UN partners with Google to prepare global data for AI agents

The United Nations announced on Thursday that it is partnering with Google to streamline AI access to its extensive global statistical archives.

Named the UN System Data Commons, this new infrastructure leverages Google’s open-source Data Commons platform, enabling users to query statistics from various UN agencies using natural language. It supersedes the legacy UNData portal, which relied on traditional database interfaces. The updated platform also integrates the Model Context Protocol (MCP), a standard facilitating direct AI connections to external data repositories.

While users increasingly rely on AI for information, many systems still fail to consistently deliver authoritative data. During a virtual briefing, João Pedro Azevedo, UNICEF’s chief statistician, revealed that a benchmark of six large language models across 133,000 responses regarding global development indicators yielded an average accuracy of just 21.2%.

Azevedo noted that the evaluation included OpenAI’s GPT-4o and GPT-4o-mini, Anthropic’s Claude Sonnet 4.5 and Haiku 4.5, as well as Google’s Gemini 2.5 Flash and Gemini 2.0 Flash.

According to Azevedo, approximately 60% of responses failed to provide a usable figure, often due to models hedging their answers. Furthermore, when the same queries were re-run on identical model versions two days later, models that did provide a number matched their previous output only about half the time.

This study, currently a UNICEF working paper pending journal submission, has not yet undergone peer review. The organization intends to publish its methodology, code, and dataset alongside the final paper.

UNICEF has also reported a significant surge in traffic from generative AI assistants to its data portal, which attracts over 6 million monthly visits. Azevedo told TechCrunch that referrals from ChatGPT links to the site increased by 67% year-over-year between January 1 and September 14. These referrals represented 6.4% of all sessions this year, with UNICEF estimating that AI assistants now drive approximately one in ten visits overall.

The UN stated that 26 of its entities have committed to the Data Commons, with data from nearly 20 agencies available at launch. The organization aims to migrate 80% of the UN system’s statistical datasets to the platform by 2027.

UN System Data Commons.Image Credits:Google

“We are orders of magnitude more advanced in scale, scope, and flexibility, connecting for the first time across so many agencies across the UN system,” said Shantanu Mukherjee, acting director of the UN Statistics Division. “And [we are] taking this moment to also make our data AI-ready.”

Google.org contributed $2 million in capacity-building funding and technical support to establish the platform’s core infrastructure. Prem Ramaswami, who leads Google’s Data Commons team, told TechCrunch that the system is hosted on a UN-governed instance and is designed to eventually be maintained, operated, and scaled independently by the UN.

“We have taken a “train-the-trainer” approach throughout the rollout, and we have already seen the UN system team ramp up quickly,” Ramaswami said.

Google launched Data Commons in 2018 to organize public datasets from diverse sources into a unified framework. Last year, it added MCP support, allowing AI agents to directly query Data Commons for statistics and their sources.

The UN’s platform also tracks the origin of each statistic, allowing users to trace AI-retrieved data back to its original UN source. Azevedo emphasized the importance of this transparency as reliance on AI tools for information retrieval grows.

Beyond enabling AI agents to retrieve individual statistics, Google demonstrated how an MCP-connected system could aggregate multiple indicators to generate dashboards, charts, and written analysis without manual data combination.

In one demonstration, Google tasked an AI system with assessing the impact of the U.S. President’s Emergency Plan for AIDS Relief in Africa. The system identified relevant UN statistics on HIV infections, AIDS mortality, and life expectancy, producing an infographic based on these metrics.

However, providing AI systems with authoritative data does not guarantee authoritative conclusions. “Because models can misinterpret nuance, a human should always review the outputs before citing or publishing them,” Ramaswami said.