Since the release of large language models such as ChatGPT, researchers and practitioners alike have rushed to examine whether working with generative AI actually improves creative performance. On the surface, the promise seems obvious: AI systems can rapidly generate text, images, and ideas, suggesting a natural complement to human creativity. Yet, despite a growing body of empirical research over the past several years, the evidence remains strikingly inconsistent.
Across studies examining human–AI collaboration in creative tasks – from ideation and writing to art and design – findings are mixed. Some studies report that human–AI collaboration enhances creativity, others find no effect, and still others show that jointly produced outputs are less creative than those generated by humans alone. Importantly, these contradictory results emerge even when researchers study similar tasks and use comparable measures of creativity. Taken together, the literature has yet to reach a credible consensus on whether, when, or why working with AI improves creative outcomes.
Our research suggests that these inconsistencies are not accidental. Rather, they stem from a shared assumption underlying most prior studies: that humans and AI will naturally figure out how to collaborate effectively. As a result, much of the existing research focuses on single-round interactions, implicitly treating creativity as an outcome of using AI rather than as a process shaped by how collaboration unfolds over time. This approach overlooks a critical question of how humans and AI learn to work together creatively.
We argue that the problem is not whether AI is capable of contributing to creativity, but whether human–AI collaboration is designed to support it. Without deliberate structure, repeated collaboration does not automatically lead to improvement. In fact, our studies show that joint creativity often stagnates over time unless organizations intervene in how humans and AI co-create. This insight shifts attention away from the debate over whether AI helps or hurts creativity and toward a more actionable question for leaders: how should human–AI collaboration be designed so that creative performance improves rather than plateaus?
We found three distinct activities that are particularly ripe for human–AI collaboration:
Responsive refinement, which is where humans generate ideas, and AI provides feedback on those ideas. This approach takes advantage of AI’s ability to rapidly spot practical constraints, market parallels, or implementation challenges that humans might miss.
Generative expansion, which is when humans provide direction, and AI generates new ideas. This approach can work well when humans provide very specific prompts, often building on what has gone before, so that the AI doesn’t converge on safe and practical solutions.
Bidirectional development, which is when both humans and AI are offering suggested improvements. Humans critique and reshape AI suggestions while AI analyzes and enhances human concepts. This is where the real augmentation happens, but it requires the most deliberate structure.
Each of these approaches requires us to tailor our behavior. For instance, in “responsive refinement,” we need to present ideas clearly so that AI can provide meaningful feedback. In “generative expansion,” meanwhile, we need to be on alert for any convergence towards standard solutions from the AI. Lastly, in “bidirectional development,” we need to treat AI responses not as final outputs but as material for further refinement.
“Netflix provides a good example of these creative production workflows.”
So, what can managers take from these findings? We think our research offers a roadmap that directly counters some of the perception challenges identified in previous studies into human-AI collaboration.
The first step is to understand that simply giving people AI tools won’t result in productivity improvements or new innovations. Instead, we need to provide people with templates and workflows that help to guide the co-creation process. It’s especially important to highlight that human–AI co-creation is a long-term process rather than a once-and-done idea generation. For instance, managers might establish a clear sequence where AI generates, humans critique and redirect, AI refines, and humans synthesize.
The next step is to really reinforce that this is an ongoing collaborative process. For collaboration to be effective, both humans and AI must learn about one another, a process that occurs over time and through repeated engagements. You should structure projects so that teams can work with AI over time and build a shared context and a collaborative cadence with the technology.
The third step is to train people so that they develop these collaboration skills. Too often, people view AI as some kind of idea vending machine, whereby they input a prompt and receive an output that is then either used or discarded. The real value comes when we use AI repeatedly. As such, you should train employees to critique AI suggestions productively, to identify which elements to keep and which to challenge, and to guide AI toward unexplored territory rather than letting it converge on safe answers.
The final step is to break down creative tasks, as this allows a clear assignment of bits of the process to humans or AI. This is likely to include distinct tasks for idea generation, critique, synthesis, and refinement. These tasks should be assigned based on complementary strengths. For instance, humans tend to thrive at recognizing novel connections and judging true innovation, while AI excels at rapid iteration and identifying feasibility constraints.
Netflix provides a good example of these creative production workflows. Humans and AI at the firm collaborate on scripts and content generation. The company breaks down script development into a number of sub-activities, such as idea generation, evaluation, and selection, and these subtasks each involve a varying level of human or AI input. For instance, humans are key to generating new storylines and early drafts, whereas AI plays a bigger role in character development and narrative pacing. These insights help to guide refinement and positioning decisions around content.
Managers can reinforce this shift by making the co-creation process visible.
Perhaps the most important task, however, is to counter the narrative that AI-augmented creativity is beneath or above human creativity. It’s not a matter of taking sides in the man vs machine debate but rather redefining just what creativity can look like in the age of AI. Our research shows that creative performance depends on how collaboration between humans and AI is structured over time.
Rather than debating outcomes in isolation, organizations should focus attention on the process of co-creation. When human–AI collaboration is treated as a one-off interaction, creative performance tends to stagnate. When it is designed as an ongoing, structured process – where humans deliberately guide, critique, and refine ideas in collaboration with AI – creative outcomes improve over time. What matters, then, is not the presence of AI, but whether teams are supported in engaging in effective idea co-development.
Managers can reinforce this shift by making the co-creation process visible. Highlighting how ideas evolve across multiple rounds – how humans steer direction, apply judgment, and set constraints while AI contributes generative input and analytical support – helps teams understand what effective collaboration actually looks like. This reframes AI not as an autonomous creator or a replacement for human creativity, but as a component within a human-directed system designed to sustain creative improvement.
AI has the potential to enhance our creativity, but it’s equally clear that we need to deliberately design that collaboration for it to work.
The evidence is clear. AI has the potential to enhance our creativity, but it’s equally clear that we need to deliberately design that collaboration for it to work. This requires moving beyond the simplistic question of whether AI is creative to the more productive question of how we structure human–AI partnerships for creative augmentation.
Organizations that get this right will have a significant advantage. By deliberately designing human–AI collaboration, they can explore a wider solution space, iterate more effectively on promising ideas, and achieve sustained improvements in joint creativity over time. In contrast, organizations that treat AI as a plug-and-play creativity tool often see creative performance plateau. Without structured guidance for idea co-development, repeated human–AI collaboration fails to produce learning, limiting both the quality of outcomes and the return on investment in AI-enabled work.
The choice isn’t between human creativity and AI efficiency. It’s between intentional design of human–AI collaboration and hoping that creativity emerges spontaneously. Our research shows that hope isn’t enough. But with the right structure, guidance, and ongoing support, human–AI pairs can achieve something neither could accomplish alone: sustained creative augmentation that gets better over time.
The question for leaders isn’t whether to use AI in creative work. The question is whether you’ll invest in helping your people learn to truly collaborate with it. The creativity of your organization may depend on the answer.
Amid widespread predictions of the bursting of the AI bubble, José Parra Moyano demonstrates how leaders can focus on optimizing initiatives with strong value, data readiness, and employee endorsement. ...
Confusing efficiency with productivity is to mistake speed for direction, and execution for value. Hamilton Mann explains how to avoid the pitfalls in your AI transformation...
Audio available
To create the best customer experience, CXOs should leverage AI to build dynamic systems that adapt in real time to customer needs. We explore how leading companies are using this approach today....
For multinationals, especially in strategically sensitive sectors such as energy, finance, and technology, traditional defense policies offer inspiration as they develop robust AI strategies. ...
Explore first person business intelligence from top minds curated for a global executive audience