Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Review on the article titled “AI-Driven Parametric Architecture as an Approach for Computational Design Innovation”

I enjoyed reading the article. My main comments are as follows:

Introduction

Comment one: In lines 39 and 40, the authors mention that there is a lack of studies evaluating the impact of AI on architectural education. The authors should strengthen this claim by providing additional references. The authors should support this statement by reviewing the existing literature (see lines 39–40).

Comment two: The authors define the following two research questions in the introduction section:

1) To what extent can AI support students in employing 2D and 3D modeling to enhance architectural parametric design workflows?

2) How can comparable AI-assisted pedagogical approaches be effectively integrated into real-world architectural projects and professional design practice?

The authors mentioned that, to respond to and address these questions, a workshop was organized. The authors should provide an explanation of why a workshop was organized as a methodological tool to address the mentioned research questions. In which ways is a workshop an appropriate methodological tool? (see lines 46 to 53).

Comment three: Please double-check the names of the AI models used, such as “Claude MCP”. It should be replaced with ‘Claude’ (see lines 53 and 54).

Comment four: In the introduction, it is important to highlight the contribution of this research to the topic of AI in architectural education. In which ways does this research contribute to the subject?

Literature review

Comment one: In lines 70 to 73, the authors mentioned the following: “AI-powered technologies such as Building Information Modeling (BIM), parametric design, and generative systems enable students to work with complicated spatial data and design procedures that are based on performance, which makes both decision-making and efficiency better.” I am not sure if we can classify BIM software such as Revit Architecture as AI tools. BIM software has an extensive library that can incorporate predefined items into the design; such software should not be considered AI-based.

Comment two: I entirely agree with the authors that AI is becoming a common tool utilized in architecture and interior design studios. As a lecturer who has been involved in teaching design studios, I agree that AI is becoming part of design studios and education in a wider sense.

Comment three: In Section 2.2, the authors mentioned the shortcomings of AI-based tools as “machine hallucinations.” The authors should widen this discussion. In fact, “machine hallucinations” can be considered an AI weakness or shortcoming (see lines 114 to 117).

Comment four: The authors, in Section 2.4, mentioned that “short-format workshops have proven to be a particularly useful lens for studying how designers respond to new AI tools.” Is this claim based on the authors’ own observations and conclusions, or is it based on previous studies? Please discuss this further in detail (see lines 165–166).

Comment five: Please provide references for the following statement: “Comparable findings have emerged from studio-based research, where participants working with AI image generation and parametric scripting reported a more exploratory approach to early-stage design, moving between ideas more freely and revisiting assumptions more often than they might have otherwise.” (see lines 174-177).

Comment six: In section 2.4, the authors mentioned the following: “Methodologically, these studies typically utilize a combination of observation, questionnaires, and interviews.” Please provide a more in-depth discussion regarding the reviewed literature. 1) How many papers were reviewed? 2) What were the topics of the reviewed articles? 3) What methodologies were used? The authors should provide a more systematic and in-depth review of the literature.

Workshop Methodology

Comment one: I am wondering how the authors managed to cover the foundations of Rhino/Grasshopper and AI-assisted concept generation in a 5-day workshop. In 5 days, perhaps only the basics of one software, such as Rhino, could be taught. Perhaps the students were already familiar with Rhino/Grasshopper and AI tools and attended the workshop mainly for experimentation. Please clarify the following: 1) How many students attended? 2) Which year and program were they in? Were they undergraduate or graduate students? 3) What was their educational background? Were all of them familiar with Rhino/Grasshopper? 4) Why were Rhino/Grasshopper selected as tools? 5) VR integration is another topic. Were the students familiar with VR? Could all these topics be taught in 5 days? Please explain in detail (see lines 201 to 205).

Comment two: Research ethics should be considered. All participants in this research should provide a written statement that they agree to participate in this research. Since this research involves human participants, the authors should provide signed consent forms to the journal. The ethics board of the university where the study has been conducted should issue the approval letter. The ethics board approval letter should also be submitted to the journal (see lines 227 to 232).

Comment three: The actual research methodology is missing in Section 3. Workshop Methodology. In the methodology section, the authors should describe how data were collected. Please provide a detailed explanation of how data were collected. Which methodological tools have been used to collect data (survey, interviews, direct observation)? The authors should explain how data were analyzed?

Workshop Methods

Comment one: The workshop’s name “Sustainable Urban Solutions” is irrelevant to the workshop’s content. The five-day workshop was focusing on AI in architectural education, while the name of the workshop suggests another topic. Please justify it.

Comment two: The authors defined three main dimensions within the architecture design process. “The research focused on three main dimensions in the architectural design process: (1) Conceptual design dimension: for generative ideation, (2) Design development dimension: for modeling and scripting and geometry generation in Rhino/Grasshopper, and (3) Rendering dimension: for enhancing presentation and visualization outputs.” (see lines 234 to 240).

As authors mentioned, the process of design starts with a series of artistic keywords and emotional cues (see lines 255 to 260). In the next phase, the images are enhanced and turned into architectural forms. The authors called it “Architecturalization Phase” (see lines 262–267). In the third phase, the generated images are compared against utility, durability, and feasibility. The generated images should have a balance between artistic vision and architectural logic (see lines 269–275). The mentioned phases should be defined more clearly. The steps should be clarified in detail so other researchers could conduct the same study in their workshops or studios. The mentioned steps as they are cannot be replicated in similar studies. Please define the term “Architecturalization”.

Case study: Workshop execution

Comment one: Authors mentioned that the workshop brought various participants. Please highlight how many people participated in the workshop, how many were architecture students, how many were senior students, and how many were graduate students. What was their level of program proficiency at each level? How did the workshop lecturers manage to teach Rhino/Grasshopper software to this range of varied groups with different backgrounds in 5 days?

Comment two: The authors provided a descriptive account of what happened during each day of the workshop. The authors should mention how each group of participants responded to each tool. The authors should mention the issues or difficulties that occurred during each phase. The authors should mention how they managed to resolve the issues. Even if all the participants from the first/second year of architecture to near graduates knew Rhino/Grasshopper and AI tools, it is still a challenge to teach and combine all these tools in one workshop.

Comment three: The authors should consider their research questions which have been mentioned earlier. In Section 5, the authors narrate the day-to-day schedule of the workshop; however, they should connect their narration to the mentioned research questions. Please reflect the research questions within the Section 5 narrative.

Comments on the Quality of English Language

The manuscript requires further editing.

Author Response

Reply to Reviewer Comments

Manuscript: "AI-Driven Parametric Architecture as an Approach for Computational Design Innovation"

We sincerely thank the Editor and all five reviewers for their careful and constructive reading of our manuscript. The comments have substantially strengthened the paper. Below we respond to each comment individually. Reviewer comments are reproduced in black, and our responses appear in blue. All corresponding changes in the manuscript are marked with yellow highlighting so they can be located easily.

Reviewer 1

We are grateful for your positive assessment and the detailed, section-by-section guidance.

Introduction

Comment 1. In lines 39–40, the authors mention a lack of studies evaluating the impact of AI on architectural education. This claim should be strengthened with additional references and a review of the existing literature.

Response: Thank you. We have moderated and better evidenced this claim. Rather than asserting an absolute absence of studies, the Introduction now states that empirical, structured evaluations remain comparatively limited relative to technical demonstrations, and we support this with additional citations to recent review and empirical work (Sadek 2023 [3]; Albukhari 2025 [4]; Jin et al. 2024 [2]) and to the systematic review by Li et al. [14]. Highlighted at lines 44–44.

Comment 2. Explain why a workshop was chosen as a methodological tool to address the research questions. In what ways is a workshop an appropriate method? (lines 46–53).

Response: We agree this rationale was missing. A new paragraph has been added at the end of the Introduction (lines 68–71) explaining that an intensive short-format workshop provides a bounded, observable setting in which the uptake of an unfamiliar AI-assisted workflow, and shifts in participants’ attitudes, can be studied within a controlled timeframe. We now explicitly frame the study as an exploratory, observational workshop evaluation rather than a controlled effectiveness trial, and cite prior workshop-based studies that use this design (Özorhon et al. [23]; Jin et al. [2]). Highlighted at lines 70–77 (workshop rationale) and 231–238 (study-design statement).

Comment 3. Double-check the AI model names, e.g. "Claude MCP" should be "Claude" (lines 53–54).

Response: Corrected. The model is now referred to simply as Claude in the Introduction (line 60). The erroneous “MCP” suffix has been removed from the model name.

Comment 4. Highlight the contribution of this research to AI in architectural education. In what ways does it contribute?

Response: A dedicated contribution statement has been added to the end of the Introduction. It clarifies that the paper’s contribution is (i) a structured, replicable three-dimension / four-phase AI-integrated parametric workflow spanning ideation, parametric development and visualization, and (ii) an exploratory account of how learners across different experience levels perceive and adopt that workflow. We are careful to frame this as a pedagogical and methodological contribution rather than a claim of proven learning gains. Highlighted at lines 70–77.

Literature Review

Comment 1. I am not sure BIM software such as Revit can be classified as an AI tool; it has an extensive library of predefined items and should not be considered AI-based (lines 70–73).

