The effect of response order and prompting on generative artificial intelligence creativity: An experimental study protocol

IntroductionAs generative artificial intelligence (GenAI) tools become more integrated into creative workflows, understanding how to optimize interactions with these systems to maximize creative output becomes increasingly important. While GenAI chatbots like ChatGPT have demonstrated high levels of creative ability on creativity tasks, there is limited empirical evidence on how to structure prompts and interact with GenAI to elicit their most original outputs. In humans, creative idea generation often follows a “serial order” effect, where later ideas in a brainstorming session tend to be more original than early ones. It remains unknown whether this phenomenon occurs in GenAI or whether repeated prompting enhances GenAI idea originality. This study investigates how the position of an idea in a response list and repeated prompting influence the originality of GenAI outputs on divergent thinking tasks.Methods We will evaluate nine leading GenAI chatbots (ChatGPT-4o, ChatGPT-o3, Microsoft Copilot, DeepSeek-V3, DeepSeek-R1, Gemini 2.5 Pro, Claude Sonnet 4, xAI’s Grok 3, and Perplexity AI) using the Alternate Uses Task. Each model will complete 50 independent sessions per task: each session will involve the generation of 25 original uses for an AUT item, with two further prompts to elicit further ideas. This will facilitate the study of the effect of response order and iterative prompting on GenAI idea originality. We plan to score the originality of each response using the Open Creativity Scoring with AI (Ocsai) tool. We will analyze the effect of response order and repeated prompting using repeated-measures ANOVAs and post-hoc comparisons. Ethics and dissemination The results of this experimental study will be shared in a relevant peer-reviewed journal and presented at local, national, and international conferences. As only data from GenAI will be used, we will not require ethics approval.