The United Nations is betting that Google can help it become the world’s AI-ready repository of official statistics. The UN announced Thursday that it is working with Google on the UN System Data Commons, a platform built on Google’s Data Commons open-source software. The project replaces the UNData portal, which required browsing through a traditional database interface.
The UN has placed its huge collection of official statistics under a standard known as the Model Context Protocol, or MCP, which enables AI systems to link directly to outside data sources. Twenty-six UN entities have signed on to the Data Commons, and data from nearly 20 will be available at launch. The organization’s stated goal is to bring 80% of its statistical datasets onto the platform by 2027.
The Accuracy Test
Prior to the announcement, UNICEF put AI models through a trial to see how accurately they could answer questions using its own data. The trial pitted six large language models against each other, gauging their performance across more than 133,000 responses tied to global development indicators. The average accuracy score came in at 21.2%.
The tested models included OpenAI’s GPT-4o and GPT-4o-mini, Anthropic’s Claude Sonnet 4.5 and Haiku 4.5, and Google’s Gemini 2.5 Flash and Gemini 2.0 Flash. About three in five responses did not provide a usable number, often because models hedged their answers. When the same questions were run again on the same model versions about two days later, models that gave a number both times returned the same number only about half the time.
This research piece is a working document from UNICEF that has not yet been reviewed by independent experts. It is currently being prepared for publication in a journal. The methodology, code, and data behind it will be made available separately.
The Data Behind the Door
UNICEF’s data website receives more than 6 million visits a month and is among the agency’s most popular sites. Referrals from users clicking links in ChatGPT answers to the site rose 67% year-over-year between January 1 and September 14. AI assistants overall now account for about one in 10 visits to the website.
An AI system linked to the UN data via MCP combined numbers on HIV infections, AIDS deaths, and lifespan into a single visual display showing how the U.S. President’s Emergency Plan for AIDS Relief has affected Africa.
Who Is Building What
Google.org provided $2 million in capacity-building funding and technical support to set up the platform’s core infrastructure. Prem Ramaswami, who leads Google’s Data Commons team, said the system is hosted on a UN-governed instance and is intended to eventually be maintained, operated, and scaled independently by the UN.
Shantanu Mukherjee, who leads the UN Statistics Division on an interim basis, described the platform as “orders of magnitude more advanced in scale, scope, and flexibility.”.
Ramaswami warned against “models can misinterpret nuance”, and he said that a person should always check AI outputs before citing or publishing them.
The Sequence of the Project
The arrangement follows a clear order:
- Google builds the core infrastructure with $2 million in funding.
- The system runs on a UN-governed instance.
- The UN intends to maintain, operate, and scale the platform independently.
The UNICEF accuracy test demonstrates the scope of the issue. Six models were involved, with 133,000 responses generated, and the average accuracy stood at 21.2%. These figures are far from reassuring. They signal what AI can currently accomplish with official data, which explains why the UN is moving quickly to make its data more accessible.
The UN Data Commons begins operation with nearly 20 datasets already made available. The stated aim is to make 80% of the UN’s statistical datasets accessible by 2027. It remains uncertain if the UN can reach that deadline, or if it can operate the system without outside help.
Where the paper stands
The paper backs the UN’s move to make its official statistics machine-readable for AI systems, and is against any licensing regime that would freeze today’s leaders in place and lock out smaller firms. The MCP standard, which lets AI systems link directly to outside data sources, is a step toward opening the UN’s vast trove of statistics to developers without requiring them to jump through costly regulatory hoops.
The UNICEF accuracy test shows the limits of current models: 21.2% correct across 133,000 responses, with many hedges and inconsistent results when the same questions were asked again. That is not a reason to slow down, but it is a reason to proceed with eyes wide open.
The $2 million from Google.org is real money, but the design calls for the UN to take over maintenance and scaling on its own. The paper wants to see that happen, rather than the UN depending on a tech giant for ongoing support.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.