UN AI Hub stage
LeadersGoldDiscovery

Trusted global data commons using StatGPT

  • Date
    8 July 2026
    Timeframe
    15:00 - 15:30 CEST
    Duration
    30 minutes
    • Days
      Hours
      Min
      Sec

    Generative AI promises to transform how the world accesses economic data — but today’s models generate statistics from patterns, not from actual databases. IMF experiments show that leading AI chatbots return the correct published figure only 17–34 percent of the time. For monetary policy, fiscal planning, and development decisions, that gap is unacceptable.

    The Global Trusted Data Commons is a coalition-led initiative to fix this foundation. National and international statistical organizations — including the IMF, BIS, OECD, UN, World Bank, Eurostat, and national statistical offices — are making their authoritative datasets AI-ready by ensuring every dataset is API-accessible via SDMX, enriched with structured metadata, and tested for reliable natural-language retrieval before admission.

    StatGPT is the IMF-developed access layer built on top of the Commons. It retrieves data directly from official sources in real time — never generating or approximating figures. Every answer cites its source, eliminating hallucination. StatGPT is free, open-source, and globally available at statgpt.org, and through Model Context Protocol (MCP) connectors it enables Claude, ChatGPT, and Gemini to access the same trusted data.

    This session includes a live demonstration of StatGPT querying the Global Trusted Data Commons.

    Are you sure you want to remove this speaker?