Author: Dr Basil Mahfouz is a collective intelligence researcher at UCL developing AI-enabled systems to improve how complex decisions are made.
Policy decisions depend on scientific evidence. Climate change, pandemics, food systems, and artificial intelligence governance all rely on research to define what is credible and actionable. The challenge is not a lack of science, but which science is selected and which is missed. The processes that shape this selection matter.
Across most domains, policymakers face more research than they can process. Decisions must be made under time pressure and uncertainty. To manage this, policy systems rely on established pathways that filter which evidence is seen, trusted, and used. This filtering is necessary. It is also a source of risk.
Our recent analysis of more than 230,000 scientific papers cited in intergovernmental organisation documents shows a high level of concentration. Across 23 fields, between 0.7% and 4.4% of scientists produce around 30% of the research used in global policy. These scientists are embedded in dense international networks and often linked to advisory processes. Their work is taken up quickly and cited across multiple organisations, even when substantively similar research by others is available.
This structure may help policy systems operate quickly. It creates shared reference points and allows rapid synthesis of complex evidence. But it also constrains what is visible. Systems become better at recognising familiar forms of expertise than identifying relevant but less connected research. Initial analyses suggest that they are also less likely to detect emerging topics, early findings, or research that does not yet fit established policy frameworks.
If concentration reflected only research quality, this would be less concerning. However, evidence suggests otherwise. Standard indicator of research quality, such as citation impact and journal metrics, only weakly explain which research is used in policy. More importantly, when comparing highly similar studies, differences in uptake remain. Relevant evidence can be available but not selected.
This means policy systems are not only filtering noise. They are also missing signal.
This pattern is not temporary. It is persistent. Even as institutions attempt to broaden participation and diversify expertise, the underlying structure of knowledge flows remains stable over time.
From a risk perspective, this matters. Policy systems depend on evidence to identify and manage risk. If the evidence base is narrow, decisions are more likely to be based on incomplete or outdated information. Systems become better at recognising familiar problems and less capable of detecting emerging ones.
This has already been observed in practice. In fast-moving domains, policy can rely on older research even when newer relevant evidence is available. This reflects a broader limitation. Policy systems do not always incorporate the full range of knowledge available at the time decisions are made.
In complex domains, this creates a structural vulnerability. Climate, health, food systems, and inequality are interconnected. Understanding these interactions requires integrating diverse forms of knowledge. When evidence pathways are narrow, this integration becomes harder, and blind spots become more likely.
The question is therefore not only whose science counts. It is which knowledge remains unseen.
Sources:
Mahfouz, B., Capra, L., & Mulgan, G. (2026). When evidence exists but is not used: Diagnosing policy responsiveness in COVID-19 education policy (Version 1). Research Square. https://doi.org/10.21203/rs.3.rs-8642639/v1
Mahfouz, B., Capra, L., & Mulgan, G. (2025). Uncovering drivers of climate research in policy with pretrained language models. Patterns, 6, Article 101342.
Asatani, K., Iwata, Y., Tomokiyo, Y., Mahfouz, B., Yarime, M., & Sakata, I. (2026). Structure of scientific knowledge flows to intergovernmental organizations. Proceedings of the National Academy of Sciences, 123(17), e2514861123. https://doi.org/10.1073/pnas.2514861123
Mahfouz, B., Capra, L., & Mulgan, G. (2025). Assessing the influence of research quality on policy citations: Quantitative analysis finds non-academic factors more likely to influence how papers get cited in SDG policy. Sustainable Development, 33(2), 1848–1860. https://doi.org/10.1002/sd.3214