Abstract
Although generative artificial intelligence models (GenAI models) can generate and interpret images, there is still limited research available to better understand the possible impact of these tools in higher education jewellery design studios. Against this backdrop, we addressed a practical curricular challenge: Students need opportunities to learn how to work with GenAI models, but these tools cannot be uncritically integrated due to possible AI plagiarism, reinforcement of stereotypes, weakened craft development, and ethical, legal, and intellectual-property concerns. We selected ChatGPT-4o as the focal model, because it could operate multimodally, understand conversation in context, generate images, and support dialogue across iterative design exchanges.
The primary purpose of this study was therefore not merely to test a technology, but to develop grounded guidelines for the responsible educational use of GenAI models in jewellery design studios. To achieve this, we answered three sequential research questions: (1) To what extent can ChatGPT-4o be useful during jewellery design? (2) Which factors may drive its use? and (3) How can responsible use be promoted in higher education studios?
The study is framed by an understanding of jewellery design as both an artistic and technical practice in which creativity, symbolism, cultural meaning, craftsmanship, ergonomics, and material knowledge must be integrated into wearable artefacts. Within this context, GenAI models cannot simply be treated as a neutral convenience. We situate ChatGPT-4o within wider debates about creativity, originality, authorship, stereotyping, cultural appropriation, plagiarism, and the legal status of AI-generated work. A particularly important conceptual move is the distinction between creativity and originality. We argue that AI-generated jewellery images may appear creative in the sense that they combine forms in new ways, but these designs are not necessarily original in a stronger design sense. Their value must instead be judged in relation to contextual suitability, stylistic repetition, practical feasibility, and the comfort with which the jewellery could actually be worn. This distinction allowed us to evaluate ChatGPT-4o not as an autonomous designer, but as a tool whose outputs require critical curation and informed human judgement.
Methodologically, the inquiry was conducted within a pragmatic, multiphase mixed-methods design. Ethical clearance had been obtained for the broader AI-related work, no human participants were involved, and the article emphasises transparency, critical verification, and source acknowledgement. We noted that ChatGPT-4o could be viewed in a dual role during the study: primarily as a tool that depends on user input to generate responses, but also because of its conversational behaviour, as a non-human research participant whose outputs could be interrogated, redirected, and critically assessed. The research was organised into three connected phases because each phase generated findings that informed the next. Phase 1 consisted of an exploratory case study in which the model was prompted to generate jewellery designs, engage in design-related dialogue, interpret stylistic instructions, respond to thematic and conceptual prompts, visualise rough sketches, and describe possible manufacturing processes for its own designs. In phase 2, we applied the adapted Artificial Intelligence Acceptance Prediction Model (AIAPM-II) to identify the factors likely to drive adoption and to calculate an acceptance prediction score. During phase 3 we used the combined findings of the first two phases to formulate guidelines for the responsible use of ChatGPT-4o in jewellery design studios. The overall design was therefore explicitly sequential, practical, and developmental: first exploring capability, then predicting uptake, and finally translating those findings into educational guidance.
During the qualitative phase, we collected both visual and textual data by means of five groups of prompts: material-based prompts, style-based prompts, thematic or conceptual prompts, prompts exploring collaboration with the designer’s rough sketches, and prompts investigating the extent to which the model could explain manufacturing processes for its own proposed designs. These outputs were then analysed using five criteria: originality and contextual appropriateness, the model’s usefulness at different phases of the design process, its ability to visualise and improve sketches, its capacity to provide workable manufacturing plans, and its ethical and practical implications, especially regarding stereotyping, cultural appropriation, and feasibility.
The first major finding is that ChatGPT-4o can generate jewellery designs and variations within seconds, which makes it potentially useful during the inspiration and idea-development phases of design. The model was able to produce visual responses rapidly, follow broad stylistic parameters, and generate alternative concepts that could stimulate discussion and refinement. In this sense, it can help students use time more efficiently and can broaden their exposure to visual possibilities, aesthetics, and stylistic directions. We show that ChatGPT-4o can be useful in establishing a “baseline” of design tendencies, enabling users to identify patterns, recurring motifs, and material-style associations that may then be questioned or deliberately redirected. These strengths support the argument that the model has educational value, not because it replaces design thinking, but because it can accelerate early ideation and provide a provocative starting point for critique and development.
