UN launches AI-ready data platform powered by Google to improve statistics accuracy

The United Nations announced a collaboration with Google this week to transform how its sprawling collection of international statistics can be accessed by artificial intelligence systems. The initiative aims to overcome a critical problem: AI models have proven remarkably unreliable at retrieving and presenting authoritative development data.

The accuracy problem#

A UNICEF study benchmarking six major language models revealed the scale of the challenge. Across more than 133,000 test responses to questions about global development indicators, the models achieved an average accuracy score of just 21.2%. The research examined models from OpenAI (GPT-4o and GPT-4o-mini), Anthropic (Claude Sonnet 4.5 and Haiku 4.5), and Google (Gemini 2.5 Flash and Gemini 2.0 Flash).

The problems extended beyond simple inaccuracy. About three-fifths of responses failed to provide any usable numerical answer, typically because the models qualified their statements with hedging language. Even more troubling: when the same questions were repeated on identical model versions just two days later, models that offered a number both times returned the identical number only about half the time, suggesting fundamental inconsistency rather than systematic bias.

What the new system offers#

The UN System Data Commons, built on Google's open-source Data Commons platform, replaces the legacy UNData portal with a modernized interface. The new system supports natural-language queries, allowing users to search UN agency statistics more intuitively than the traditional database browse-and-search approach.

Critically, it supports the Model Context Protocol (MCP), a standard that enables AI systems to directly connect to and query external data sources. This architectural choice means AI agents can retrieve individual statistics along with their source attribution, allowing users to trace AI-generated information back to original UN agencies. Google demonstrated the system pulling together multiple indicators to automatically generate dashboards, charts, and analytical summaries—a capability that previously required manual data aggregation.

Rollout and ambitions#

Twenty-six UN entities have already committed to the platform, with data from nearly twenty available at launch. The organization targets bringing 80% of the UN system's statistical datasets onto the platform by 2027. Google.org contributed $2 million in capacity-building funding and technical support, with the platform hosted on UN-governed infrastructure and designed for eventual independent operation and scaling by the organization.