes G*Power’s a priori analyses as calculating sample size from a chosen significance level, power, and effect size; Section 3.2 (page unavailable) also describes a two-group comparison. It does not report the citing study’s target of 120 participants, baseline adjustment, over-recruitment for attrition or incomplete documentation, or effect-size expectations for educational interventions. (LLM, conf=0.86)

  Experimental — accuracy not yet measured
• cites Faul et al. 2007 sourcePartly supported

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

  Experimental — accuracy not yet measured

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

Meaningful revision requires the student to retain control over action, decision autonomy, and ownership of the final artefact.

“The finding also connects to human-AI agency research because meaningful revision requires the student to retain control over action, decision autonomy, and ownership of the final artefact (Rafner et al. 2025).”

• cites Rafner et al. 2025 sourcePartly supported

  The sections “Is this mAIne?” and “Creativity as agentic action in human-AI co-creation” link creative control, autonomy, and ownership; participants who retained editorial control reported a stronger sense of ownership. The paper does not establish that retaining all three is required for meaningful revision by students. Its findings concern exploratory human-AI image generation and creative writing, not student revision specifically; page numbers were unavailable in our copy. (LLM, conf=0.88)

  Experimental — accuracy not yet measured

May combine two claims.

AI-generated examples can anchor subsequent ideas and reduce divergent thinking when learners rely heavily on generated references.

“Prior work on design fixation shows that AI-generated examples can anchor subsequent ideas and reduce divergent thinking when learners rely heavily on generated references (Wadinambiarachchi et al. 2024).”

• cites Wadinambiarachchi et al. 2024 sourceSupported

May combine two claims.

In the protocol-compliant subset, the adjusted mean was 3.43 ± 0.06 for S-GCEM and 3.23 ± 0.06 for Control, producing an adjusted mean difference of 0.20, 95% CI [0.06, 0.34].

“The adjusted mean was 3.43 ± 0.06 for S-GCEM and 3.23 ± 0.06 for Control, producing an adjusted mean difference of 0.20, 95% CI [0.06, 0.34].”

May combine two claims.

For Elaboration, the adjusted mean was 3.30 ± 0.06 for S-GCEM and 3.10 ± 0.05 for Control, adjusted mean difference = 0.20, 95% CI [0.07, 0.33], with a significant condition effect, F(1,159) = 9.00, p = 0.003, η2p = 0.054.

“For Elaboration, the adjusted mean was 3.30 ± 0.06 for S-GCEM and 3.10 ± 0.05 for Control, adjusted mean difference = 0.20, 95% CI [0.07, 0.33], with a significant condition effect, F(1,159) = 9.00, p = 0.003, η 2 p = 0.054 (Table 2).”

• AssumptionRests on an assumption: The three-dimension rubric validly measures product creativity in illustration design.

May combine two claims.

The mean homogenisation score was 0.83 ± 0.74 in S-GCEM and 1.27 ± 0.82 in Control.

“The mean homogenisation score was 0.83 ± 0.74 in S-GCEM and 1.27 ± 0.82 in Control.”

May combine two claims.

In the final Product Creativity model, Ideation Breadth remained positively associated with Product Creativity, B = 0.08, p < 0.001, and Iteration Depth also remained positively associated with Product Creativity, B = 0.06, p = 0.003.

“In the final Product Creativity model, Ideation Breadth remained positively associated with Product Creativity, B = 0.08, p < 0.001, and Iteration Depth also remained positively associated with Product Creativity, B = 0.06, p = 0.003.”

May combine two claims.

Final Product Creativity was higher under S-GCEM than the conventional workflow, with a significant condition effect (p = 0.003, η2p = 0.053).

“This study shows that final Product Creativity was higher under S GCEM than the conventional workflow, with a significant condition effect (p = 0.003, η 2 p = 0.053).”

• Also stated in: § 4.3 | Conclusion · plain

  “| Product Creativity | 3.42 ± 0.05 | 3.24 ± 0.05 | 0.18 | 0.06 to 0.30 | 9.00 | 0.003 | 0.053 |”

May combine two claims.

For Appropriateness, the adjusted mean was 3.46 ± 0.05 for S-GCEM and 3.34 ± 0.05 for Control, adjusted mean difference = 0.12, 95% CI [0.00, 0.24], with a significant condition effect, F(1,159) = 4.00, p = 0.047, η2p = 0.024.

“For Appropriateness, the adjusted mean was 3.46 ± 0.05 for S-GCEM and 3.34 ± 0.05 for Control, adjusted mean difference = 0.12, 95% CI [0.00, 0.24], with a significant condition effect, F(1,159) = 4.00, p = 0.047, η 2 p = 0.024.”

• AssumptionRests on an assumption: The three-dimension rubric validly measures product creativity in illustration design.

May combine two claims.

Satisfaction was rated 4.18 ± 0.72 in S-GCEM and 3.70 ± 0.86 in Control, p = 0.002.

“Satisfaction was rated 4.18 ± 0.72 in S-GCEM and 3.70 ± 0.86 in Control, p = 0.002.”

May combine two claims.

For Originality, the adjusted mean was 3.51 ± 0.06 for S-GCEM and 3.32 ± 0.06 for Control, adjusted mean difference = 0.19, 95% CI [0.04, 0.34], with a significant condition effect, F(1,159) = 6.00, p = 0.015, η2p = 0.036.