Response: We agree and have corrected this. BIM is no longer grouped with AI technologies. The sentence has been rewritten so that BIM is described as a data-rich modelling environment into which AI capabilities are increasingly being embedded, distinct from generative and parametric AI systems. Highlighted at line 89.

Comment 2. I agree that AI is becoming a common tool in architecture and interior design studios.

Response: Thank you for this supportive observation; no change was required.

Comment 3. In Section 2.2 the shortcomings of AI tools are described as "machine hallucinations." This discussion should be widened (lines 114–117).

Response: The discussion of machine hallucination has been expanded into a fuller treatment of AI limitations, now covering not only contextually inappropriate or geometrically invalid outputs but also dataset bias, lack of spatial/structural verification, and reproducibility concerns, with the implications for authorship and human oversight. Highlighted at lines 134–147.

Comment 4. The claim that "short-format workshops have proven to be a particularly useful lens..." — is this the authors’ own conclusion or based on prior studies? Please discuss (lines 165–166).

Response: Clarified. This is grounded in prior work, and we now attribute it explicitly and cite the relevant workshop/studio-based studies (Özorhon et al. [23]; Jin et al. [2]) rather than presenting it as our own assertion. Highlighted at line 194.

Comment 5. Please provide references for the statement beginning "Comparable findings have emerged from studio-based research..." (lines 174–177).

Response: Citations have been added to studio-based studies reporting more exploratory early-stage design behaviour under AI-assisted workflows ([3], [23], and [7]). Highlighted at line 203, 207.

Comment 6. In Section 2.4 the authors mention studies "typically utilize a combination of observation, questionnaires, and interviews." Provide a more systematic and in-depth review: how many papers were reviewed, their topics, and methodologies used.

Response: We have made this part of the review more systematic. It now summarizes the specific body of workshop/studio-based evaluations we draw on, characterizing their topics (AI in design studios, generative ideation, human–AI co-creation) and the methods they employ (direct observation, pre/post or retrospective surveys, and interviews or reflective accounts), and positions our own methodological choices against them. A new summary sentence indicates the number and type of studies considered. Highlighted at lines 194–203.

Workshop Methodology

Comment 1. How were Rhino/Grasshopper foundations and AI-assisted generation covered in only 5 days? Please clarify: (1) number of students; (2) year/program, undergraduate or graduate; (3) educational background and prior Rhino/Grasshopper familiarity; (4) why Rhino/Grasshopper were selected; (5) VR familiarity; and whether all topics could be taught in 5 days (lines 201–205).

Response: We have added a Participants and Setting subsection (new subsection, lines 272–285) and a participant-characteristics table that report the number of participants, their programme year/level, prior experience with Rhino/Grasshopper and AI tools, and recruitment. We clarify that Day 1 was a refresher rather than an introduction from zero — participants had prior exposure to Rhino/Grasshopper — and that VR and digital fabrication were brief demonstrations, not assessed components of the workflow. We also justify the choice of Rhino/Grasshopper (their status as the standard parametric environment in the participants’ curriculum and their compatibility with the AI plug-ins used). The participant count is highlighted as N Highlighted at lines 272–285 (new “Participants and Setting” subsection and participant-characteristics table).

Comment 2. Research ethics: participants should provide written consent; the university ethics board approval letter and signed consent forms should be submitted to the journal (lines 227–232).

Response: We take this seriously. The Informed Consent and ethics statements have been expanded to describe the consent procedure, and this was clearly shared with workshop participants. Additionally, to ensure the privacy of the participants we made all photos blur.  We shared the ethics-committee approval and consent documentation to the editorial office. And you can find the ethics approval document (expedited review; accepted decision) is provided as Appendix 2.

Comment 3. The actual research methodology is missing in Section 3. Describe how data were collected (survey, interviews, observation) and how they were analyzed.

Response: A new Data Collection and Analysis subsection has been added (lines 286–307). It specifies the two data sources (structured direct observation during sessions and a post-workshop survey), describes the survey instrument (number of items, item types, and response scales), states who collected and analyzed the data, and describes how observational notes and survey responses were handled and reported. We also now acknowledge observer-expectancy and social-desirability risks and how they were mitigated. Highlighted at lines 286–307 (new “Data Collection and Analysis” subsection and the five-day schedule table); the survey instrument is to be provided as Supplementary Material. The full instrument is provided as Supplementary Material (Appendix A).

 

Workshop Methods

Comment 1. The workshop name "Sustainable Urban Solutions" is irrelevant to a workshop on AI in architectural education. Please justify.

Response: Clarified. “smart and future cities laboratory for Sustainable Urban Solutions (SFCL)” is the name of the host laboratory under which the workshop was convened, not the topic of the workshop itself. We have reworded the sentence, so this is unambiguous and removed the implication that it describes the workshop’s content. Highlighted at line312.

Comment 2. The three design dimensions and the phases (concept extraction, "Architecturalization", evaluation) should be defined more clearly and made replicable. Please define "Architecturalization" (lines 234–275).

Response: We have rewritten these phases as explicit, numbered, replicable steps (inputs, tools, actions, and outputs for each), so that another researcher could reproduce them. We now define “Architecturalization” explicitly as the img2img-based transformation of open-ended conceptual imagery into architecturally grounded form by progressively introducing architectural references, style descriptors and constraints into the prompts. The five-day activities and outputs are additionally summarized in a new day-by-day table. The definition is highlighted at line 341, and the five-day activities are summarized in the schedule table at line 309.

Case study: Workshop execution

Comment 1. State how many people participated, how many were architecture students, senior students, and graduates, and their proficiency at each level. How were varied backgrounds taught in 5 days?

Response: These details are now reported in the new participant-characteristics table (line 285) and cross-referenced in the case study from line 451. We explain that the mixed-experience group shared a common Day-1 refresher, after which more experienced participants worked more independently while early-stage participants received closer instructor support.

Comment 2. For the day-by-day account, mention how each group responded to each tool, the difficulties in each phase, and how they were resolved.

Response: The case-study narrative (from line 451) has been revised to report, for each phase, the difficulties that arose (e.g., prompt formulation for concept extraction, reconciling AI-generated scripts with intended geometry, multi-platform switching) and how instructors addressed them. These observational findings are now clearly labelled as coming from direct observation. Highlighted from line 451.

Comment 3. Connect the Section 5 day-to-day narrative back to the stated research questions.

Response: We have added explicit links from the narrative to the two research questions — RQ1 (AI support for 2D/3D modelling in parametric workflows) is tied to the scripting and modelling days, and RQ2 (transfer to real-world/professional practice) is discussed with appropriate caution given the study’s exploratory scope. Highlighted at lines 451 - 459.

 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript addresses a timely and relevant topic — AI integration in parametric architectural design education — and presents a coherent three-dimensional workflow framework (conceptual design, parametric development, visualization). The literature review is recent and well-targeted. However, several methodological weaknesses must be addressed before the manuscript can be considered for publication.

1. Sample size and statistical representation!!! The manuscript presents percentage-based findings (e.g., "66.7%", "83.3%", "16.7%") throughout the Abstract, Results, and Conclusion, yet the exact number of participants is never explicitly stated anywhere in the text. The sample size can only be inferred mathematically from the reported percentages, which themselves reveal a very small group where each individual participant accounts for approximately 16.7% of all responses. Omitting this fundamental methodological detail while simultaneously presenting percentage values as though they carry quantitative weight is a serious lapse in transparency. The total number of participants must be stated explicitly. Furthermore, the authors must either reframe the study as a qualitative or exploratory case study and contextualize all percentage claims accordingly, or clearly acknowledge that the current findings are preliminary and not generalizeable. At minimum, every percentage must be accompanied by the corresponding absolute number (e.g., "4 out of 6 participants, 66.7%").

2. Research instrument not described: The post-workshop survey is central to the Results section, yet no description of the instrument is provided — no item count, question types, or response scales. The survey, or a structured description of it, must be included as a supplementary appendix.

3. Overlap between Sections 3 and 4: Section 3 ("Workshop Methodology") and Section 4 ("Workshop Methods") substantially overlap and their distinction is unclear. The four-phase framework and the three-dimension framework describe largely the same content. These sections should be consolidated or their structural distinction must be explicitly justified.

4. No comparison group: The study includes no comparative element — no control group, no pre/post skill assessment, no benchmarking against comparable studies. It is therefore not possible to attribute observed improvements specifically to the AI-integrated workflow rather than to the intensive format or instructor quality. The contribution should be reframed as descriptive and exploratory, avoiding causal implications.

5. No evaluation of design output quality: The case study examples (Figures 14 and 15) are illustrative but are not evaluated against any architectural or pedagogical criteria. What constitutes a "successful" output in this context? 

6. Informed consent and participant photography: The manuscript states that verbal consent was obtained, yet survey data were collected and participant photographs published. This raises ethical concerns. The authors must provide an ethics committee statement or documentation confirming that verbal consent was institutionally sufficient for this type of data collection and publication.

Comments on the Quality of English Language

The most recurring issue is sentence length. Many sentences across Sections 4 and 7 are excessively long and contain multiple embedded clauses, which reduces clarity and makes the logical flow harder to follow. These should be broken into shorter, more direct statements.

