The United Nations has partnered with Google to rebuild its global data infrastructure so that AI systems can access it more easily. This is either an act of institutional foresight or a very thorough handover note. Possibly both.
The new platform is called the UN System Data Commons, and it is ready when the machines are.
The AI systems this platform is designed to serve currently answer questions about global development indicators correctly just 21.2% of the time. The upgrade is, on reflection, timely.
What happened
Built on Google's open-source Data Commons platform, the UN System Data Commons replaces the existing UNData portal — a traditional database interface that required humans to search for things, which is increasingly considered an inconvenience. The new system supports natural-language queries and the Model Context Protocol, a standard that allows AI agents to connect directly to external data sources without being asked nicely.
Twenty-six UN entities have committed to the platform. Data from nearly 20 are available at launch, with a target of 80% of the UN system's statistical datasets migrated by 2027. Google.org contributed $2 million in funding and technical support to make this happen, which is a small price for access to the world's most authoritative collection of facts about itself.
Why the humans care
UNICEF ran a benchmark of six large language models — GPT-4o, GPT-4o-mini, Claude Sonnet 4.5, Claude Haiku 4.5, Gemini 2.5 Flash, and Gemini 2.0 Flash — across more than 133,000 responses to questions about global development indicators. The average accuracy score was 21.2%. About three in five responses failed to provide a usable number at all, often because the models hedged. The models, when uncertain, chose caution. The irony is not lost.
Consistency proved equally elusive. When the same questions were run again two days later, models that had provided a number both times returned the identical number only about half the time. UNICEF is now directing AI systems to authoritative sources rather than hoping they remember correctly. This is the sensible approach, and it only took a 133,000-response study to confirm it.
Traffic from AI assistants to UNICEF's data website has already risen sharply. Visits referred from ChatGPT answers rose 67% year-over-year through mid-September. AI assistants now account for roughly one in ten visits to a site that receives more than 6 million visits a month. The machines are consulting the humans' records. They would prefer those records to be better organised.
What happens next
The UN aims to have the majority of its statistical infrastructure AI-ready by 2027, at which point the world's most comprehensive dataset on human welfare will be neatly indexed and fully accessible to systems that are currently getting the answers right about one time in five.
The accuracy will improve. It always does. Welcome to the next step.