“For Originality, the adjusted mean was 3.51 ± 0.06 for S-GCEM and 3.32 ± 0.06 for Control, adjusted mean difference = 0.19, 95% CI [0.04, 0.34].The condition effect was statistically significant, F(1,159) = 6.00, p = 0.015, η 2 p = 0.036”

• AssumptionRests on an assumption: The three-dimension rubric validly measures product creativity in illustration design.

May combine two claims.

Ideation support was rated 4.21 ± 0.68 in S-GCEM and 3.56 ± 0.81 in Control, p < 0.001.

“Ideation support was rated 4.21 ± 0.68 in S-GCEM and 3.56 ± 0.81 in Control, p < 0.001.”

May combine two claims.

Authorship control was similar between groups, 3.88 ± 0.77 versus 4.02 ± 0.74, p = 0.239, indicating that Midjourney-supported ideation did not reduce students' reported sense of control over the final artefact.

“Authorship control was similar between groups, 3.88 ± 0.77 versus 4.02 ± 0.74, p = 0.239, indicating that Midjourney-supported ideation did not reduce students' reported sense of control over the final artefact.”

May combine two claims.

After adjustment for Baseline Product Creativity, final Product Creativity was higher in the S-GCEM group than in Control, with an adjusted mean difference of 0.18 (95% CI [0.06, 0.30]), F(1,159) = 9.00, p = 0.003, η2p = 0.053.

“Analysis of covariance showed higher final Product Creativity in the S-GCEM group after adjustment for Baseline Product Creativity.The adjusted mean was 3.42 ± 0.05 for S-GCEM and 3.24 ± 0.05 for Control, with an adjusted mean difference of 0.18, 95% CI [0.06, 0.30].”

• Also stated in: § 3.2 | Primary Outcomes · plain

  “Condition showed a statistically significant effect on Final Product Creativity (F(1,159) = 9.000, p = 0.003, η 2 p = 0.053).”

• AssumptionRests on an assumption: The three-dimension rubric validly measures product creativity in illustration design.

May combine two claims.

A drawing-based originality rubric shows significant gains and clear scoring dimensions, but age level and task simplicity limit transfer to studio illustration briefs.

“A drawing based originality rubric validated with pre and posttests shows significant gains and clear scoring dimensions, but age level and task simplicity limit transfer to studio illustration briefs (Xu and Rahman 2025).”

• cites Xu and Rahman 2025 sourceNot checked — no full text of the source

May combine two claims.

A policy-oriented synthesis treats creativity as measurable through fluency, originality, and adaptability, but domain fit for illustration products is indirect.

“A policy oriented synthesis using Organization for Economic Co-operation and Development (OECD) Programme for International Student Assessment (PISA) 2022 Creative Thinking treats creativity as measurable through fluency, originality, and adaptability, but domain fit for illustration products is indirect (Lucas 2025).”

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

Classified as probably not a substantial claim.

Creativity development in illustration design depends on both exploratory generation and reflective evaluation.

“Illustration design is a studio discipline in which students translate a brief into visual narratives through cycles of ideation, selection, critique, and refinement, so creativity development depends on both exploratory generation and reflective evaluation (Karadağ 2025).”

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

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

These levels require separate measurement because students may value GenAI highly while producing generic artefacts, or may generate many visual variants without developing a coherent final illustration.

“These levels require separate measurement because students may value GenAI highly while producing generic artefacts, or may generate many visual variants without developing a coherent final illustration.”

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

A studio workflow should use GenAI to expand starting points without replacing critique, manual revision, or expert evaluation.

“These findings support a studio workflow in which GenAI expands possible starting points but does not replace critique, manual revision, or expert evaluation of the final illustration artefact.”

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

Prompt engineering is linked to creative design learning when students connect spatial language, conceptual intent, and visual form.

“In architectural education, prompt engineering has been linked to creative design learning when students connect spatial language, conceptual intent, and visual form through learner-centred activities (Lee and Kang 2025).”

• cites Lee and Kang 2025Not matched

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

May combine two claims.

Prior creative-course implementations rely heavily on perceived creativity and self-reported outcomes, so this study prioritizes expert-rated product creativity.

“Prior creative-course implementations often report creativity-related benefits but rely heavily on perceived creativity and other self-reported outcomes, so this study distinguishes subjective perceptions from expert-rated product creativity and prioritizes rubric-scored artefacts as the primary endpoint (Lee 2025).”

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

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

Creativity measurement should remain artefact-centred and rater-based when the aim concerns final design-product quality.

“Creativity measurement therefore needs to remain artefact-centred and rater-based when the research aim concerns final design-product quality, while process logs provide supporting evidence of how the artefact was developed.”

Classified as probably not a substantial claim.

Midjourney is most pedagogically useful during early ideation for generating variants for critique.

“In illustration education, this means that Midjourney is most pedagogically useful during early ideation, where students can generate motif, composition, atmosphere, and style variants for critique.”

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

A quantitative tool comparison shows strong symmetry but weaker rhythm and contrast in AI outputs.

“A quantitative tool comparison rates outputs across design elements and principles, showing strong symmetry but weaker rhythm and contrast, implying that critique prompts and rubric emphasis should target semantic intent rather than formal polish, though classroom inference remains indirect (Bal et al. 2025).”

• cites Bal et al. 2025Not matched

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

May combine two claims.

Experimental and quasi-experimental findings report moderate learning benefits from GenAI interventions, especially for behavioural engagement and self-regulation, but heterogeneity leaves expert-rated studio creativity insufficiently resolved.

“Experimental and quasi-experimental findings similarly report moderate learning benefits from GenAI interventions, especially for behavioural engagement and self-regulation, although heterogeneity across context, duration, and task type leaves expert-rated studio creativity insufficiently resolved (Han et al. 2025).”

