UN launches AI-ready data platform with Google to fix agent accuracy

The United Nations has launched a new platform designed to make its global statistics accessible to AI systems, addressing a critical accuracy problem that has plagued large language models attempting to answer questions about international development data.

The UN System Data Commons, announced Thursday and built on Google's open-source Data Commons platform, replaces the existing UNData portal with a system that supports natural-language queries and the Model Context Protocol (MCP), a standard enabling AI agents to connect directly to external data sources.

Why it matters

As more people turn to AI assistants for information, the inability of these systems to reliably surface authoritative data creates real risks for decision-makers, researchers, and the public. A UN benchmark revealing 21% accuracy for leading AI models on development questions underscores why structured access to verified sources matters — particularly for data that informs policy and humanitarian work.

The accuracy crisis

A UNICEF benchmark tested six major large language models across more than 133,000 responses to questions about global development indicators, producing an average accuracy score of just 21.2%, according to João Pedro Azevedo, the agency's chief statistician. The test covered 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 failed to provide a usable number at all, often because models hedged their answers. When the same questions were posed to the same model versions approximately two days later, models that provided numbers both times returned identical figures only about half the time.

The study is a UNICEF working paper being prepared for journal submission and has not yet been peer-reviewed. The organization plans to release its methodology, code, and data alongside the paper.

Growing AI traffic to UN data

UNICEF has observed a sharp increase in traffic from generative AI assistants to its data website, which receives more than 6 million visits monthly. Visits from users clicking links in ChatGPT answers rose 67% year-over-year between January 1 and September 14, according to Azevedo. Such referrals accounted for 6.4% of all sessions this year, while UNICEF estimates AI assistants overall now represent about one in 10 visits.

Platform scope and governance

Twenty-six UN entities have committed to the Data Commons, with data from nearly 20 available at launch. The organization aims to bring 80% of the UN system's statistical datasets onto the platform by 2027, according to Shantanu Mukherjee, acting director of the UN Statistics Division.

Google.org provided $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, said the system is hosted on a UN-governed instance and is intended to eventually be maintained, operated, and scaled independently by the UN.

The platform tracks the provenance of each statistic, allowing users to trace data retrieved by an AI system back to the original UN source. Google demonstrated how an AI system connected to UN data through MCP could pull together multiple indicators to generate dashboards, charts, and written analysis.

However, Ramaswami cautioned that authoritative data does not guarantee authoritative conclusions. "Because models can misinterpret nuance, a human should always review the outputs before citing or publishing them," he said.

These details were first reported by TechCrunch.