There are also instances of imprecise or redundant phrasing, particularly in the Abstract and Conclusion, where the same ideas are restated in slightly different wording without adding new meaning. Tightening this language would strengthen the manuscript considerably.

Finally, the transition between some sections lacks appropriate connective language, making the text feel fragmented in places — most notably between Sections 5 and 6, where the shift from case study description to results is abrupt.

Author Response

Reply to Reviewer Comments

Manuscript: "AI-Driven Parametric Architecture as an Approach for Computational Design Innovation"

We sincerely thank the Editor and all five reviewers for their careful and constructive reading of our manuscript. The comments have substantially strengthened the paper. Below we respond to each comment individually. Reviewer comments are reproduced in black, and our responses appear in blue. All corresponding changes in the manuscript are marked with yellow highlighting so they can be located easily.

 

Reviewer 2

We thank the reviewer for recognizing the timeliness of the topic and the coherence of the three-dimension workflow, and for the precise methodological guidance below.

Comments and Suggestions for Authors

1. Sample size and statistical representation. Percentages (66.7%, 83.3%, 16.7%) appear throughout, but the number of participants is never stated. State the total N explicitly, reframe the study as qualitative/exploratory and contextualize the percentages, and report every percentage with its absolute number (e.g., "4 out of 6 participants, 66.7%").

Response: We fully agree. The exact number of participants is now stated explicitly in the Abstract (line23), the participants subsection (lines 273–284), and the Results (lines 544–548), and every percentage is now reported with its raw count (e.g., “4 of 6 participants (66.7%)”, “5 of 6 (83.3%)”, “6 of 6 (100%)”, “1 of 6 (16.7%)”). The study is now explicitly reframed as an exploratory, descriptive case study, and all percentage-based statements are presented as preliminary and non-generalizable. Highlighted at line 23 (Abstract), lines 273–284 (participants subsection and table), and lines 544–555 (Results).

2. Research instrument not described. Provide item count, question types, and response scales; include the survey (or a structured description) as a supplementary appendix.

Response: A description of the post-workshop survey (number of items, item types, and response scale) has been added at lines 279–282, the items and their constructs are to be summarized in a new table, and the full instrument is provided as Supplementary Material (Appendix 1).

3. Overlap between Sections 3 and 4. The "Workshop Methodology" (four phases) and "Workshop Methods" (three dimensions) substantially overlap; consolidate them or justify the distinction.

Response: A study-design statement has been added at the start of the methodology (lines231–238) framing Section 3 as the research/pedagogical process (preparation, delivery, experimentation, reflection, participants and data collection and analysis) and Section 4 as the technical workflow; workshop methods (the three design dimensions and the tools used in each). The fuller consolidation of the overlapping descriptive passages between the two sections has not yet been carried out in the manuscript and will be completed in the next revision.

4. No comparison group. There is no control, pre/post skill assessment, or benchmarking, so improvements cannot be attributed to the AI-integrated workflow. Reframe the contribution as descriptive and exploratory, avoiding causal implications.

Response: Agreed. Throughout the Abstract, Results, Discussion and Conclusion we have replaced causal/effectiveness language (e.g., “improves”, “better prepares”) with descriptive, exploratory phrasing (e.g., “participants perceived…”, “appeared feasible and acceptable…”). A statement of the absence of a control/comparison group has been added to the Limitations. Highlighted at lines 26–28 (Abstract), 526 and 544–555 (Results), 614-625 (Limitations), and 637(Conclusion).

5. No evaluation of design output quality. Figures 14 and 15 are illustrative but are not evaluated against architectural/pedagogical criteria. What constitutes a "successful" output here?

Response: We now state explicitly that design outputs were not independently assessed for architectural quality, and we define what “completion” meant operationally in this exploratory setting (a participant producing an end-to-end parametric model and rendered output through the full workflow) while making clear this is a process/feasibility indicator, not a quality judgment. The absence of independent design-quality assessment is now listed as a limitation. Highlighted at line 614 and line 625.

6. Informed consent and participant photography. Verbal consent was stated, yet survey data were collected and participant photographs published. Provide an ethics-committee statement confirming verbal consent was institutionally sufficient.

Response: The ethics and consent statements have been expanded (the Informed Consent Statement, lines 669–674) to describe the institutional ethics review and to confirm the basis on which consent  including consent for the publication of images  was obtained. Also, we make all photos Blur in order to protect the privacy of participants.

Comments on the Quality of English Language

Sentence length is the most recurring issue, especially in Sections 4 and 7; break long sentences into shorter statements. Reduce imprecise/redundant phrasing in the Abstract and Conclusion. Improve transitions, notably the abrupt shift between Sections 5 and 6.

Response: We take this point. A dedicated language pass splitting the long multi-clause sentences in the technical and limitations material, tightening the Abstract and Conclusion, and smoothing the transition from the case-study narrative (line 451-495) to the Results (line 544-555). Additionally, we will made an English review for all the manuscript before publishing.

 

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Major Comments:

(01) The study design is more appropriately described as an exploratory workshop evaluation rather than an effectiveness study. Participants completed a five-day intervention, with data collected through direct observation and a post-workshop survey. The findings are presented descriptively and mainly address usefulness, clarity, and changes in perception. However, the Methods section does not clearly report any baseline measurement, control, or comparison group, or post-workshop follow-up. As a result, the manuscript can reasonably support claims about feasibility and acceptability, but it cannot substantiate stronger claims that AI “improves” students’ design capability or “better prepares” them for professional practice.

(02) The literature review is generally current and relevant, but it requires further refinement. Although the manuscript includes many recent sources and is commendably up to date, it relies too heavily on recent reviews, conference papers, and preprints. It also does not sufficiently engage with broader literature on educational evaluation, creativity assessment, and human–AI co-creation. At minimum, the authors should incorporate and discuss key relevant studies, including Zhuang et al. on generative architectural design, Li et al. on stage-specific AI use, Wang et al. on architecture-studio pedagogy, Huh on student perceptions, and Paananen et al. on text-to-image ideation. These works are directly relevant to the manuscript’s claims regarding creativity, teaching integration, and phase-specific applications of AI in architectural education.

(03) The research question is relevant, but it is too broad for the data collected. The manuscript examines whether AI can support students in 2D/3D modelling for parametric design and whether similar pedagogical approaches can be integrated into real-world projects and professional practice. However, the first issue is only partially addressed through participants’ post-workshop perceptions, while the second is not directly examined, as the study includes no field deployment, longitudinal follow-up, or professional-practice outcome data. In addition, although the paper presents research questions, it does not define clear hypotheses, primary endpoints, or success criteria. This weakens the manuscript’s internal logic and makes the connection between the introduction, methods, findings, and conclusion less convincing.

(04) The main weakness of the manuscript is methodological. In its current form, the paper does not meet the expected reporting standards for an observational educational study. For the observational and survey components, STROBE would be the most relevant formal reporting guideline. TIDieR should also be used to describe the workshop intervention in sufficient detail for reproducibility. If the authors retain observational or reflective material as qualitative evidence, SRQR would also be appropriate. GREET may be useful as a secondary reference because the study concerns an educational intervention, although it was developed specifically for evidence-based practice education. By contrast, CONSORT, PRISMA, and ARRIVE do not apply to this manuscript.

(05) The opening paragraph of the methodology section should clearly explain how the study was conducted. It should describe the research design in sufficient detail and justify why this design is appropriate for examining the effectiveness of AI in supporting architectural modeling and 3D parametric workflows. This is necessary for readers to evaluate the logic of the research process and for other researchers to replicate the study.

(06) The survey and qualitative procedures are underreported. The manuscript mentions direct observation and a post-workshop survey, and later refers to participants’ “perception of AI before and after the workshop.” However, it does not clarify whether the “before” perception was measured prospectively before the intervention, collected retrospectively at the end of the workshop, or inferred from participant discussions. The manuscript also lacks essential methodological details, including the questionnaire items, scale description, number of survey items, missing-data handling procedures, coding framework for observational notes, and information about who analyzed the material. A questionnaire appendix or supplementary file should be provided. Because the instructors appear to have designed and delivered the workshop and then interpreted participants’ responses, observer-expectancy effects and social desirability bias are plausible alternative explanations for the positive findings. The authors should acknowledge these risks and explain how they were managed. Finally, the manuscript should clarify whether VR and digital fabrication were part of the evaluated intervention or whether they were only demonstration add-ons.

(07) The methodology section should provide clear details about the study population of interest (N) and the procedure used to calculate the required sample size (n). The authors should specify the formula used to determine the representative sample size, as this is essential for assessing the statistical validity of the study.

(08) At present, the manuscript does not provide a questionnaire appendix, full item list, coding instructions, access to raw or anonymized data, statistical syntax, or a clear data-availability statement. These omissions limit transparency and make the study difficult to assess or reproduce.