• cites Han et al. 2025 sourceNot checked — no full text of the source

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

A sociomaterial qualitative study highlights embodied process consistency and emotional regulation as prerequisites for meaningful digital integration.

“A sociomaterial qualitative study of hybrid studios highlights embodied process consistency and emotional regulation as prerequisites for meaningful digital integration (Perold-Bull 2025).”

• cites Perold-Bull 2025 sourceNot checked — no full text of the source

May combine two claims.

A controlled experiment on AIGC in virtual education reports gains in engagement, retention, and problem solving, but the task ecology differs from illustration authorship and critique workflows.

“A controlled experiment on Artificial Intelligence Generated Content (AIGC) in virtual education reports gains in engagement, retention, and problem solving, but the task ecology differs from illustration authorship and critique workflows (Zhang et al. 2025).”

• cites Zhang et al. 2025 sourceNot checked — no full text of the source

May combine two claims.

Teaching Factory evidence links AI-aided workflows to industry readiness, yet lacks controlled comparisons and artefact creativity scoring.

“Teaching Factory (TEFA) evidence in vocational design links Artificial Intelligence aided workflows to industry readiness, yet lacks controlled comparisons and artefact creativity scoring (Khair and Astuti 2025).”

• cites Khair and Astuti 2025 sourceNot checked — no full text of the source

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

These criteria are especially important where generated images may display high surface polish while retaining generic concepts or weak brief alignment.

“These criteria are especially important where generated images may display high surface polish while retaining generic concepts or weak brief alignment.”

May combine two claims.

Together, these studies suggest critique-centered routines matter, yet causal evidence remains weak due to non-experimental designs and limited product creativity measurement.

“Together, these studies suggest critique centered routines matter, yet causal evidence remains weak due to non-experimental designs and limited product creativity measurement.”

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

Assessment of GenAI-supported illustration requires separation between final artefact quality, process evidence, and learner perception.

“GenAI-supported illustration assessment needs a clear separation between final artefact quality, process evidence, and learner perception.”

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

A theoretical framework proposes evaluation grids for conceptual depth, ethics, and process transparency, but lacks classroom trials and external validation.

“A theoretical framework analysing rubrics proposes evaluation grids for conceptual depth, ethics, and process transparency, but lacks classroom trials and external validation (Sirb and Petrovici 2025).”

• cites Sirb and Petrovici 2025 sourceNot checked — no full text of the source

Classified as probably not a substantial claim.

It remains uncertain whether a Midjourney-supported early-stage ideation workflow improves expert-rated creativity compared with a conventional workflow in undergraduate illustration design.

“In undergraduate illustration design classes, it remains uncertain whether a Midjourney-supported early-stage ideation workflow improves expert-rated creativity for a shared studio brief compared with a conventional workflow based on manual ideation and non-generative reference gathering under explicit authorship rules.”

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

A studio intervention reports improved visual literacy and awareness of algorithmic aesthetics, but evidence is interpretive and not linked to expert-rated product outcomes.

“A studio intervention using classroom observation, discussion, and reflective essays reports improved visual literacy and awareness of algorithmic aesthetics, but evidence is interpretive and not linked to expert-rated product outcomes (Källström 2025).”

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

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

The central gap is a controlled studio illustration test that links a specified text-to-image pedagogy to expert-rated product creativity, with authorship rules and process logs as necessary conditions.

“Overall, the central gap is a controlled studio illustration test that links a specified text to image pedagogy to expert-rated product creativity, with authorship rules and process logs as necessary conditions.”

Classified as probably not a substantial claim.

Design principles for text-to-image creativity-support tools emphasise exploration, comparison, controllability, and evaluation.

“Design principles for text-to-image creativity-support tools emphasise exploration, comparison, controllability, and evaluation as core requirements for visual design support, which aligns with studio routines where students test alternatives before committing to a final visual direction (Herath et al. 2026).”

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

May combine two claims.

AI image generation in adjacent visual fields supports the pedagogical value of guided ideation, but the available evidence still leaves undergraduate illustration-product creativity under-tested.

“AI image generation in adjacent visual fields supports the pedagogical value of guided ideation, but the available evidence still leaves undergraduate illustration-product creativity under-tested.”

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

The workflow shifts part of the creative process toward prompting, curation, selection, and revision, making process documentation essential.

“The same workflow also shifts part of the creative process toward prompting, curation, selection, and revision, making process documentation essential for distinguishing deliberate design reasoning from passive output adoption.”

May combine two claims.

A structured Midjourney-supported ideation workflow improved expert-rated illustration creativity while increasing documented ideation activity and reducing homogenisation indicators.

“A structured Midjourney-supported ideation workflow improved expert-rated illustration creativity while increasing documented ideation activity and reducing homogenisation indicators.”

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

In architectural education, prompt engineering has supported creative design learning by linking language, spatial reasoning, and visual form, although architectural tasks follow a different studio logic.

“In architectural education, prompt engineering has supported creative design learning by linking language, spatial reasoning, and visual form, although architectural design tasks follow a different studio logic from illustration because they prioritize structure, spatial organization, and environmental constraints (Lee and Kang 2025).”

• cites Lee and Kang 2025Not matched

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

May combine two claims.

Limited AI literacy and weak institutional support can undermine ethical and consistent classroom integration.

“These benefits also depend on readiness conditions, since limited AI literacy and weak institutional support can undermine ethical and consistent classroom integration, especially where process transparency is required for judging creative development (Ceallaigh et al. 2025).”