However, we also found clear limitations. Although ChatGPT-4o could follow stylistic parameters, it also tended to stereotype. It frequently returned to familiar visual clichés, relied on repeated templates, and sometimes associated particular metals with predictable stylistic codes. In some cases, it produced implausible or flawed outputs: one design included nonsensical text, several outputs raised questions about wearability, and some technically specific requests were not interpreted accurately. Even where the generated images appeared attractive, the model could drift toward standardised or commercially familiar jewellery conventions rather than genuinely distinctive design solutions. These weaknesses matter educationally because they show that GenAI models can easily reinforce safe, formulaic, or culturally unexamined design habits unless its outputs are subjected to rigorous critique.
A further important finding is that these shortcomings do not necessarily make the model useless; rather, they create opportunities for the development of critical thinking and problem-solving skills. We demonstrate that follow-up prompts could guide ChatGPT-4o toward more innovative and wearable outcomes. Through iterative prompting, questioning, and refinement, the model could be nudged beyond weaker initial responses and led toward stronger conceptual or practical solutions. This suggests that the educational value of ChatGPT-4o lies partly in the dialogue it provokes. Students can learn to evaluate outputs, identify faults, refine prompts, and make defensible design decisions. The study therefore treats prompting as a meaningful studio skill. In this view, creativity does not reside only in the AI output itself, but also in the user’s ability to curate, redirect, and transform what the system produces.
Phase 2 strengthened the argument for responsible integration by showing that adoption is highly likely. Using the AIAPM-II, we found that all of the model’s major adoption factors could drive its use in jewellery design studios. ChatGPT-4o was judged to offer advantages over the status quo, to be easy enough to use, useful, testable, visibly productive, compatible with user needs, and cost-effective. Two factors, however, were only partially confirmed: low input and high-quality outputs. This qualification is important, because it indicates that although the model is accessible and attractive, skilled prompting and iterative refinement are still necessary if users want stronger results. Even with these caveats, the overall acceptance prediction score was 9 out of 10, which led to the conclusion that policy and strategy are necessary. In other words, the model is sufficiently usable and appealing that students are likely to adopt it, but that very likelihood increases the urgency of responsible educational guidance.
Therefore, we developed guidelines for responsible use during phase 3. These guidelines are grounded in the view that GenAI models should be treated as functional tools that can be integrated purposefully into curricula rather than tools to be used casually or invisibly. We emphasise transparency as a central principle: contributions made by GenAI models should be acknowledged and documented. It also stresses the need for critical evaluation of AI outputs, plagiarism prevention, cultural sensitivity, and the continued development of students’ own design judgement and craftsmanship. Students should not present AI-generated outputs as wholly their own work, and they carry the responsibility to ensure that final designs do not infringe copyright or intellectual property rights. The guidelines are also shaped by concern for maintaining studio learning outcomes that extend beyond finished artefacts, including reflective judgement, responsible decision-making, and the sustained development of manufacturing skills.
The study’s broader contribution lies in showing that the responsible educational value of ChatGPT-4o is conditional rather than absolute. The tool is not presented as a replacement for the designer, nor as a reliable source of originality, feasibility, or ethical judgement. Instead, it is positioned as a potentially valuable assistant during inspiration and concept refinement, provided that its outputs remain subject to expert scrutiny and pedagogical structure. The article contributes a baseline for future research by documenting how ChatGPT-4o performed in a jewellery design context at a particular moment in the evolution of GenAI models. Although ChatGPT-5 became available during the review process, the authors argue that this does not diminish the relevance of the study. On the contrary, the article establishes an important benchmark against which the performance of later models can be measured to revise guidelines if needed. As GenAI models are continuously changing, we conclude that the responsible-use guidelines must be reviewed regularly to remain valid.
Keywords: adapted artificial intelligence adoption prediction model (AIAPM-II); artificial intelligence adoption prediction model (AIAPM); artificial intelligence designs; artificial intelligence plagiarism; ChatGPT-4o; generative artificial intelligence (GenAI), higher education; jewellery design
- This article’s featured image was created byGoogle DeepMind and obtained from Pexels.