A study of GenAI usage by Design Students: Analysis of Survey Results and Journals of AI practices at the Politecnico di Milano in 2025/2026

Synthesis: Broadbent and colleagues survey design students at the Politecnico di Milano's School of Design and pair the results with AI-use journals kept during a Masters research methods course, finding that GenAI use is very frequent but concentrated in the early research and writing phases of the design process, and that students' perceived project ownership and Creativity are largely unaffected. The study's central tension is the research phase itself: students offload literature search, summarization and editing to AI while remaining skeptical of its outputs (65% distrust them; 85% systematically modify them), suggesting both a pragmatic, verification-driven literacy and a possible devaluing of research as a creative act.

Overview

The introduction situates the study against a large existing literature — a bibliometric review of more than 2,700 higher-education GenAI studies, Anthropic's analysis of over one million chatbot conversations (7% academic, coursework above 12%), the UCLA Class of 2025 survey (73% using GenAI for coursework), a four-university Australian survey of over 8,000 respondents (over 80% used GenAI for study tasks), and the UK HEPI survey (95% using GenAI, 94% for assessed work). What those general-purpose studies cannot answer, the authors argue, is what happens in a discipline with a structured multi-phase process. Design runs from secondary research, through primary research (interviews, field observation, co-design workshops), to concepting, prototyping, testing and development. The paper's guiding questions are therefore design-specific: does GenAI support the whole process or only particular phases, and how does its use affect students' sense of authorship and ownership of their projects?

Study Design & Method

• Instrument. The survey was run in September/October 2025 in the School of Design of Politecnico di Milano. It adapted — with permission and under a Creative Commons license — the Australian student survey by Chung et al. for comparison, adding domain-specific questions on which phases of a design process students used GenAI for, and on perceived effects on Creativity and project ownership. It comprised 36 questions, offered in both English and Italian, and forms part of a multi-institutional project ("Student Perspectives on AI in Higher Education") with TU Delft and Aalto University.
• Sample. 280 students completed the questionnaire, spanning undergraduates and Masters students across all branches of design and all years of enrollment. This page reports the English version, answered by 223 students; the authors state a comparison with the Italian results showed no significant differences, and that the respondents included both domestic and international students. Item-level n varies (e.g. n=223 for some tables, n=231, n=228, n=219, n=218, n=202 for others).
• Journals, a second data source. In the 2025/2026 Research Methods course of the Masters in Product Service System Design, students kept a journal of their GenAI use on two group assignments (groups of 5): an in-depth secondary research report, and a primary research study of interviews, in situ visits and observations on the year's theme, "Water scarcity in Sicily". A form — designed by the author — broke each research type into 15 subtasks. Overall 100 students completed it; it was mandatory but its content was explicitly excluded from assessment and did not contribute to marks.
• Analysis. Statistical analysis tested for differences by gender, year of enrollment, design domain and language (finding none), and a correlation analysis examined ownership/creativity responses against reported AI-assisted activities. The journals were analyzed thematically to identify workflows and verification strategies.

Key Findings

• Use is frequent and routine. 71% of respondents used GenAI daily or weekly for academic purposes, and 59% regularly for personal use. 65% said they do not trust what GenAI generates, and 85% said they systematically modify its output to suit their needs. 70% reported being highly concerned about inaccuracy as a reason discouraging use.
• Activities cluster in research and writing. The most common uses were editing or improving writing (77.5%), finding information or conducting research (74.5%), brainstorming (68.0%), step-by-step teaching (64.9%) and summarizing notes or readings (61.0%). Lower down were generating code or formulas (38.1%), artwork and diagrams (32.9%) and completing part or all of an assignment (18.6%).
• Tools converge on a few platforms. ChatGPT was near-universal (206 of 513 tool mentions across 228 respondents), followed by Figma (65), NotebookLM (49), Perplexity (40), Microsoft Copilot (40), Midjourney (38) and Claude (38); specialized multimedia generators were named by very few respondents (7 non-users).
• Ownership and creativity are largely intact. Only 31% reported a perceived loss of ownership or creative agency when using GenAI; on "I feel that I am less creative since I started using AI" (n=219) 6.4% strongly agreed and 24.7% somewhat agreed, with 28.8% neutral. Neither research-related activities nor brainstorming correlated significantly with a sense of lost creativity or ownership — the sole activity significantly associated with such feelings was using GenAI for prototype generation.
• Journals confirm heavy secondary-research reliance and light primary-research reliance. Almost all reported phases showed near-universal adoption for finding initial information, summarizing papers, transcribing/summarizing interviews, and editing or polishing text; by contrast, developing an argument, planning and interpreting primary research, and structuring data collection were usually kept in students' own control. Students described cross-checking AI output with academic databases and institutional reports, and all said they manually verified AI-generated sources or summaries, often through group discussion.
• Concrete numbers on mistrust in the journals' domain. Students used GenAI for maps, diagrams, comparison tables and transcription, but did not use available tools for synthetic participants, personas or videos, despite these having been presented in class. The journals were filled unevenly, some highly detailed and others generic.

What this means for practice

• Learners. Verify every AI-supplied source yourself: the journals show students cross-checking output against academic databases and institutional reports and manually verifying AI-generated references and summaries, often through group discussion.
• Learners. Keep track of which phases you are delegating: students most often handed off editing (77.5%) and finding information (74.5%) while retaining planning, argument development and interpretation, and the authors read that split as a signal about which work the curriculum values.
• Learners. Give primary research the same standing as AI-assisted secondary research: adoption was near-universal for finding information, summarizing papers and polishing text, but light in developing and interpreting primary research.
• Learners. Move from consuming GenAI to making with it, through custom GPTs, low/no-code app creation or adapting open-source models, the shift the authors propose for design education.
• Learners. Re-examine your habits yearly rather than treating them as settled: the authors recommend repeating the survey as a longitudinal instrument because growing familiarity can slide into face-value acceptance of AI output.

Limitations

• Single-institution survey: 280 students completed the questionnaire at the Politecnico di Milano School of Design and this analysis reports the 223 English-version respondents, with no significant differences found by gender, year of enrollment, design domain or language.
• The journal data are self-reports gathered in a context of evaluation, so they carry self-validation and express what students take to be acceptable AI use for an assignment; the course itself framed experimenting with GenAI as acceptable as long as it was reported.
• The journals were filled unevenly, some highly detailed and others generic, and overall 100 students completed the form on two group assignments, so as a corpus they give only a partial picture of practice.
• Item-level n varies across the survey tables (n = 202-231) and the ownership, Creativity and Trust measures are self-report, so results such as the 31% reporting a perceived loss of ownership rest on different respondent subsets.

Citation

Stefana Broadbent, et al. (2026). A study of GenAI usage by Design Students: Analysis of Survey Results and Journals of AI practices at the Politecnico di Milano in 2025/2026. .