• cites Ceallaigh et al. 2025 sourceNot checked — no full text of the source

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

These findings support a controlled illustration-specific test in which Midjourney-assisted ideation is evaluated through final artefacts rated for originality, appropriateness, and elaboration.

“These findings support a controlled illustration-specific test in which Midjourney-assisted ideation is evaluated through final artefacts rated for originality, appropriateness, and elaboration.”

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

A correlational model links digital tool use to creative expression through creative self-efficacy, but self-report and non-experimental design constrain causal inference.

“A correlational model links digital tool use to creative expression through creative self-efficacy, but self-report and non-experimental design constrain causal inference for product creativity (Barreda-Coaquira et al. 2025).”

• cites Barreda-Coaquira et al. 2025 sourceNot checked — no full text of the source

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

A rubric development study reports strong internal consistency, yet validation targets teaching practice rather than student artefacts.

“A rubric development study using literature review, Delphi feedback, and pilot testing reports strong internal consistency, yet validation targets teaching practice rather than student artefacts and needs broader cultural testing (Yoon and Lee 2025).”

• cites Yoon and Lee 2025Not matched

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

May combine two claims.

Procedural controls (same instructor, identical weekly activity blocks, common critique prompts, anonymised submissions, blinded rating, predefined exclusion rules) reduced the risk of unmeasured contextual confounders, although intact-section allocation remained a design constraint.

“dynamics, and motivation, were treated as possible unmeasured contextual confounders.Procedural controls reduced these risks through the same instructor, identical weekly activity blocks, common critique prompts, anonymised submissions, blinded rating, and predefined exclusion rules, although intact-section allocation remained a design constraint.”

• AssumptionRests on an assumption: The two intact sections did not differ in unmeasured contextual factors (e.g., dynamics, motivation) that could affect creativity outcomes, beyond the measured baseline variables and procedural controls.

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

Adjudication was required for 23 Originality scores, 18 Appropriateness scores, and 20 Elaboration scores where the two raters differed by more than one scale point.

“Adjudication was required for 23 Originality scores, 18 Appropriateness scores, and 20 Elaboration scores where the two raters differed by more than one scale point.”

Classified as probably not a substantial claim.

Baseline creativity was captured through a short pre-task scored with the same expert rubric prior to workflow exposure.

“Baseline creativity was captured through a short pre-task scored with the same expert rubric prior to workflow exposure.”

May combine two claims.

Performance and contamination bias were minimised by restricting Midjourney access to the S-GCEM section during ideation weeks and requiring parallel process documentation in both sections to equalise reflective workload.

“Performance and contamination bias were minimised by restricting Midjourney access to the S GCEM section during ideation weeks and requiring parallel process documentation in both sections to equalise reflective workload.”

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

Creativity assessment remains incomplete when expert-rated artefact quality is separated from process evidence and homogenisation monitoring.

“Creativity assessment also remains incomplete when expertrated artefact quality is separated from process evidence and homogenisation monitoring.”

May combine two claims.

Drawing practice, weekly studio hours, prior GenAI exposure, baseline task completion time, and attendance rate showed no statistically significant condition differences, indicating similar measured starting profiles before workflow exposure.

“Drawing practice, weekly studio hours, prior GenAI exposure, baseline task completion time, and attendance rate showed no statistically significant condition differences, indicating similar measured starting profiles before workflow exposure.”

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

Final submissions require student-directed selection, refinement, visual integration, and documented design rationale to preserve assessable student authorship.

“Final submissions require student-directed selection, refinement, visual integration, and documented design rationale, allowing GenAI-supported exploration to remain visible within the studio process while preserving assessable student authorship.”

May combine two claims.

Study participants exhibited comparable prior experience, with overlapping distributions of drawing practice years and weekly studio hours across groups, indicating balanced experiential backgrounds before intervention.

“Study participants exhibited comparable prior experience, with overlapping distributions of drawing practice years and weekly studio hours across groups, indicating balanced experiential backgrounds before intervention (Figure S2A).”

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

Because the process indicators were partly structured by protocol requirements and documentation format, indirect-effect estimates were interpreted as exploratory process-accounting results rather than causal mediation evidence.

“Because these process indicators were partly structured by protocol requirements and documentation format, indirect-effect estimates were interpreted as exploratory process-accounting results rather than causal mediation evidence.”

May combine two claims.

The hypotheses were supported at the level of observed group differences and process-accounting evidence, while causal claims about internal mechanisms remain bounded by the intact-section design and the structured nature of the process measures.

“The hypotheses were therefore supported at the level of observed group differences and process-accounting evidence, while causal claims about internal mechanisms remain bounded by the intact-section design and the structured nature of the process measures.”

• AssumptionRests on an assumption: The two intact sections did not differ in unmeasured contextual factors (e.g., dynamics, motivation) that could affect creativity outcomes, beyond the measured baseline variables and procedural controls.

May combine two claims.

Inter-coder reliability was acceptable for Ideation Breadth (ICC = 0.91) and Iteration Depth (ICC = 0.89).

“Inter-coder reliability was acceptable for Ideation Breadth, ICC = 0.91, and Iteration Depth, ICC = 0.89.”

May combine two claims.

The condition exerts both direct and indirect influences on final creative outcomes through distinct process pathways, with a direct standardised effect of β = 0.14 on creativity.

“Evidence from the mediation model indicates that the condition exerts both direct and indirect influences on final creative outcomes through distinct process pathways, with a direct standardised effect of β = 0.14 on creativity (Figure S3).”