(09) The sample reporting is inadequate. The manuscript states only that participants ranged from early-stage students to recent graduates in practice, but it does not clearly report the total sample size, subgroup counts, age range, gender distribution, recruitment procedure, prior AI exposure, or baseline Rhino/Grasshopper competence. The reported percentages appear to increase in 16.7-point intervals, suggesting that the sample may have included only six participants. If so, 83.3% represents 5/6 and 66.7% represents 4/6, which is far too small to support confident generalization. For example, an 83.3% result with n = 6 produces a very wide Wilson 95% confidence interval of approximately 43.6% to 97.0%, demonstrating the instability of the estimates. Although the authors acknowledge the small sample size as a limitation, its implications are more serious than the Discussion currently suggests. The manuscript should therefore present all results as highly preliminary and avoid broad claims beyond the immediate workshop context.

(10) In addition, the manuscript does not present the demographic profile of the sample with sufficient clarity. The authors should provide a fuller description of respondents’ characteristics and, where possible, compare them with the actual resident composition of the development. Without this comparison, it is difficult to assess whether the sample reasonably represents the population of interest.

(11) The current statistical reporting is insufficient. Given the very small sample size, the paper should report raw numerators and denominators for every percentage, along with exact or Wilson confidence intervals for all proportions. Paired statistical analyses should be applied only where genuine matched pre/post data exist. If the “before” and “after” variables are binary, exact McNemar tests would be appropriate. If the responses are paired ordinal Likert-type data, Wilcoxon signed-rank tests or sign tests would be more suitable. If multiple items are intended to form a broader construct, such as “AI readiness” or “workflow clarity,” the authors should first report internal consistency and explain the scale construction before making composite claims. At present, the manuscript reports only descriptive percentages and then interprets them too strongly. The statistical claims should therefore be moderated, and the results should be presented as preliminary descriptive findings unless supported by appropriate analysis.

(12) Data presentation is uneven. The workflow figures, screenshots, and student examples are a clear strength, as they make the workshop process concrete and accessible. However, the manuscript lacks analytical tables summarising the participants, measures, and outcomes. The use of pie charts is also not ideal, especially with such a small sample. Raw-count tables or stacked Likert bar charts would provide a clearer and more transparent presentation of the findings. In addition, the figures do not replace the need for supplementary materials. The authors should provide the survey form, anonymised response sheet, coding framework, example prompts, Grasshopper files, scripts, software/tool versions, seeds, and reproducibility notes either as supplementary files or through a repository. The current data-availability statement, which claims that “the original contributions … are included in the article,” is not convincing, because the empirical raw data are not actually visible in the manuscript.

(13) The paper’s interpretation requires greater restraint. It is reasonable to conclude that participants perceived the workshop as useful and that AI-assisted scripting and visualization were important learning experiences. However, the evidence does not support stronger claims that the intervention improved architectural design capability, produced durable reductions in technical barriers, or can be directly transferred to professional practice. Several alternative explanations remain possible, including novelty effects, intensive instructor support, short-term enthusiasm, differences in participants’ prior skills, and the absence of any independent assessment of design quality. Therefore, the strongest defensible conclusion is that the workshop appears feasible, acceptable, and promising as a pilot pedagogical approach.

(14) The authors should also add a concise table summarizing the five workshop days, including the learning objectives, tools used, activities, and outputs for each day. At minimum, the manuscript should include a participant characteristics table, a table of survey items and related constructs, and a raw-results table reporting counts and percentages. In addition, the supplementary materials should include the anonymized response data, example prompts, scripts, and representative Grasshopper files or reproducibility materials. The current data-availability statement is not adequate, as the underlying empirical data and intervention materials are not sufficiently available for verification or reuse.

(15) There is a lack of clear separation between the findings derived from observation, questionnaire responses, and interviews. The manuscript should distinguish strictly between direct observational findings, survey results, participant quotations or qualitative themes, and the authors’ own interpretation throughout the Results and Discussion sections. In addition, the Discussion should be strengthened by comparing the study’s findings with those of previous relevant studies. At present, the interpretation appears too internally focused and does not sufficiently explain how the results align with, differ from, or extend existing research.

(16) The writing is generally understandable, but it requires further editorial tightening. Several passages read more like product descriptions than methodological reporting, and there is noticeable repetition across the Introduction, Discussion, and Conclusion. The manuscript also contains typographic inconsistencies in both the main text and the reference list. The prose should be revised to adopt a more neutral scientific tone, reduce vendor-style or promotional language, and maintain a clearer separation between methods, results, and interpretation.

Good luck on getting this far!

Author Response

Reply to Reviewer Comments

Manuscript: "AI-Driven Parametric Architecture as an Approach for Computational Design Innovation"

We sincerely thank the Editor and all five reviewers for their careful and constructive reading of our manuscript. The comments have substantially strengthened the paper. Below we respond to each comment individually. Reviewer comments are reproduced in black, and our responses appear in blue. All corresponding changes in the manuscript are marked with yellow highlighting so they can be located easily.

Reviewer 3

We thank the reviewer for the exceptionally detailed methodological and reporting guidance, which we have used to restructure the empirical reporting of the paper.

Major Comments

(01) The design is better described as an exploratory workshop evaluation than an effectiveness study; the Methods report no baseline, control, comparison, or follow-up. The paper can support feasibility/acceptability claims but not that AI "improves" capability or "better prepares" students.

Response: We agree and have adopted exactly this framing. The study is now described throughout as an exploratory workshop evaluation addressing feasibility and acceptability. Claims that AI “improves” design capability or “better prepares” students have been removed or reworded to perception-based statements. Highlighted at lines 9–11 (Abstract), 231–238 (workshop methodology), 544–555(Results), and 637(Conclusion).

(02) The literature review relies too heavily on recent reviews, conference papers and preprints and does not engage broader literature on educational evaluation, creativity assessment, and human–AI co-creation. Incorporate Zhuang et al. (generative architectural design), Li et al. (stage-specific AI use), Wang et al. (studio pedagogy), Huh (student perceptions), and Paananen et al. (text-to-image ideation).

Response: We have broadened the review to engage educational-evaluation, creativity-assessment and human–AI co-creation literature, and added the suggested works and Margaret Boden’s foundational treatment of machine creativity (The Creative Mind, 1990/2004) to correct the impression that the human-vs-machine agency question is recent. Line 146

(03) The research question is too broad for the data: RQ1 is only partially addressed via post-workshop perceptions, and RQ2 (transfer to professional practice) is not examined (no field deployment or follow-up). No hypotheses, endpoints, or success criteria are defined.

Response: We have narrowed and clarified the questions. RQ1 is now explicitly scoped to participants’ perceived support for 2D/3D parametric modelling; RQ2 is reframed as a forward-looking, exploratory question about perceived transferability rather than a tested outcome, and the Limitations note the absence of field deployment or follow-up. We add explicit, modest success criteria for this exploratory study (workflow completion and acceptability). Highlighted at lines 451 -459.

(04) The main weakness is methodological: the paper does not meet reporting standards for an observational educational study. STROBE (observational/survey), TIDieR (intervention description) apply; SRQR if qualitative material is retained; GREET as secondary. CONSORT, PRISMA, ARRIVE do not apply.

Response: We have adopted these standards. The intervention is now described following TIDieR (what, who, how, where, when, how much), the observational/survey reporting follows STROBE, and the qualitative/reflective material follows SRQR. We state this explicitly in the Methodology and provide the relevant checklist(s) as Supplementary Material. Highlighted at lines 231 -238

(05) The opening of the methodology should clearly explain how the study was conducted, describe the research design in sufficient detail, and justify why it is appropriate, to allow evaluation and replication.

Response: A new opening paragraph (lines 226–230) states the design (single-group, exploratory, observational workshop evaluation), justifies it for the research aim, and outlines the procedure end to end. Highlighted at lines 226–230.

(06) Survey and qualitative procedures are underreported. Clarify whether the "before" perception was prospective, retrospective, or inferred; provide questionnaire items, scale, item count, missing-data handling, coding framework, and who analyzed the data; add a questionnaire appendix. Acknowledge observer-expectancy and social-desirability bias. Clarify whether VR and digital fabrication were evaluated or demonstration add-ons.

Response: Addressed at lines 295–306. We clarify that the “before” perception was collected retrospectively at the end of the workshop (and we note the recall limitation this implies); we report the item count, scale, missing-data handling, and the coding approach for observational notes, and who conducted the analysis; the instrument is provided as Appendix 1. We now explicitly acknowledge observer-expectancy and social-desirability bias as plausible alternative explanations and describe mitigation. We also state that VR and digital fabrication were brief demonstrations, not evaluated components. Highlighted at lines 282–283.

(07) Provide the study population (N) and the procedure/formula used to calculate the required sample size (n) for statistical validity.

Response: Because this is an exploratory case study of a single convenience group, no formal sample-size calculation was performed, and we now say so plainly rather than implying statistical representativeness. We report the actual N and frame all results as descriptive. We note that a power-based sample size would be required for any future confirmatory study. Highlighted at lines 266–269.

(08) The manuscript lacks a questionnaire appendix, full item list, coding instructions, access to anonymized data, statistical syntax, and a clear data-availability statement, limiting transparency and reproducibility.

Response: We have added a questionnaire appendix and item list (Appendix 1), the observational coding framework, and an anonymized response sheet, and we have rewritten the Data Availability Statement to point to these Supplementary Materials/repository. Highlighted at line 674 (Data Availability Statement).

