Why Organizational Hierarchy Determines How Far AI Research Actually Spreads

When new scientific research emerges, organizations often assume it will naturally circulate among employees. A new study conducted at the World Bank shows that ideas do not spread evenly inside large institutions, and the first recipient of the information can strongly influence how far it travels.

Researchers from the World Bank's Development Research Group, together with academics from the University of Warwick and Warwick Business School, examined how scientific evidence about generative artificial intelligence moves within a major global bureaucracy. Instead of focusing only on whether AI improves productivity or creativity, the team explored a more fundamental question: how do organizations actually learn?

A Large-Scale Test Inside the World Bank

To answer this question, the researchers carried out a large field experiment at the World Bank's headquarters in Washington, D.C., involving more than 6,000 employees across 511 divisions. In each division, one employee was randomly selected to receive summaries of leading academic studies on ChatGPT's impact at work. These employees were referred to as "seeds."

The studies shared with them showed that ChatGPT significantly improved writing productivity and quality, and in another experiment, enhanced creativity in storytelling tasks. After completing a survey, each seed received an infographic and a link to the full research paper. The researchers then monitored whether the seed intended to share the findings, whether colleagues clicked on the materials, and whether coworkers could recall details of the research two to three weeks later.

The Power of Position

The clearest finding was that hierarchy plays a decisive role.

When the initial recipient of the evidence was a senior employee, such as a Senior Economist or Director, the information spread further. Senior staff were more likely to report that they intended to share the research. Divisions with senior seeds recorded greater engagement with the study materials. Most importantly, colleagues of senior seeds were more likely to remember at least one specific detail of the research weeks later.

Overall recall levels across the organization were low. Only a small percentage of colleagues could remember even one notable feature of the study. Yet within this limited diffusion, the difference between senior and junior starting points was meaningful. Ideas gained more traction when they began higher in the organizational structure.

What Did Not Influence Sharing

The researchers also tested whether social norms or credibility signals would affect the spread of evidence. Some seeds were told that most colleagues were enthusiastic users of ChatGPT, while others were told that adoption was relatively low. Although these messages changed perceptions about peer behavior, they did not affect whether seeds shared the evidence or whether colleagues remembered it.

Similarly, emphasizing the research's credibility, such as publication in prestigious journals or affiliations with well-known institutions, had little impact on diffusion. Even when recipients rated the studies as slightly more credible, this did not lead to wider sharing.

In simple terms, peer approval and academic prestige did not drive the spread of information. Formal rank did.

Lessons for the AI Era

The findings come at a time when generative AI is already widely used inside the World Bank. A majority of surveyed employees reported having a ChatGPT account, and many had used it in their professional work. Support for these tools was relatively strong. Yet even in a highly educated and research-focused environment, most employees did not retain details of rigorous evidence shared just weeks earlier.

For organizations navigating rapid technological change, the message is clear. Evidence alone does not guarantee diffusion. Choosing the right messenger may be just as important as crafting the right message. Senior staff appear more likely to pass along information and more likely to be heard when they do so.

At the same time, the results highlight a potential bottleneck. If ideas spread more easily from the top, innovations emerging from junior employees may struggle to gain visibility. As artificial intelligence continues to reshape workplaces, understanding how knowledge flows within organizations may be as important as understanding the technology itself.