TLDR: A comprehensive meta-analysis of 17 experimental studies conducted between 2022 and 2025 has found no significant evidence that generative artificial intelligence consistently outperforms humans in generating original or useful creative ideas. Despite the growing adoption of GenAI tools, the research indicates that human creativity still holds its ground.
Despite the considerable hype surrounding generative artificial intelligence (GenAI) and its perceived creative capabilities, a recent meta-analysis spanning 17 experimental studies from January 2022 to January 2025 reveals that these advanced systems have not yet demonstrably surpassed human creativity. The meta-analysis, which included 115 effect sizes comparing human and GenAI-prompted creative idea generation, found only a small, non-significant pooled effect size favoring GenAI.
Initial analyses within the meta-analysis did suggest a greater degree of originality in ideas generated by GenAI. However, a crucial sensitivity analysis indicated that this observation was largely driven by a few outlier studies that reported exceptionally large effects. When these influential studies were accounted for, no significant differences in originality or usefulness emerged between human-generated ideas and those produced by prominent GenAI models, including GPT-3, GPT-3.5, and GPT-4.
The researchers meticulously checked for publication bias using funnel plots, asymmetry tests, and p-curve analysis, finding no evidence of such bias. This strengthens the validity of the findings, suggesting that the results are a true reflection of the current state of GenAI’s creative performance relative to humans.
The implications of these findings are significant, prompting theoretical and methodological questions about the fundamental nature of GenAI’s creative output. The study suggests that rather than generating genuinely novel and contextually situated ideas, GenAI may primarily retrieve statistically likely exemplars of creativity based on its vast training data. This perspective underscores the unique dimensions of human creativity, which are often embedded in sociocultural and sociotechnical contexts, emphasizing ethical sensitivity and interpersonal grounding—qualities not yet fully captured by current GenAI outputs.
Looking ahead, the authors advocate for future research to explore GenAI-powered creativity in more ecologically valid settings. This includes a focus on co-creative human-AI processes, where factors such as task design, participant diversity, and evolving cultural frameworks are acknowledged as key influences on how creativity is evaluated and fostered.
Further supporting the notion of human creative resilience, a separate study published in PsyPost on July 25, 2025, highlighted the enduring power of human collaboration. This experiment found that pairs of humans working together consistently produced more original ideas than individuals collaborating with artificial intelligence or relying on internet search tools. The findings from this study suggest that human interaction retains a distinct creative advantage, particularly in generating novel concepts.
Participants in human-to-human collaborative settings also reported increased confidence in their own creative abilities. As the researchers of that study concluded, “creativity is a unique human endowment that is not easily replicated by AI.”
Also Read:
- Human Creativity Remains Paramount Amidst AI-Driven Content Deluge, Bain & Company Report Finds
- Generative AI Workshops Drive Transformative Change in Business Operations
While another experiment conducted at Bath Spa University RAK showed a slight preference for AI-generated ideas (52.9%) over human-generated ones (47.1%) for a specific prompt (‘Activities for Freshers’ Week 2025′), it also noted an equal representation of both in the top five ideas. This study also revealed that approximately 62% of respondents harbored concerns about generative AI posing long-term risks to human creativity. These varied findings underscore the complex and evolving landscape of human-AI creative interaction, but the overarching meta-analysis provides a grounded perspective on GenAI’s current capabilities.