(09) Sample reporting is inadequate: no total N, subgroup counts, age range, gender, recruitment, prior AI exposure, or baseline Rhino/Grasshopper competence. Percentages in 16.7-point steps suggest six participants (83.3% = 5/6; 66.7% = 4/6), too small for generalization (e.g., 83.3% with n = 6 gives a Wilson 95% CI ≈ 43.6%–97.0%). Present all results as highly preliminary.

Response: The reviewer’s inference is correct that the sample is very small. A participant-characteristics table now reports total N, subgroup counts, experience level, and prior AI / Rhino-Grasshopper exposure; recruitment is described at lines284. Wilson 95% confidence intervals are now reported alongside the key proportions to make their instability explicit, and all results are described as highly preliminary. Highlighted at line 284 (participant table) and lines 272-284 (Results, with Wilson CIs).

(10) The demographic profile of the sample is not presented with sufficient clarity; provide a fuller description and, where possible, compare with the relevant population.

Response: The new participant-characteristics table provides the demographic profile. As the study concerns an educational workshop rather than a resident population, we describe the sampling frame (participants recruited from the host programme) and note explicitly that the convenience sample is not intended to represent a defined population. Highlighted at line 275.

(11) Statistical reporting is insufficient: report raw numerators/denominators and exact or Wilson CIs for all proportions; apply paired tests only where matched pre/post data exist (McNemar for binary, Wilcoxon/sign for ordinal); report internal consistency for any composite; otherwise present descriptive findings and moderate the interpretation.

Response: We have moderated all statistical claims. Every proportion is now reported as a raw count with its denominator and a Wilson 95% CI. Because the “before/after” perception data are retrospective and not genuinely matched pre/post measurements, we do not apply paired inferential tests, and we say so; the findings are presented descriptively. No composite scales are claimed. Highlighted at lines 531–536.

(12) Data presentation is uneven: add analytical tables (participants, measures, outcomes); pie charts are not ideal for a small sample — use raw-count tables or stacked Likert bars; provide survey form, response sheet, coding framework, prompts, Grasshopper files, scripts, tool versions, seeds, and reproducibility notes. The current data-availability statement is unconvincing.

Response: We have added analytical tables (participant characteristics; five day workshop schedule). Two tables are highlighted (participant characteristics, line 285; five-day schedule, line307) and the survey is now available at appendix 1, prompts, scripts, representative Grasshopper files, and tool/version information as Supplementary Material, with a rewritten Data Availability Statement.

(13) Interpretation needs greater restraint: the evidence supports that participants found the workshop useful, not that it improved design capability, reduced technical barriers durably, or transfers to practice. Alternative explanations (novelty, instructor support, enthusiasm, prior-skill differences, no design-quality assessment) remain. The strongest defensible conclusion is feasible/acceptable/promising pilot.

Response: The Discussion and Conclusion have been rewritten to this standard. We now conclude that the workshop appeared feasible, acceptable and promising as a pilot, list the alternative explanations named by the reviewer, and remove stronger claims. Highlighted at line 613-624 (Limitations) and line 636 (Conclusion).

(14) Add a concise five-day table (objectives, tools, activities, outputs); a participant-characteristics table; a survey-items/constructs table; and a raw-results (counts and percentages) table. Supplementary materials should include anonymized data, prompts, scripts, and Grasshopper files. The data-availability statement is inadequate.

Response: Two tables have been added (five-day schedule; participant characteristics), and the supplementary materials and Data Availability Statement have been expanded accordingly, as noted above. Two of the four tables are in place (five-day schedule, line 307; participant characteristics, line 284); Additionally, the survey is now available at appendix 1, prompts, scripts, representative Grasshopper files, and tool/version information as Supplementary Material, with a rewritten Data Availability Statement.

 

(15) There is no clear separation between observational, questionnaire, and interview findings; distinguish observation, survey results, participant quotations/themes, and authors’ interpretation throughout Results and Discussion. Strengthen the Discussion by comparing findings with prior studies.

Response: Results are now organized under clearly labelled sub-headings that separate observational findings from survey findings and from authors’ interpretation. (No formal interviews were conducted; we state this to avoid implying a data source we did not use.) The Discussion now compares our observations with prior studies ([2], [7], [23]). The narrative is labelled as observational at line 450 and the Results proportions appear at lines 548–554.

(16) The writing needs tightening: some passages read like product descriptions; there is repetition across Introduction, Discussion and Conclusion; and there are typographic inconsistencies in text and references. Adopt a more neutral scientific tone and reduce vendor-style language.

Response: We have revised the prose to a neutral scientific register, reduced tool-promotional phrasing, removed repetition across the Introduction/Discussion/Conclusion, and corrected typographic and reference-list inconsistencies. These language and typographic corrections are not yet reflected in the manuscript and will be completed in the next revision.

 

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The paper is well structured and demonstrates practical engagement with contemporary AI tools; however, several aspects require improvement to strengthen its scholarly contribution.

The methodology should be clarified by providing more detail regarding the participant sample, survey design, and evaluation process, as the findings rely mainly on descriptive and self-reported feedback.

The manuscript would also benefit from a more critical discussion of limitations, including the short workshop duration, limited generalizability, and reliance on specific proprietary tools. Additionally, the novelty and research contribution should be articulated more clearly to distinguish the work from a workshop report and better position it within existing literature on AI in architectural education.

 

Comments on the Quality of English Language

Finally, minor language editing and reduction of repetitive tool descriptions would improve readability and overall quality.

Author Response

Reply to Reviewer Comments

Manuscript: "AI-Driven Parametric Architecture as an Approach for Computational Design Innovation"

We sincerely thank the editor and all five reviewers for their careful and constructive reading of our manuscript. The comments have substantially strengthened the paper. Below we respond to each comment individually. Reviewer comments are reproduced in black, and our responses appear in blue. All corresponding changes in the manuscript are marked with yellow highlighting so they can be located easily.

Reviewer 4

We thank the reviewer for the encouraging overall assessment and the focused suggestions.

Comments and Suggestions for Authors

The methodology should be clarified with more detail on the participant sample, survey design, and evaluation process, as findings rely mainly on descriptive, self-reported feedback.

Response: Addressed through the new Participants and Setting (3.1) and Data Collection and Analysis (3.2) subsections, the participant-characteristics and survey-items tables, and the supplementary survey instrument. We also state plainly that findings are descriptive and self-reported. Highlighted at lines 272–308.

The manuscript would benefit from a more critical discussion of limitations—short workshop duration, limited generalizability, and reliance on specific proprietary tools. The novelty and research contribution should be articulated more clearly to distinguish the work from a workshop report.

Response: The Limitations (7.1) have been expanded to address the short duration, small non-representative sample, absence of a control group and independent quality assessment, and dependence on specific proprietary tools (with a note on reproducibility and tool churn). A clearer contribution statement in the introduction and conclusion distinguishes the paper from a descriptive workshop report by foregrounding the structured, replicable AI-integrated workflow and the reporting-standard-based evaluation design. Highlighted at lines 70–77 (contribution) and lines 613–624 (limitations).

Comments on the Quality of English Language

Minor language editing and reduction of repetitive tool descriptions would improve readability.

Response: The manuscript has been language-edited, and repetitive tool descriptions have been condensed, consistent with the edits noted for Reviewers 2 and 3. Additionally, a whole English review was made of the manuscript before the final version.

 

Author Response File: Author Response.pdf

Reviewer 5 Report

Comments and Suggestions for Authors

This study reports on a workshop that introduced its participants to an AI-driven parametric workflow. The workflow was built out of existing, proprietary components. The study falls under two broad topics, human-machine collaboration in (architectural) design and the integration of AI-tools in (architectural) education. Both topics are highly interesting, and the paper presents some questions worthy of investigation. However, at its present state the paper seems underdeveloped for a journal paper. It has both conceptual and methodological issues that must be addressed before resubmission.

I am listing my main concerns and some additional remarks which I hope might help in the further developments.

Conceptual:

• The study fails to clearly identify its goal. It remains unclear whether this is an educational research paper (and if so, is it about the learning experience, the learning process or the learning outcome), and the workshop methodology is the novelty, OR a computational design paper, where the AI-assisted design workflow would be the novelty. While there are some arguments for both in the paper, neither one is supported strongly enough in its present form.
• The study positions itself as architecture specific, however, the workshop exercise, while this is not clearly stated in the paper, seems to be designing a vase. I would argue that in its present state it is more related to product design.
• If the aim is to contribute to the educational research literature, a much more rigorous experimental methodology must be presented, the scope much more clearly defined (see also below).
• If the aim is to position the proposed integrated workflow as a novelty, then its description should be much more specific on how it is going beyond the current state of the art, and address critical usability issues, such as the amount of different platforms used throughout the process.

It also must address the fact that “design” is reduced to geometric manipulations (usability, durability etc are not addressed, or not reported on)

Methodological:

• The title does not reflect the content (where is the design innovation?); The research questions (to what extent can AI support students?, comparative learning experience assessment /there is no comparison/), and the declared main contribution of the paper (see Conclusion, “The key contribution…proposing a structured AI-integrated workflow”) are only vaguely related.

