Augmented (or Not): The Human–GenAI Tango in Generating Creative Ideas

This study addresses a burgeoning question: under what conditions does human–GenAI collaboration expand the solution space and produce higher-quality ideas in terms of novelty, utility, and feasibility, and what do humans still add to this process? This question is becoming increasingly important as recent studies on human–GenAI collaboration in creative and innovation processes show that the desired effect of augmentation is far from automatic. Adopting a search-based view of creativity, we address this question through an experimental study (n = 350) and focus on how different configurations of human–GenAI collaboration shape that search. The findings show that simply involving GPT-4 in idea generation, without changing how search is seeded or structured, does not improve idea quality in terms of novelty, utility or feasibility. Providing participants with an initial list of ideas generated by GPT-4 increases the perceived utility of final ideas but does not enhance novelty or feasibility. By contrast, seeding the search with GPT-4 ideas generated under transformational creativity principles significantly boosts both novelty and utility, at the cost of feasibility. Semantic similarity analyses further indicate that utility is predominantly shaped by GenAI’s initial suggestions, whereas divergence from these seeds reflects human contributions that matter more for novelty and, to a lesser extent, feasibility. Together, these results show that GenAI augments ideation mainly when its role in seeding and structuring joint search is deliberately designed, and they clarify the distinctive, complementary roles of humans and GenAI in shaping creative outcomes.