Evaluating an AI Evidence Intermediary to Improve Research Use in Sri Lanka’s Economic Policymaking
How to cite
Wong, Pui Hang; Silva, Tiloka de; Mahakalanda, Indra; Laar, Mindel van de; Gunasiri, Chethika; Moore, Lindsey. 2026. "Evaluating an AI Evidence Intermediary to Improve Research Use in Sri Lanka’s Economic Policymaking." AEA RCT Registry. AEARCTR-0018171. https://doi.org/10.1257/rct.18171-2.0
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
Governments and development agencies routinely conduct evaluations as part of their workflow. However, this knowledge may not feed back to future policymaking due to information overload. The rise of large language models (LLMs) and artificial intelligence provides a perfect solution to expedite policy learning. However, whether and how these AI agents facilities the use of evidence in policymaking remains untested. To address this question, we examine whether access to a generative AI platform powered by a domain-specific LLM improves the use of evidence in the policymaking process among government officials in Sri Lanka.
Intervention
Participating government officials will be granted access to a generative AI–powered evidence platform that enables users to search, retrieve, and interact with evidence from policy documents produced by the Sri Lankan government and evaluation reports from international development organizations such as USAID. Participants can make an inquiry, and the platform will provide responses based on the input reports and documents contained in the library. Core functions of the platform include: (i) search across a curated corpus using a domain-specific language model; (ii) assisted summarization through a chat interface; and (iii) transparent citation trails that allow users to trace major findings from source documents and search for information within them.
Experimental design
The target population consists of mid- and senior-level government officials who are responsible for drafting or reviewing policy documents, including policy briefs, policy reports, and strategy papers. 94 officials will be randomly assigned to either the treatment group or the control group, with 47 individuals in each group.
As part of a broader partnership with the Sri Lankan government, an additional group of 47 government officials will participate in a voluntary capacity-building program that includes access to the AI platform and complementary training activities. The impact of this capacity-building intervention will be evaluated separately using a mixed-methods approach combining quantitative and qualitative data.
Design details
Cited by
No citation data available for this registration yet.