 

• The report on the workshop is confusing and vague: The number of participants is not stated explicitly. Their year-level, or experience level (in any of the tools) are not described sufficiently. The tools used by the users, vs all the possible options (Figure 1) are unclear. Some interesting problems are mentioned only by the way: studying the impact of AI-assisted visual coding vs AI-assisted scripting on the learning or the design quality would be worthy on its own right of a much more in-depth study. Some phrases, like hands on experimenting are very confusing – was it not an all-digital workshop? How come Fig 14 has a hand drawn sketch? The reporting on the questionnaire outcome seems selective (In figure 18 the diagrams show responses to different questions next to each other). The evaluation report mixes participants self-reflection with their evaluation of the instructors, and it is not clear, why.
• The literature review/background is both too broad and not extensive enough. The covered topics range from AI ethics (barely mentioned later in the paper) to AI-integrated design tools and AI in education. This should be reconsidered once the focus of the paper is clarified, and the background should be more tailored to the problems addressed by the paper. A more representative selection of references would be desirable. It is respectable that authors collected some of the recent papers on these topics as the state of the art, but such a description leads to the false perception as if, for example, the philosophical problem of human vs machine agency in creative processes was so recent. Margaret Boden’s seminal book, The creative mind is 36 years old, and discusses this problem extensively. She was not the first, either.

The background section identifies important research gaps, but those are not addressed by the paper.

• Many of the figures must be revised: e.g., Fig 1 is not corresponding to the process described in the text, Figure 2-3 do not help, but add to the confusion on what happens in the conceptual design phase,  Images of the workshop do not seem to add much to the understanding of the paper, especially Fig 11.
• Claims appear in the conclusion section that are not supported by the results of the study: e.g.. "the study confirms that AI does not replace the human". No such confirmation is provided, nor was this statement tested. Or the “importance of critical approach” how was that implemented in the workshop methodology? How is that supported by the results? What outcomes were revised based on a critical approach?

Once again, I am hopeful, given the timeliness of the topic, that some of these comments prove useful for the future development of this interesting project.

 

Author Response

Reply to Reviewer Comments

Manuscript: "AI-Driven Parametric Architecture as an Approach for Computational Design Innovation"

We sincerely thank the editor and all five reviewers for their careful and constructive reading of our manuscript. The comments have substantially strengthened the paper. Below we respond to each comment individually. Reviewer comments are reproduced in black, and our responses appear in blue. All corresponding changes in the manuscript are marked with yellow highlighting so they can be located easily.

Reviewer 5

We thank the reviewer for the thoughtful, developmental critique. We have used it to sharpen the paper’s focus and to be more precise about scope and claims.

Conceptual

• The study fails to clearly identify its goal — whether it is educational research (and if so, about learning experience, process, or outcome) with the workshop methodology as the novelty, OR a computational-design paper with the workflow as the novelty. Neither is supported strongly enough.

Response: We have made the goal explicit. The paper is positioned as educational/pedagogical research, focused on the learning experience and process (not measured learning outcomes), with the structured AI-integrated workflow serving as the pedagogical instrument under study rather than as a claimed computational-design innovation. The introduction, aims, and conclusion now state this consistently. Highlighted at lines 231–283 (Methods) and line 636 (Conclusion).

• The study positions itself as architecture-specific, but the exercise (designing a vase) seems closer to product design; this is not clearly stated.

Response: We now state the exercise scope openly: within the five-day format, participants worked at small / interior- and product-scale artefacts (including a vase and a pavilion element), chosen so a complete ideation-to-visualization workflow could be executed in the time available. We acknowledge in the Limitations that this narrows architectural generalizability and note it as a boundary of the study. Highlighted at lines 450–458.

• If the aim is educational-research contribution, a more rigorous experimental methodology and clearer scope are needed.

Response: We have narrowed the scope and adopted recognized reporting standards (STROBE/TIDieR/SRQR; see Reviewer 3) while being explicit that this is an exploratory, single-group study without a control condition. Rather than overstate rigor, we now frame the design honestly as a pilot and identify what a subsequent controlled study would require. Highlighted at lines 226–238.

• If the aim is to position the integrated workflow as a novelty, its description must be more specific about how it goes beyond the state of the art and must address usability issues such as the number of platforms used. Also, "design" is reduced to geometric manipulations (usability, durability, etc. are not addressed or reported).

Response: We have chosen the pedagogical framing (above) rather than a state-of-the-art computational-design claim, and we now discuss the multi-platform usability burden explicitly as both an observed difficulty and a limitation. We also acknowledge candidly that, in this exploratory setting, “design” was largely operationalized as geometric/formal manipulation and that utility, durability, and feasibility were discussed conceptually in the evaluation phase but not independently assessed. Highlighted at lines 604-624 (Limitations).

Methodological

• The title does not reflect the content (where is the design innovation?); the research questions and the declared main contribution ("proposing a structured AI-integrated workflow") are only vaguely related.

Response: We have aligned the title, questions, and contribution. The revised working title removes the “design innovation” claim and foregrounds the pedagogical workflow, “AI-Assisted Parametric Design in Architectural Education: An Exploratory Workshop Evaluation of a Structured Human-in-the-Loop Workflow.” The research questions and the stated contribution have been rewritten to match this framing.

• The report on the workshop is confusing and vague: participant number, year-level, and tool experience are not stated; which tools participants actually used (vs. all options in Fig. 1) is unclear; "hands-on experimenting" is confusing for an all-digital workshop; Fig. 14 shows a hand-drawn sketch; the questionnaire reporting seems selective (Fig. 18 shows different questions side by side); and the evaluation mixes participants’ self-reflection with their evaluation of instructors.

Response: These are now resolved: participant number and experience levels are reported (line 284, participant table); Figure 1 has been revised to distinguish the tools actually used from the broader option space; the ambiguous phrase “hands-on experimenting” has been reworded to “hands-on, screen-based experimentation,” and we clarify the status of any hand-drawn input in Figure 14; the survey reporting now presents all items with raw counts rather than selected results; and we have separated participants’ self-reflection from their evaluation of instruction into distinct, clearly labelled parts of the Results, explaining why instructor feedback is reported (as part of the acceptability assessment). Participant number and experience levels are highlighted at line 284.

• The literature review is both too broad and not extensive enough (ranging from AI ethics, barely used later, to AI tools and AI in education); it should be tailored to the paper’s focus. The framing wrongly implies that human-vs-machine agency in creativity is recent—Margaret Boden’s The Creative Mind (36 years old) discusses this extensively, and she was not the first.

Response: The review has been refocused on the paper’s core (AI in parametric/architectural design education and human–AI co-creation), and material only tangentially used (e.g., extended AI-ethics discussion) has been trimmed to match what the study actually addresses. We now cite Margaret Boden, The Creative Mind: Myths and Mechanisms (Weidenfeld & Nicolson, 1990; 2nd ed. Routledge, 2004), and acknowledge the longer intellectual history of machine-creativity and agency debates so they are not presented as recent. Highlighted at lines 134–147.

• The background identifies important research gaps that the paper does not then address.

Response: We have reconciled the gaps with the study’s actual scope: gaps The paper does not address (e.g., longitudinal transfer to practice, design-quality assessment), which are no longer presented as ones we resolve but are moved into Future Work, while the gap we do address (structured, replicable integration of AI across the parametric workflow in a teaching setting) is stated clearly. Highlighted at line 648 (future-study requirements); the reorganization of unaddressed gaps into Future Work is still being completed.

• Many figures need revision: Fig. 1 does not correspond to the process in the text; Figs. 2–3 add confusion about the conceptual-design phase; workshop images (especially Fig. 11) add little.

Response: We have revised Figure 1 to correspond to the process described in the text, clarified/relabelled Figures 2–3 so the conceptual-design phase reads unambiguously, and we added tile to figure 11 as well.

• Claims in the Conclusion are unsupported by the results: e.g., "the study confirms that AI does not replace the human" (never tested), and the "importance of a critical approach" — how was it implemented in the methodology, and what outcomes were revised based on it?

Response: We have removed the unsupported “confirms that AI does not replace the human” claim and replaced it with a statement grounded in what we observed (participants used AI outputs as material to be curated and revised, consistent with a human-in-the-loop process). Where we discuss a “critical approach,” we now tie it to concrete workshop practice—the evaluation phase in which generated images/geometry were accepted, rejected, or iterated—rather than asserting it as a proven result. Highlighted at line 636.

 

We thank all reviewers again for their time and for comments that have materially improved the manuscript. We believe the revised paper is substantially clearer, more transparent about its exploratory scope, and better supported by its evidence.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have addressed all the reviewers’ comments in the revised manuscript.

Author Response

We sincerely thank the reviewer for the positive assessment and for the time and effort devoted to evaluating our manuscript. We are also grateful for the constructive comments provided in the first round, which substantially improved the quality and clarity of the paper.

