The UN and Google are making global statistics AI-ready, using structured data, MCP, and provenance to help AI systems access and trace authoritative sources.
The United Nations is now working with Google to address that gap through the UN System Data Commons, a new platform that brings statistics from multiple UN agencies into a common, machine-readable environment.
Launched on September 17, 2026, the platform is built on Google’s open-source Data Commons and supports natural-language search as well as the Model Context Protocol (MCP). It is intended to give users a simpler way to explore UN statistics while allowing AI agents to retrieve authoritative data directly.
The data problem behind AI answers
AI systems can produce convincing answers even when the underlying information is incomplete or incorrectly interpreted. That becomes a bigger concern when the subject is development data, public health, poverty, education, or other areas where figures can vary depending on definitions, geography, and time period.
The UN’s data is already widely used for such analysis, but much of it has historically been distributed across individual organisations and platforms.
The new Data Commons is designed to bring those datasets into a connected framework. Google says the platform automatically integrates metrics, timelines, and geographic boundaries, allowing users to work across datasets rather than manually joining information from separate sources. That changes the task from simply finding a statistic to understanding how different indicators relate to one another.
Making UN statistics easier for AI to use
The platform replaces much of the traditional database-search experience with natural-language exploration. A user can ask questions in ordinary language rather than first learning how a particular statistical database is organised.
The bigger change, however, is aimed at AI systems. Through MCP, AI agents can connect to the UN System Data Commons and retrieve statistics directly. Google demonstrated how an AI assistant could combine multiple UN indicators and turn them into charts, graphs, infographics, or written analysis.
This matters because the model does not have to rely solely on information embedded in its training data. Instead, it can retrieve current information from the underlying data source when responding to a request. For organisations using AI for research, that distinction can reduce one of the risks associated with relying on static model knowledge.
Provenance becomes part of the AI workflow
Access to authoritative data does not automatically make an AI-generated answer accurate. The UN System Data Commons therefore includes information about the origin of individual statistics, allowing users to trace figures back to their underlying sources. Google explicitly advises users to review the underlying sources before citing critical figures, even when AI is working with grounded data.
Prem Ramaswami, Head of Data Commons at Google, also stressed the continuing role of human review because models can misinterpret nuance.
That point is important for enterprise and policy users. Connecting an AI system to an authoritative database can improve the quality of the information available to the model, but it does not remove the need to check how that information has been interpreted.
The scale of the UN data challenge
The initiative is being developed as a shared platform rather than a single new database containing every UN dataset. At launch, 26 UN entities had committed to the initiative, with data from nearly 20 agencies available. The UN’s longer-term goal is to bring 80% of statistical datasets from the UN system onto the platform by 2027.
The distinction matters because the challenge is partly technical and partly organisational. Making data AI-ready requires different agencies to expose information in formats that can be connected, understood, and reused consistently.
The UN’s own work on the Data Commons has identified interoperability, harmonised metadata, data governance, and sustainable infrastructure as important parts of making official statistics ready for modern digital and AI tools.
Why AI needs better public data infrastructure
The need for this infrastructure becomes clearer when AI performance is measured against factual development questions. A UNICEF working paper cited in the supplied material found that six AI models achieved an average accuracy of 21.2% across more than 133,000 responses involving development indicators. The research was described as a working paper and had not been peer-reviewed.
The finding does not mean that AI systems are broadly inaccurate across all tasks. It does, however, illustrate a specific problem: general-purpose AI models can struggle when asked to retrieve and interpret detailed statistical information. The Data Commons approach is designed to address part of that problem by moving the source of truth closer to the AI workflow.
From search engine to data interface
The broader change is therefore not simply about replacing the UNData portal with a newer search interface. It is about preparing official data for a world in which AI systems increasingly sit between people and information.
A researcher may ask an AI assistant to compare health indicators across countries. A policymaker could request a combination of demographic and economic statistics. An analyst could ask an agent to build a visualisation from several datasets.
In each case, the quality of the answer depends partly on whether the AI can access structured, authoritative, and traceable information. That makes data infrastructure an increasingly important part of AI infrastructure.
The next test is consistency
The UN System Data Commons gives AI systems a more direct route to authoritative statistics, but its longer-term value will depend on how widely the system is adopted, how consistently datasets are structured, and how effectively provenance and governance are maintained as more agencies come on board.
The UN’s target of covering 80% of its statistical datasets by 2027 points to the scale of that task. The larger shift is already clear, though. As AI becomes an interface for research and decision-making, governments and institutions will need to rethink how their public data is published. Making information available online was the first step. Making it machine-readable, interoperable, traceable, and usable by AI is becoming the next one.