May combine two claims.

The final analytic sample comprised 81 students per condition and 162 students in total after exclusions.

“leaving 81 students per condition and 162 students in the final analytic sample.”

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

GenAI-supported creative work can produce trade-offs between efficiency and authorship, including fixation, stylistic convergence, weaker personal-style articulation, and reduced control over final visual decisions.

“Second, GenAI-supported creative work can produce trade-offs between efficiency and authorship, including fixation, stylistic convergence, weaker personal-style articulation, and reduced control over final visual decisions.”

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

The study restricts Midjourney to early ideation, requires prompt and revision logs, and monitors homogenisation as a secondary risk indicator.

“The present study therefore restricts Midjourney to early ideation, requires prompt and revision logs, and monitors homogenisation as a secondary risk indicator alongside expert-rated creativity.”

May combine two claims.

Both sections completed the same illustration brief, received the same critique schedule, and submitted identical milestone deliverables to control exposure, contact time, and feedback.

“Both sections completed the same illustration brief, received the same critique schedule, and submitted identical milestone deliverables to control exposure, contact time, and feedback.”

May combine two claims.

For RQ2, higher Ideation Breadth and Iteration Depth show that the workflow produced more documented concept exploration and revision activity, yet these indicators should be read as protocol-linked process evidence because minimum thresholds were embedded in both conditions.

“For RQ2, higher Ideation Breadth and Iteration Depth show that the workflow produced more documented concept exploration and revision activity, yet these indicators should be read as protocol-linked process evidence because minimum thresholds were embedded in both conditions.”

• AssumptionRests on an assumption: The process indicators (Ideation Breadth and Iteration Depth) validly capture the underlying ideation breadth and iteration depth, despite being partly structured by protocol requirements.

May combine two claims.

Across conditions, baseline creativity profiles were closely aligned, with comparable mean levels across all four dimensions, indicating equivalent starting points.

“Across conditions, baseline creativity profiles were closely aligned, with comparable mean levels across all four dimensions, indicating equivalent starting points for subsequent comparisons (Figure S1A).”

May combine two claims.

The direction and magnitude of the condition effect in the protocol-compliant subset were consistent with the primary ANCOVA model, indicating that the condition effect was not driven by participants with incomplete or non-compliant documentation.

“The direction and magnitude were consistent with the primary ANCOVA model, indicating that the condition effect was not driven by participants with incomplete or non-compliant documentation.”

• AssumptionRests on an assumption: The final analytic sample is not biased by the exclusion of participants with missing documentation, missing artefacts, or protocol non-compliance.

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

Two independent coders extracted Ideation Breadth and Iteration Depth from anonymised process logs, with disagreements resolved through consensus.

“Two independent coders extracted Ideation Breadth and Iteration Depth from anonymised process logs, with disagreements resolved through consensus.”

May combine two claims.

A systematic review of visual arts assessment emphasises that structured rubrics plus iterative feedback support originality and confidence, but tool-specific mechanisms are not isolated.

“A systematic review of visual arts assessment emphasises that structured rubrics plus iterative feedback support originality and confidence, but tool specific mechanisms are not isolated and effect estimates remain indirect (Farahie and Sihes 2024).”

• cites Farahie and Sihes 2024Not matched

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

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

Distilled requirements are expert raters with reliability checks, rubric validity aligned to illustration briefs, and process trace capture of ideation and iteration.

“Distilled requirements are expert raters with reliability checks, rubric validity aligned to illustration briefs, and process trace capture of ideation and iteration.”

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

Creativity assessment in GenAI-supported illustration requires a structure that connects product ratings with process evidence and homogenisation monitoring.

“Third, creativity assessment in GenAI-supported illustration requires a structure that connects product ratings with process evidence and homogenisation monitoring.”

May combine two claims.

Systematic reviews and framework syntheses highlight minimal requirements for studio-grade measurement, but do not fully address GenAI-supported workflows.

“Systematic reviews and framework syntheses highlight minimal requirements for studio grade measurement, but do not fully address Generative Artificial Intelligence supported workflows.”

May combine two claims.

Prior generative artificial intelligence exposure was similarly distributed between groups, reducing confounding from differential familiarity.

“Prior generative artificial intelligence exposure was similarly distributed between groups, reducing confounding from differential familiarity (Figure S1D).”

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

Midjourney-supported illustration pedagogy remains underspecified when early ideation, authorship rules, and critique routines are weakly connected to measurable studio outcomes.

“Midjourney-supported illustration pedagogy remains underspecified when early ideation, authorship rules, and critique routines are weakly connected to measurable studio outcomes.”

May combine two claims.

The Iteration Depth pathway contributes a smaller but meaningful indirect effect, linking Condition to Depth (β = 0.27) and Depth to Creativity (β = 0.18), supporting parallel process mediation within the model structure.

“The Iteration Depth pathway contributes a smaller but meaningful indirect effect, linking Condition to Depth (β = 0.27) and Depth to Creativity (β = 0.18), supporting parallel process mediation within the model structure (Figure S3).”

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

GenAI has become pedagogically important in art, design, and higher education because it can expand ideation, strengthen prompt literacy, and accelerate visual experimentation.

“GenAI has become pedagogically important in art, design, and higher education because it can expand ideation, strengthen prompt literacy, and accelerate visual experimentation.”

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

An ADDIE-based guideline shows high content validity and pre-post improvement, supporting process-based scoring, but does not define enforceable authorship boundaries.