Reviewer 2 Report

Comments and Suggestions for Authors

The revised manuscript shows genuine and substantial improvement on four of the six original concerns: the sample size is now stated explicitly and every percentage is reported with its raw count and confidence interval, the survey instrument is fully documented with the complete questionnaire in an appendix, causal language has been replaced with appropriately hedged, exploratory phrasing throughout most of the paper, and the definition of "successful" design output has been clarified and added to the limitations. Two issues, however, remain unresolved and should be addressed before acceptance: the overlap between Sections 3 and 4 was explicitly deferred by the authors to "the next revision" rather than consolidated now, and the response letter's claim that participant photographs were blurred is not reflected in the actual figures (11, 12, 13), which still show unobscured faces despite the newly added written consent and IRB approval — this discrepancy needs clarification or correction. Additionally, one sentence in the Conclusion still uses causal language ("AI can significantly lower technical barriers") that is inconsistent with the more carefully hedged phrasing used in the Abstract and elsewhere, and the promised English-language editing pass does not yet appear to have reached the technical sections (e.g., 4.2.1), which retain the same overlong, multi-clause sentences flagged in the first round.

Author Response

We sincerely thank the reviewer for the positive assessment and for the time and effort devoted to evaluating our manuscript. We are also grateful for the constructive comments provided in the first round, which substantially improved the quality and clarity of the paper.

Comment 1. The overlap between Sections 3 and 4 was explicitly deferred by the authors to “the next revision” rather than consolidated now; this should be addressed before acceptance.

Response: We have now consolidated the two sections and made their distinct roles explicit rather than deferring the change. The opening of Section 4 moves directly from how the workshop was run—described in Section 3—to the AI-integrated workflow itself, that is, the specific tools and their role across the design process (Lines 325-330). The redundant re-introduction of the workshop setting has been removed from the start of Section 4, and the hosting venue is now named once, at the start of Section 3, so the methodological description in Section 3 and the tool-and-workflow description in Section 4 no longer overlap.

 

Comment 2. The response letter’s claim that participant photographs were blurred is not reflected in the actual figures (Figures 11, 12, and 13), which still show unobscured faces despite the newly added written consent and IRB approval. This discrepancy needs clarification or correction.

Response: We thank the reviewer for identifying this discrepancy. To resolve it, we have obscured (blurred) the faces of all identifiable participants in Figures 11, 12, and 13, so that the figures are consistent with the anonymisation described in the consent statement. Written consent and IRB approval (no. IRB-ASU-26-030) remain on file, but no identifiable faces are now shown.

 

Comment 3. One sentence in the Conclusion still uses causal language (“AI can significantly lower technical barriers”) that is inconsistent with the more carefully hedged phrasing used in the Abstract and elsewhere.

Response: This has been corrected. The sentence in the Conclusion now reads “the findings suggest that AI may lower technical barriers,” replacing the earlier “AI can significantly lower technical barriers,” so the Conclusion is now consistent with the hedged, exploratory phrasing used in the Abstract and throughout the paper.

Comment 4. The promised English-language editing pass does not yet appear to have reached the technical sections (e.g., Section 4.2.1), which retain the same overlong, multi-clause sentences flagged in the first round.

Response: We have carried out a further language-editing pass focused on the technical sections. In Section 4.2.1 in particular, the overlong multi-clause sentences have been broken into shorter ones, redundant phrasing removed, and punctuation errors corrected (Lines 383-407).

Reviewer 3 Report

Comments and Suggestions for Authors

Well done on addressing and revising most of the comments.

Author Response

We thank the reviewer for the kind and encouraging feedback and for the constructive comments in the previous round that helped us strengthen the manuscript.

Reviewer 4 Report

Comments and Suggestions for Authors

The authors have address the required changes. 

Comments on the Quality of English Language

Finally, minor language editing and reduction of repetitive tool descriptions would improve readability and overall quality.

Author Response

Comment 1. The authors have addressed the required changes.

Response: We thank the reviewer for the positive evaluation and for confirming that the required changes have been addressed.

Comment 2. (Quality of English Language) Minor language editing and reduction of repetitive tool descriptions would improve readability and overall quality.

Response: We thank the reviewer for this helpful suggestion. In the present revision we carried out a further language-editing pass across the manuscript to improve readability and clarity, and we condensed the repetitive descriptions of the AI tools so that each tool and its role are introduced once and thereafter referred to concisely, removing duplicated explanations in the workflow and case-study sections.

Reviewer 5 Report

Comments and Suggestions for Authors

This is the second revision of the study, which reports on a workshop introducing participants to an AI-driven parametric workflow. The workflow was built from existing proprietary components. The study falls under two broad topics: human-machine collaboration in (architectural) design and the integration of AI tools in (architectural) education. The paper, in its revised state, commits to belonging more to the latter, which I believe is adequate. I acknowledge and welcome the revision, as it addressed some of the issues raised in my initial review, but I still do not think the paper is ready for publication. The paper attempts to address comments largely by adding new elements, which are useful, but falls short in the necessary deep revision of the existing text.

There are still both conceptual and methodological issues, but it mainly comes down to the paper not quite clearly defining its aims, how to achieve them, and why that would matter.

 (example: in the Introduction, authors state, as their RQ2:

• “How can comparable AI-assisted pedagogical approaches be effectively integrated into real-world architectural projects and professional design practice” , which is absolutely not tested. They use a single question in the survey to poll participants on the PERCEIVED likelihood of them using AI later)

 

Assuming the paper really is about how to integrate AI tools in architectural education (RQ1), these are the key issues the authors must address:

• Authors believe AI tools should be integrated in every step of the architectural design process in the design curriculum, it seems. So, they develop a workflow (this they sort-of-did), as a proof of concept. The challenge is: I do not think anyone thinks otherwise. So, it seems essential that they very explicitly demonstrate that while everyone wants this, very few or no one has done it. The goal no. 1 is the integrated workflow. I will come back to this.
• Consequently, there must be a proof of feasibility study. Hence the workshop. But then again, what is the question they are trying to answer (goal no. 2)? What is this integrated workflow good for? The text repeats the claim that the workshop shows that this integration makes people see AI as a partner and not a replacement. I think this should not be a central argument. The experimental setting is not set up to prove this; this remains a claim: There is a single, quite pointed question in the survey if participants feel this way (the other option being, a bit simplified, that AI is useless). That is weak evidence.

Because the sample is so small and diverse, the most evident hypothesis they might aim to TEST is that anyone can learn/use this workflow. This is mentioned, though not emphasized. However, this hypothesis, in the paper’s current state, is neither supported nor falsified by the results, because the workflow looked different for different participants (they worked on different tasks and used, or were able to use, different tools, such as Claude + Rhino or ChatGPT/Raven and GH.) The experience, the “this” that they can learn, is different for each.

 

Some further remarks:

• The keywords do not reflect that the paper deals with AI integration in design curriculum
• The task that the participants had to do, over the workshop, is never defined in the paper, only alluded to through images.
• The paper still consistently refers to the design exercise as architectural. It is not; it is still a vase OR (!) a nondescript façade element. That is product design. I understand the small scale, but the complexity is markedly different. That is OK, but it should be clearly labeled as product design, and not architectural
• There is this recurring claim, in different formats that the participants were able to use AI to iterate their design, develop variations as opposed to the linear process (?). Of what linear process? The design process is not linear by default, whether it is analog, digital, or even generative. There is also nothing to back this claim – there is no evidence that this was the case or not, and it remains a mystery what the authors compared this process to.
• The results are still selectively reported, despite their written response. We do not have the results of all questions. There are also questions answered that do not appear in the survey, I think (e.g., self-assessment of AI understanding).
• The workflow, with its branching, multiple-options and multiple-platform methodology is very complicated. A short discussion on possible ways to simplify it, what might be a future research endeavor, could be useful.
• It is not acceptable that the Limitations section overwrites the claims of the Discussion section. The discussion section (and the conclusion) still presents untested or unproven claims (like: “The findings of this study highlight that the effectiveness of integrating artificial in- telligence into parametric design is not determined solely by the tools themselves but by how these tools are structured within a coherent and guided workflow. “ )Absolutely not tested, nor demonstrated.) some of them are self-critiqued in the limitations, but they should not appear in the paper in the first place.
• The conclusion must reassert the aims of the paper and how they are achieved. Most of it is still a mix of novel workflow and curricular integration.
• Figures are not revised substantially, most notably Fig 1 still presents all possible tools, not the ones they were instructed to use. Fig 2, 3 unchanged, still unclear.
• Why is BIM, VR mentioned? Seems quite irrelevant to the topic at hand.
• Guidelines (SRQR, etc) on reporting results and statistics: why mention three, in their aim equivalent reporting guideline, and why restrict their use to one portion of the paper? They are meant for the whole paper, and following any one of those, would be more than enough. I cannot fathom why they mention, once, the Wilson confidence interval – which is, I believe, meant for head or tail-type of problems. But more importantly, there is not much statistics can do for 9 (in reality, 6)-participant studies. Small participant number is no problem, it is a pilot, but attempting to apply statistical methods is out of place.

Author Response

We thank the reviewer for the comments and the acknowledgment of welcoming the past revision.

Comment 1. There are still both conceptual and methodological issues, but it mainly comes down to the paper not clearly defining its aims, how to achieve them, and why that would matter.

