Negotiating Creative Agency with Generative AI in Design Learning: A Multilevel Meta-Analysis and Meta-Thematic Synthesis of Design Creativity, Authorship, and Reflection (2020–2026)

Negotiating Creative Agency with Generative AI in Design Learning Purpose and Scope This study investigates how generative artificial intelligence (GenAI) influences creative thinking and creative agency in higher education design disciplines. As GenAI tools—including text-to-image systems, large language models, and multimodal platforms—become embedded in design practice and studio pedagogy, questions arise about whether these technologies augment or attenuate design creativity. The research addresses a critical gap in the literature: while evidence on GenAI's effects remains fragmented and theoretically underdeveloped, the field lacks integrated understanding of the mechanisms through which GenAI shapes creative cognition, authorship, and professional identity formation. Methodology The study employs a comprehensive three-stage research design: PRISMA-compliant systematic review of empirical research published between January 2020 and March 2026 Multilevel meta-analysis of quantitative creative thinking outcomes Meta-thematic synthesis of qualitative findings Searches across ten databases and grey literature sources identified 24,318 initial records, with 412 studies meeting inclusion criteria. The meta-analysis encompassed 138 studies with 347 independent effect sizes (N = 38,247), while the meta-thematic synthesis analyzed 224 qualitative studies. Key Findings Quantitative Results: GenAI-supported instruction produced a moderate positive overall effect on creative thinking (g = 0.47, 95% CI [0.36, 0.58], p < .001) Substantial heterogeneity was observed (I² = 81.4%) Dimensional analysis revealed an efficiency-originality divergence: effects were significantly stronger for elaboration (g = 0.72) and fluency (g = 0.58) than for flexibility (g = 0.31) and originality (g = 0.22, n.s.) Significant Moderators: Structured prompt engineering training (g = 0.63 vs. 0.31 without) Reflective scaffolding (g = 0.59 vs. 0.29 without) Collaborative AI roles (g = 0.64 vs. 0.28 as shortcut) Intervention duration of 4–8 weeks (g = 0.61) Higher baseline creative self-efficacy (g = 0.61 vs. 0.27) Qualitative Mechanisms:Thematic synthesis identified three cross-cutting negotiation processes: Attributional negotiation – how designers assign creative credit between themselves and AI Scaffolding negotiation – calibration of AI reliance based on task demands and metacognitive awareness Identity negotiation – reconciliation of professional design identity with distributed creative agency Theoretical Contribution The study advances a Human-AI Creative Agency Negotiation (HCAN) framework that synthesizes AI-TPACK, reflective practice, distributed agency, and designerly ways of knowing. The framework explains why GenAI redistributes rather than simply enhances or erodes creative agency, and reframes design pedagogy from tool-centric training toward cultivation of metacognitive, reflective, and adaptive creative competencies. Practical Implications The findings support pedagogical approaches emphasizing: Structured prompt engineering as metacognitive scaffolding Reflective documentation and process-visible assessment Collaborative AI integration positioning AI as partner rather than shortcut Sustained engagement over 4–8 week intervention periods Limitations The review acknowledges predominance of studies from China and the USA, underrepresentation of fashion and architecture disciplines, heterogeneity of outcome measures, and reliance on short-term, course-specific studies with self-report data.