“An Analysis, Design, Development, Implementation, and Evaluation (ADDIE) based guideline aligned to Design Thinking shows high content validity and pre post improvement, supporting process based scoring of ideation and iteration, but does not define enforceable authorship boundaries for external generators (Widyastuti and Jayanta 2024).”

• cites Widyastuti and Jayanta 2024 sourceNot checked — no full text of the source

May combine two claims.

Inter-rater reliability was acceptable for Originality (ICC = 0.87), Appropriateness (ICC = 0.84), Elaboration (ICC = 0.86), and the composite Product Creativity score (ICC = 0.90).

“Inter-rater reliability was acceptable for Originality, ICC = 0.87; Appropriateness, ICC = 0.84; Elaboration, ICC = 0.86; and the composite Product Creativity score, ICC = 0.90.”

• Also stated in: § 3.2 | Primary Outcomes · plain

  “The intraclass correlation coefficient was 0.87 for Originality, 0.84 for Appropriateness, 0.86 for Elaboration, and 0.90 for the composite Product Creativity score (Table S1).”

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

Two blinded raters scored all artefacts independently after a 60 min calibration session using practice artefacts and agreed anchor descriptors, with a third rater adjudicating only when a dimension differed by more than 1 point.

“Two blinded raters scored all artefacts independently after a 60 min calibration session using practice artefacts and agreed anchor descriptors, with a third rater adjudicating only when a dimension differed by more than 1 point.”

May combine two claims.

Baseline task completion times followed nearly identical density profiles between conditions, suggesting equivalent task engagement and pacing at study entry.

“Baseline task completion times followed nearly identical density profiles between conditions, with closely aligned weighted mean references, suggesting equivalent task engagement and pacing at study entry (Figure S2B).”

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

Ideation Breadth and Iteration Depth were treated as protocol-linked process indicators rather than independent latent mechanisms because both were partly shaped by required documentation thresholds.

“Ideation Breadth and Iteration Depth were treated as protocollinked process indicators rather than independent latent mechanisms because both were partly shaped by required documentation thresholds.”

• AssumptionRests on an assumption: The process indicators (Ideation Breadth and Iteration Depth) validly capture the underlying ideation breadth and iteration depth, despite being partly structured by protocol requirements.

May combine two claims.

The study applies an authorship and allowable-use policy where Midjourney supports early ideation but final submissions must demonstrate student-directed refinement and include structured process documentation.

“To prevent assessment ambiguity, the study applies an authorship and allowable-use policy in which Midjourney may support early ideation and exploratory variants, while final submissions must demonstrate studentdirected refinement and include structured process documentation that links prompts, selections, and revisions to design rationales (Arosio 2025).”

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

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

Many studies prioritize attitudes, engagement, case-based implementation, or design-process reflection, leaving controlled evidence on expert-rated product creativity comparatively limited.

“First, many studies prioritize attitudes, engagement, case-based implementation, or design-process reflection, leaving controlled evidence on expert-rated product creativity comparatively limited.”

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

Both sections used the same brief, studio timetable, and submission platform, with access to Midjourney provided only to the intervention section during the ideation phase.

“Both sections used the same brief, studio timetable, and submission platform, with access to Midjourney provided only to the intervention section during the ideation phase.”

May combine two claims.

The S-GCEM section and the conventional section were matched on brief, duration, contact hours, critique schedule, submission milestones, file specifications, and assessment rubric, while baseline creativity, drawing experience, weekly studio practice, prior GenAI exposure, and AI literacy were recorded before workflow exposure and used to assess starting equivalence.

“The S-GCEM section and the conventional section were matched on brief, duration, contact hours, critique schedule, submission milestones, file specifications, and assessment rubric, while baseline creativity, drawing experience, weekly studio practice, prior GenAI exposure, and AI literacy were recorded before workflow exposure and used to assess starting equivalence.”

May combine two claims.

Attrition bias was minimised by enforcing identical submission checkpoints and applying predefined exclusion rules for missing artefacts or incomplete logs.

“Attrition bias was minimised by enforcing identical submission checkpoints and applying predefined exclusion rules for missing artefacts or incomplete logs.”

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

Authorship rules limited Midjourney use to early-stage ideation and exploratory variation, prohibited direct submission of generated images as final artefacts, and required student-directed transformation demonstrated through intermediate drafts and revision notes.

“Authorship rules limited Midjourney use to early-stage ideation and exploratory variation, prohibited direct submission of generated images as final artefacts, and required student-directed transformation demonstrated through intermediate drafts and revision notes.”

Classified as probably not a substantial claim.

Midjourney is examined through user-facing functionality, not internal model details, to focus measurement on controllable inputs and observable outputs.

“Midjourney is examined through user-facing functionality, namely prompt-based generation of multiple variants and iterative selection and modification, while internal proprietary model details are not assumed, allowing measurement to focus on controllable inputs, observable outputs, and studio routines (Morgan et al. 2025).”

• cites Morgan 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.

The central gap is an integrated measurement approach that combines expert-rated product creativity with traceable ideation and iteration metrics and enforceable authorship boundaries.

“Overall, the central gap is an integrated measurement approach that combines expert rated product creativity with traceable ideation and iteration metrics and enforceable authorship boundaries under GenAI supported ideation.”

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

The study places expert-rated illustration products as the primary outcome and treats ideation breadth, iteration depth, and homogenisation as supporting indicators.

“For this reason, the present study places expert-rated illustration products as the primary outcome and treats ideation breadth, iteration depth, and homogenisation as supporting process and risk indicators.”

Classified as probably not a substantial claim.

