UN adopts Google platform to make global data AI-ready

The United Nations has launched the UN System Data Commons, a new platform built on Google's open-source Data Commons technology. Supported by $2 million in funding and technical assistance from Google.org, the system replaces the legacy UNData portal. It allows users to query global statistics using natural language and supports the Model Context Protocol (MCP). This standard enables AI models to connect directly to external databases, retrieve verified statistics, and generate charts or dashboards. Currently, 26 UN entities have committed to the project, with data from nearly 20 available at launch, and a target to integrate 80 percent of the UN's statistical datasets by 2027.

The initiative addresses a severe accuracy deficit in how modern artificial intelligence handles global development data. A working paper by UNICEF evaluated six prominent large language models—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—across more than 133,000 responses. The benchmark revealed an average accuracy score of only 21.2 percent. Approximately three out of five answers failed to provide any usable number. Furthermore, when the same questions were asked again just two days later, the models returned the identical number only about half the time.

For developers and data practitioners, this integration bridges a critical gap as user behavior shifts toward conversational search. UNICEF reported a 67 percent year-over-year increase in traffic referred by ChatGPT between January 1 and September 14, with AI assistants now driving roughly 10 percent of all visits to its data website. By utilizing the new MCP-enabled platform, developers can build AI agents that pull directly from a UN-governed instance. This ensures that applications can trace every retrieved statistic back to its original source. However, because models can still misinterpret complex nuances, human oversight remains essential before publishing any AI-generated analysis.