The United Nations developed UN System Data Commons with Google to make global statistics more easily accessible to people and AI systems. Announced on Thursday, the platform uses Google’s open-source Data Commons infrastructure, enabling natural-language queries and the tracking of UN sources. The system replaces the UNData portal, which relied on traditional search and browsing. MCP support allows AI agents to connect directly to external data. The change is being linked to the growing shift toward generative AI and the problems large language models face in conveying reliable official statistics. According to a test reported by UNICEF Chief Statistician João Pedro Azevedo, the average accuracy of OpenAI GPT-4o and GPT-4o-mini, Anthropic Claude Sonnet 4.5 and Haiku 4.5, and Google Gemini 2.5 Flash and Gemini 2.0 Flash across more than 133 thousand responses remained at 21.2%. Approximately three-fifths of the responses contained no usable numerical value; two days later, only about half of the models arrived at the same figure. The research is not yet a final peer-reviewed study; the publication of the methodology, code, and data is planned.
Why it matters
The change in the platform makes access to United Nations statistics more direct not only for human users but also for AI agents that can connect to external data. This places greater emphasis on the traceability of the data source for researchers and developers working with official data, as well as users of generative AI. However, this new form of access does not by itself solve the problem of language models accurately conveying numerical information; existing test results show that even when models reach reliable statistics, their responses still contain a significant margin of error. The fact that the research has not yet been peer-reviewed and that the methodology, code, and data will be shared later leaves open the question of how the results will be assessed.
Background
Google is not a new name in the FikirPilot archive: we have published 40 articles mentioning the name in the past 90 days; the most recent is dated September 18, 2026.