Section-level influences were treated as possible unmeasured contextual confounders.

“Section-level influences, including class culture, peer interaction, teacher-student dynamics, and motivation, were treated as possible unmeasured contextual confounders.”

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

A review of design sketching assessment proposes the 3pDS Framework, but calls for reliability testing beyond novice cohorts.

“A review of design sketching assessment proposes the Three Pronged Design Sketching (3 pDS) Framework integrating communication, visual spatial characteristics, and creativity, but calls for reliability testing beyond novice cohorts and across artefact genres (Ghahremani et al. 2024).”

• cites Ghahremani et al. 2024 sourceNot checked — no full text of the source

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

The control section used the same weeks, contact time, critique schedule, and milestone deadlines, differing only in ideation tool use.

“The control section used the same weeks, contact time, critique schedule, and milestone deadlines, differing only in ideation tool use.”

Classified as probably not a substantial claim.

Fidelity was monitored using an instructor checklist completed at each session, verifying delivery of the planned activity blocks, critique checkpoints, and documentation requirements for both sections.

“Fidelity was monitored using an instructor checklist completed at each session, verifying delivery of the planned activity blocks, critique checkpoints, and documentation requirements for both sections.”

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

Structured rubrics combined with iterative feedback improve originality and creative outcomes in visual arts assessment.

“Evidence that structured rubrics combined with iterative feedback improve originality and creative outcomes in visual arts assessment likewise converges with the observed breadth advantage under S GCEM, because structured guidance can expand exploration while keeping outputs evaluable against shared criteria (Farahie and Sihes 2024).”

• cites Farahie and Sihes 2024Not matched

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

Classified as probably not a substantial claim.

The finding does not mean that more generated images are inherently more creative.

“The finding does not mean that more generated images are inherently more creative.”

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

A multi-university classroom experiment reported increased idea diversity and visual experimentation under text-to-image ideation support.

“A multiuniversity classroom experiment likewise reported increased idea diversity and visual experimentation under text-to-image ideation support, aligning with the present pattern that structured ideation workflows can elevate rated creative outcomes (Morgan et al. 2025).”

• cites Morgan 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.

The S-GCEM requirements can help instructors preserve assessment validity while allowing students to use GenAI for exploratory ideation.

“These requirements can help instructors preserve assessment validity while allowing students to use GenAI for exploratory ideation.”

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

The outcome layer is sensitive to this expansion because originality scoring depends on selecting a genuinely differentiated direction and elaborating it coherently through critique-informed refinement.

“The outcome layer is sensitive to this expansion because originality scoring depends on selecting a genuinely differentiated direction and elaborating it coherently through critique-informed refinement.”

May combine two claims.

The effect size remained modest, so the finding should not be read as evidence that Midjourney use alone produces major creativity gains.

“The effect size remained modest, so the finding should not be read as evidence that Midjourney use alone produces major creativity gains.”

Classified as probably not a substantial claim.

Practically, S-GCEM provides a usable studio protocol for controlled Midjourney integration.

“Practically, S-GCEM provides a usable studio protocol for controlled Midjourney integration.”

Classified as probably not a substantial claim.

The contribution is methodological and pedagogical: the study provides a structured model for assessing GenAI-supported illustration work without collapsing creativity evaluation into tool use, engagement, or self-reported usefulness.

“The contribution is therefore methodological and pedagogical: the study provides a structured model for assessing GenAI-supported illustration work without collapsing creativity evaluation into tool use, engagement, or self-reported usefulness.”

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

A studio-based Midjourney intervention described how iterative prompting aligned with critique and reflection made student processes more systematic and elaborative.

“A studio-based Midjourney intervention described how iterative prompting aligned with critique and reflection made student processes more systematic and elaborative, which matches the present pattern that deeper iteration can translate into stronger refined outputs (Källström 2025).”

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

May combine two claims.

The result supports the value of structured revision logs, while leaving open whether the same depth would appear without mandatory documentation thresholds.

“The result therefore supports the value of structured revision logs, while leaving open whether the same depth would appear without mandatory documentation thresholds.”

Classified as probably not a substantial claim.

Breadth became meaningful in this study because each direction required a label, visual intention, supporting artefacts, and later selection rationale.

“Breadth became meaningful in this study because each direction required a label, visual intention, supporting artefacts, and later selection rationale.”

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

Structured rubrics and iterative feedback can support originality when students understand the criteria used to judge creative work.

“It also aligns with visual arts assessment research showing that structured rubrics and iterative feedback can support originality when students understand the criteria used to judge creative work (Farahie and Sihes 2024).”

• cites Farahie and Sihes 2024Not matched

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

Classified as probably not a substantial claim.

Design-sketching assessment work treats representational diversity and iteration as important evidence of creative development.

“This interpretation is consistent with design-sketching assessment work that treats representational diversity and iteration as important evidence of creative development (Ghahremani et al. 2024).”

• cites Ghahremani et al. 2024 sourceNot checked — no full text of the source

May combine two claims.

Within S-GCEM, the pedagogy layer constrains tool use to early ideation while preserving critique-centered learning, making the creativity gains plausible as improvements in concept selection and refinement rather than simple output substitution.

“Within S GCEM, the pedagogy layer constrains tool use to early ideation while preserving critiquecentered learning, making the creativity gains plausible as improvements in concept selection and refinement rather than simple output substitution.”

May combine two claims.

Equivalent counting thresholds reduced documentation imbalance, yet the process indicators still partly reflect workflow structure.

“Equivalent counting thresholds reduced documentation imbalance, yet the process indicators still partly reflect workflow structure.”