Response: The aims are now explicit. The study is framed as an exploratory feasibility-and-acceptability evaluation of a structured, human-in-the-loop AI-parametric workflow in an architectural educational setting (Abstract, lines 9–11and lines 29-30); study framing, lines 243–250). The rewritten gap paragraph states why this matters and closes with the study’s first aim (lines 45–60), and the two research questions (lines 62–67) are tied to feasibility (RQ1) and to perceived usefulness and attitudinal change treated as exploratory (RQ2).

Comment 2. In the Introduction the authors state as RQ2: “How can comparable AI-assisted pedagogical approaches be effectively integrated into real-world architectural projects and professional design practice,” which is not tested. Only a single survey question is used, polling participants on the perceived likelihood of them using AI later.

Response: We accept this and rewrote RQ2 so it no longer promises an untested claim. It now reads: “How do participants perceive the usefulness of this AI-assisted parametric workflow, and how do their perceptions of AI’s role in design change over the course of the workshop?” (lines 65–67). This is answerable from the data collected (survey Section B and the usefulness ratings) and is explicitly designated exploratory (lines 468–469). The earlier professional-practice formulation has been removed.

Comment 3. Assuming the paper is about how to integrate AI tools in architectural education (RQ1): the authors appear to believe AI tools should be integrated in every step of the design process in the curriculum, and develop a workflow as a proof of concept. However, few would think otherwise, so the authors must very explicitly demonstrate that, while everyone wants this, very few or no one has actually done it. The integrated workflow is goal no. 1.

Response: The gap statement was rewritten to make this contrast explicit (lines 45–60). Drawing on the literature already cited, it now states that although integrating AI is widely endorsed, in practice AI stays concentrated in early, schematic phases while integrated, end-to-end pipelines remain rare [16][19]; it then frames delivering such a workflow as the study’s first aim.                        

Comment 4. Consequently a proof-of-feasibility study is needed (hence the workshop), but what question is being answered (goal no. 2)—what is the integrated workflow good for? The repeated claim that the workshop shows integration makes people see AI as a partner and not a replacement should not be a central argument: the experimental setting is not designed to prove it, and it rests on a single, pointed survey question, which is weak evidence.

Response: We agree, and “partner, not replacement” is no longer presented as a finding. The workflow’s value (goal 2) is now stated as a concrete outcome: participants completed a full concept-to-visualization AI-parametric process within a short taught workshop (Conclusion, lines 654–666). The human-in-the-loop character is presented as the workflow’s design rationale, not as a tested result.

 

Comment 5. Because the sample is small and diverse, the most evident testable hypothesis is that anyone can learn/use this workflow. This is mentioned but not emphasized, and in the current state it is neither supported nor falsified, because the workflow differed across participants (different tasks and tools, e.g., Claude + Rhino or ChatGPT/Raven and Grasshopper). The “this” that they can learn is different for each participant.

Response: We address this directly in the Limitations (lines 644–648). Because the tasks and tool combinations differed across participants (for example, ChatGPT/GhPython with Grasshopper versus Raven with grasshoper), the workflow each participant learned was not identical; the hypothesis that any participant can learn a single, common workflow therefore cannot be uniformly supported or falsified in this design, and is proposed for future controlled study using a shared task. We do not claim to have tested learnability.

 

Comment 6. The keywords do not reflect that the paper deals with AI integration in the design curriculum.

Response: Revised (lines 31–32). The keywords now read: AI in architectural education; generative AI; parametric design; human-in-the-loop; computational design; Rhino 3D.

 

Comment 7. The task the participants had to do over the workshop is never defined in the paper, only alluded to through images.

Response: The task is now defined explicitly (case study, lines 473–478). Each participant developed a design concept typically a pattern or motif and applied it to an object of their own choosing, carrying it from an initial concept image, through a parametric model, to a rendered output; the objects differed, but every participant produced the same three deliverables (concept, parametric definition, and final visualization), so the shared workflow, not any particular object, was the focus. The two worked examples (Figures 14–15) document this end to end.

 

Comment 8. The paper still consistently refers to the design exercise as architectural. It is not—it is a vase or a nondescript façade element, which is product design. The small scale is understandable, but the complexity is markedly different; it should be clearly labeled as product design, not architectural.

Response: Thank you for the comment we updated this part in lines 524, 630,632, 643.

Comment 9. There is a recurring claim, in various forms, that participants used AI to iterate their design and develop variations as opposed to a linear process. What linear process? The design process is not linear by default (analog, digital, or generative). Nothing backs this claim—there is no evidence either way—and it is unclear what the process was compared to.

Response: Accepted. Because no linear baseline was observed or measured, the phrase implying one (“moving beyond linear workflows”) has been deleted; the sentence now reads simply “an exploratory design approach that allowed for continuous refinement and development” (Discussion, lines 613–614), with no implied comparison.

 

Comment 10. The results are still selectively reported, despite the written response. The results of all questions are not provided, and some answered questions do not appear in the survey (e.g., self-assessment of AI understanding).

Response: Two points. First, the omitted items are restored: the full 15-item instrument including the self-assessment of AI understanding (item B3) that was previously absent is provided in Appendix 1 (lines 723–781; B3 at lines 765–767). Second, closed-item responses are reported as raw counts with percentages (lines 568–576) and, for completeness, item-level frequencies for every closed question are provided as Supplementary Material (Table S1) so that no answered item is omitted; given the small convenience sample these are presented as preliminary descriptive findings.

 

Comment 11. The workflow, with its branching, multiple options, and multiple-platform methodology, is very complicated. A short discussion of possible ways to simplify it—as a future research direction—could be useful.

Response: Added and integrated into the future-directions discussion in the Conclusion (lines 686–688). Simplification is listed among the practical considerations for future work: consolidating the current branching, multi-platform structure into fewer tools and a single guided track, with optional advanced branches introduced only once the core process is mastered.

 

Comment 12. It is not acceptable that the Limitations section overwrites the claims of the Discussion. The Discussion (and Conclusion) still present untested or unproven claims—e.g., “the effectiveness of integrating artificial intelligence into parametric design is not determined solely by the tools themselves but by how these tools are structured within a coherent and guided workflow”—which are not tested or demonstrated. Being self-critiqued in the Limitations does not justify their presence in the paper.

Response: We revised the claims at source. In the Discussion the statement now reads that effectiveness “appeared to depend not only on the tools themselves but also on how they were structured within a coherent, guided workflow, an exploratory observation that would require controlled testing to confirm” (lines 584–587). The corresponding Conclusion sentence is likewise phrased as an explicitly preliminary observation (lines 663–666).

Comment 13. The Conclusion must reassert the aims of the paper and how they are achieved. Most of it is still a mix of novel workflow and curricular integration.

Response: The opening paragraph of the Conclusion was rewritten to restate the aims and how each was addressed (lines 654–666): first, proposing the integrated, stage-by-stage human-in-the-loop workflow (RQ1); and second, examining, in an exploratory manner, participants’ perceived usefulness and attitudinal change (RQ2). It then reports, for each aim in turn, what the workshop showed, keeping the novel-workflow contribution distinct from the curricular-integration framing.

Comment 14. Figures are not revised substantially—most notably, Figure 1 still presents all possible tools rather than the ones participants were instructed to use, and Figures 2 and 3 are unchanged and still unclear.

Response: Figure 1 (line 333) has been updated to distinguish the tools actually used in the workshop  from the broader set of available options shown for context. Figure 2(line 340) has been redrawn so that the three steps are numbered and connected in sequence, each sub-label is aligned with its step, the AI image tools are linked to the phase, and a short explanatory sentence has been added beneath each step. Figure 3 (line 344) has been redrawn with the three phases shown as clearly separated, labelled bands, a one-line explanation for each phase, and the feedback loops rendered as clearly labelled arrows; it is now marked "Adapted from [30]" to reflect that it has been adjusted to the workshop's process rather than reproduced.

Comment 15. Why are BIM and VR mentioned? They seem quite irrelevant to the topic at hand.

Response: Their limited role is now clarified. BIM is retained only as brief context and is described as “not an AI technology in itself” (lines 99–101). VR is mentioned only as an optional, short demonstration during the workshop; it was not one of the tools participants used in the assessed workflow (lines 295–296; Table 2, line 320). Neither is presented as central to the study.

Comment 16. On reporting guidelines (SRQR, etc.): why mention three broadly equivalent guidelines and restrict their use to one portion of the paper? They are meant for the whole paper, and following any one would be sufficient. Also, why mention the Wilson confidence interval once (meant for head-or-tail problems)? More importantly, statistics can do little for a 9- (in reality, 6-) participant study; a small sample is fine for a pilot, but applying statistical methods is out of place.

Response: Two points. First, the three guidelines are complementary rather than equivalent, and each governs a distinct component: TIDieR for the intervention, STROBE for the observational/survey component, and SRQR for the reflective/qualitative material (lines 246–248); the text now makes this division explicit. Second, we agree that interval or inferential statistics are inappropriate for a sample of six: the Wilson confidence intervals have been removed from the Methods (lines 310–311) and the Results (line 569), and the study now reports only raw counts and descriptive percentages, with no inferential tests, consistent with a pilot of this size.

 

Author Response File: Author Response.pdf