The United Nations is overhauling how the world accesses global statistics, partnering with Google to launch the UN System Data Commons. This new platform replaces the clunky UNData portal with a system built on Google’s open-source infrastructure, allowing users to query data from multiple agencies using natural language. It supports the Model Context Protocol (MCP), a standard that lets AI agents connect directly to external data sources. The move addresses a critical reliability gap in generative AI. João Pedro Azevedo, UNICEF’s chief statistician, revealed that a recent benchmark of six major large language models yielded an average accuracy score of just 21.2% when answering questions about global development indicators. The test included OpenAI’s GPT-4o, Anthropic’s Claude Sonnet 4.5, and Google’s Gemini 2.5 Flash. Roughly 60% of responses failed to provide usable numbers, often due to model hedging. Even when models did produce figures, consistency was poor. Identical queries run two days apart returned the same number only half the time.
Traffic patterns confirm the urgency. UNICEF’s data website sees over 6 million monthly visits, with referrals from AI assistants like ChatGPT jumping 67% year-over-year between January and mid-September. These AI-driven sessions now account for about 10% of all traffic. The new Commons aims to channel this demand toward authoritative sources. Twenty-six UN entities have committed to the project, with data from nearly 20 available at launch. The goal is ambitious: migrate 80% of the UN system’s statistical datasets to the platform by 2027. Shantanu Mukherjee, acting director of the UN Statistics Division, called it a leap in scale and flexibility, finally connecting disparate agencies into a single, AI-ready framework.
Google.org contributed $2 million in funding and technical support, but Prem Ramaswami, who leads Google’s Data Commons team, emphasized that the UN will eventually operate the system independently. The platform tracks data provenance, allowing users to trace AI-generated answers back to original UN sources. In a demonstration, an AI agent used MCP to pull HIV infection rates and life expectancy data, automatically generating an infographic on the impact of AIDS relief programs in Africa. Yet, access to clean data doesn’t guarantee sound reasoning. Ramaswami warned that models can still misinterpret nuance, meaning human review remains essential before citing any AI-derived conclusions.