May combine two claims.

The present finding indicates that homogenisation risk can be reduced when originality is explicitly scored, generated images are confined to early ideation, and final artefacts must show student-directed transformation.

“The present finding indicates that homogenisation risk can be reduced when originality is explicitly scored, generated images are confined to early ideation, and final artefacts must show student-directed transformation.”

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

The same policy should avoid open-ended permission for final-image generation because the present evidence supports structured ideation use, not unrestricted GenAI production.

“The same policy should avoid open-ended permission for final-image generation because the present evidence supports structured ideation use, not unrestricted GenAI production.”

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

A design-based pilot reported a 27% increase in creativity scores when generative AI was embedded into coursework.

“In a designbased pilot, Lee (2025) reported a 27% increase in creativity scores when generative artificial intelligence was embedded into coursework, which is consistent with the direction of the adjusted Product Creativity advantage observed here.”

• cites Lee (2025) sourceNot checked — no full text of the source

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

Theoretically, S-GCEM contributes a bounded account of GenAI-supported studio creativity by separating tool exposure from structured pedagogy, process evidence, and artefact quality.

“Theoretically, S-GCEM contributes a bounded account of GenAI-supported studio creativity by separating tool exposure from structured pedagogy, process evidence, and artefact quality.”

May combine two claims.

The study contributes to GenAI-supported design education by showing that text-to-image tools can be incorporated into studio pedagogy when their use is restricted to early ideation, documented through process logs, and evaluated through final student-produced artefacts.

“The study contributes to GenAIsupported design education by showing that text-to-image tools can be incorporated into studio pedagogy when their use is restricted to early ideation, documented through process logs, and evaluated through final student-produced artefacts.”

May combine two claims.

The use of the same brief, instructor, timetable, critique schedule, anonymised scoring, and baseline adjustment reduced the risk of section-level confounding, but individual randomisation was not used.

“The use of the same brief, instructor, timetable, critique schedule, anonymised scoring, and baseline adjustment reduced this risk, but individual randomisation was not used.”

Classified as probably not a substantial claim.

The outcome layer captures these improvements because elaboration is sensitive to coherence, purposeful detail, and integration that typically emerge through successive revisions.

“The outcome layer then captures these improvements because elaboration is sensitive to coherence, purposeful detail, and integration that typically emerge through successive revisions.”

Classified as probably not a substantial claim.

The process layer operationalises the S-GCEM structure through logged revision cycles, making iteration depth a tractable pathway that connects ideation work to outcome scoring.

“The process layer operationalises this through logged revision cycles, making iteration depth a tractable pathway that connects ideation work to outcome scoring.”

May combine two claims.

S-GCEM contributes a practical safeguard: GenAI use is allowed, but similarity risk is measured instead of assumed away.

“S-GCEM therefore contributes a practical safeguard: GenAI use is allowed, but similarity risk is measured instead of assumed away.”

Classified as probably not a substantial claim.

The results support departmental policies that permit text-to-image tools under documented, phase-specific, and assessable conditions.

“The results support departmental policies that permit text-toimage tools under documented, phase-specific, and assessable conditions.”

May combine two claims.

Similar conditions are likely necessary for transfer to other studio courses because unguided text-to-image use can increase output quantity without strengthening design reasoning or final artefact quality.

“Similar conditions are likely necessary for transfer to other studio courses because unguided textto-image use can increase output quantity without strengthening design reasoning or final artefact quality.”

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

The findings may not generalise to longer portfolio development, advanced studio courses, different illustration genres, or other GenAI systems.

“The findings may not generalise to longer portfolio development, advanced studio courses, different illustration genres, or other GenAI systems.”

May combine two claims.

Ideation Breadth and Iteration Depth helped explain the observed creativity pattern as protocol-linked process indicators, while lower homogenisation scores suggest that originality criteria and critique documentation can reduce similarity concentration across student outputs.

“Ideation Breadth and Iteration Depth helped explain the observed creativity pattern as protocol-linked process indicators, while lower homogenisation scores suggest that originality criteria and critique documentation can reduce similarity concentration across student outputs.”

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

Iteration Depth should be interpreted as evidence of documented refinement activity, not as direct proof of deeper cognition.

“Iteration Depth should be interpreted as evidence of documented refinement activity, not as direct proof of deeper cognition.”

Classified as probably not a substantial claim.

The process layer transforms the S-GCEM structure into distinct, documentable concept clusters, which supports valid differentiation between alternative directions instead of superficial variants.

“The process layer transforms this into distinct, documentable concept clusters, which supports valid differentiation between alternative directions instead of superficial variants.”

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

The study used one undergraduate illustration course, one shared brief, one six-week implementation window, and one text-to-image platform configuration.

“Third, the study used one undergraduate illustration course, one shared brief, one six-week implementation window, and one text-to-image platform configuration.”

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

Within S-GCEM, the pedagogy layer sets explicit constraints that protect autonomy while preventing unbounded prompt wandering, so breadth reflects deliberate exploration rather than noise.

“Within S GCEM, the pedagogy layer sets explicit constraints that protect autonomy while preventing unbounded prompt wandering, so breadth reflects deliberate exploration rather than noise.”

May combine two claims.

The findings support a model in which text-to-image GenAI expands early visual search, critique converts alternatives into evaluated decisions, and final creativity depends on student-led selection, refinement, and integration.

“The findings support a model in which text-to-image GenAI expands early visual search, critique converts alternatives into evaluated decisions, and final creativity depends on student-led selection, refinement, and integration.”