Large language models (LLMs) are increasingly used as creativity support tools (CSTs) to assist users in generating new ideas. However, this study investigates whether LLMs actually support user creativity or instead homogenize creative outputs. Using a 36-participant comparative user study, the researchers compared ChatGPT and an alternative, non-AI CST (Oblique Strategies deck) to examine the homogenization effects of LLMs on human creative ideation. The study found that users of ChatGPT produced more homogenous sets of ideas at the group level compared to users of the non-AI CST. While individual-level homogenization was not observed, ChatGPT users generated more detailed ideas but felt less responsible for them. Additionally, ChatGPT users showed higher fluency, flexibility, and elaboration in their ideas compared to non-ChatGPT users. The study also found that LLMs may contribute to homogenization by providing similar ideas to different users, and by reducing the diversity of creative outputs due to the low inferential distance between LLM outputs and finished creative products. However, users may be able to resist homogenization effects if they are given information about the types of outputs the LLM tends to produce in a particular creative context. The study suggests that LLM-based CSTs may benefit from design strategies that introduce randomness or diversity into their outputs, such as using prompts that are less predictable or encouraging users to adapt their prompting strategies. Additionally, the study highlights the importance of considering the role of user adaptation in mitigating homogenization effects, as users may be able to resist homogenization by generating new ideas or modifying LLM outputs. Overall, the study provides insights into the homogenization effects of LLMs on human creative ideation and suggests potential design strategies for mitigating these effects in future LLM-based CSTs.Large language models (LLMs) are increasingly used as creativity support tools (CSTs) to assist users in generating new ideas. However, this study investigates whether LLMs actually support user creativity or instead homogenize creative outputs. Using a 36-participant comparative user study, the researchers compared ChatGPT and an alternative, non-AI CST (Oblique Strategies deck) to examine the homogenization effects of LLMs on human creative ideation. The study found that users of ChatGPT produced more homogenous sets of ideas at the group level compared to users of the non-AI CST. While individual-level homogenization was not observed, ChatGPT users generated more detailed ideas but felt less responsible for them. Additionally, ChatGPT users showed higher fluency, flexibility, and elaboration in their ideas compared to non-ChatGPT users. The study also found that LLMs may contribute to homogenization by providing similar ideas to different users, and by reducing the diversity of creative outputs due to the low inferential distance between LLM outputs and finished creative products. However, users may be able to resist homogenization effects if they are given information about the types of outputs the LLM tends to produce in a particular creative context. The study suggests that LLM-based CSTs may benefit from design strategies that introduce randomness or diversity into their outputs, such as using prompts that are less predictable or encouraging users to adapt their prompting strategies. Additionally, the study highlights the importance of considering the role of user adaptation in mitigating homogenization effects, as users may be able to resist homogenization by generating new ideas or modifying LLM outputs. Overall, the study provides insights into the homogenization effects of LLMs on human creative ideation and suggests potential design strategies for mitigating these effects in future LLM-based CSTs.