Gen AI creativity plateaus without 'idea co-development'

Generative AI tools can accelerate early-stage brainstorming, but that initial creative boost doesn't last unless workers fundamentally change how they interact with the technology, according to new research from Rice University.

A study published in Information Systems Research tracked participants across multiple creative tasks and found that people working with gen AI chatbots plateaued in creative output over time, while those working independently continued to improve. The difference came down to how users engaged with the technology—most treated AI as an idea vending machine rather than a collaborative partner.

The creativity plateau

Researchers led by Jing Zhou, the Mary Gibbs Jones Professor of Management at Rice Business, ran participants through 10 rounds of creative problem-solving on issues like climate change. Those using a custom gen AI chatbot initially scored higher on novelty and usefulness than solo workers. But by round seven, the human-only group had caught up, while AI-assisted participants showed no further improvement.

Analyzing hundreds of human-AI conversations revealed the problem: users simply prompted for more ideas rather than working to develop better ones. They proposed concepts, waited for responses, and moved on—a transactional pattern that failed to deepen creative output.

Why it matters

This research challenges the assumption that simply deploying gen AI tools will continuously improve workplace creativity. Organizations investing in AI collaboration need to invest equally in teaching employees how to use these systems effectively. Without training in co-development strategies, companies risk expensive AI implementations that deliver diminishing returns.

The co-development strategy

The researchers identified a specific interaction pattern that does sustain creative improvement: "idea co-development." In this mode, humans and AI exchange critical feedback, combine concepts, impose constraints, and iteratively refine single ideas into stronger solutions.

For example, a user might propose a concept, ask the AI to evaluate its practicality, revise based on that critique, add real-world constraints like budget or risk factors, and continue shaping the idea through multiple exchanges. This approach treats AI as a creative partner rather than an output generator.

When participants gradually increased their use of co-development strategies across tasks, their joint creativity improved significantly. But without explicit instruction, engagement in this pattern actually declined over time.

Training makes the difference

In a final experiment, researchers provided one group with explicit training in idea co-development techniques—how to exchange feedback and refine ideas interactively with AI. A control group received no such guidance.

The trained group showed significant improvement in creative output across tasks, while the untrained group continued to plateau. The intervention demonstrated that augmented learning—the process of humans and AI adjusting to each other to improve joint performance—requires deliberate skill development.

Zhou emphasized that companies need to assess how employees actually use AI tools and train them to move beyond simple prompting. Effective training should teach co-development strategies while encouraging workers to bring distinctly human strengths to the process: social judgment, ethical reasoning, and emotional intelligence.

The research was first reported by Rice Business Wisdom and suggests that organizations should design AI systems that prompt deeper engagement—asking questions and encouraging elaboration rather than just delivering answers. With proper training, gen AI can become a lasting driver of creative improvement rather than a novelty that loses its edge.

The findings were published in Information Systems Research and detailed by Rice University.