The United Nations has been publishing its statistics for decades through a portal built for a different era, one where a researcher clicked through categories to find a number. On September 17, the organization replaced that structure with something built to be read by machines, and it turned to Google’s technology to do it.
The new system is called the UN System Data Commons, and it is rebuilt on Google’s open-source data platform. The pitch is plain: a user can ask for a figure in natural language, and the system returns it. Underneath, the platform supports the Model Context Protocol, the standard that lets AI systems connect directly to a data source, which means a model can pull a statistic the same way a researcher would, but faster.
The change replaces the old UNData, a portal that required a user to navigate through layers to find what they were looking for. The new version is built for the opposite behavior: a question in, a number out. The difference is not cosmetic; it is the difference between a library and a database.
The push came with a set of numbers that explain the urgency. UNICEF ran a test in which six major AI models answered more than 133,000 questions about global development indicators. The models’ average accuracy was 21.2 percent. Roughly three out of every five answers offered no usable figure at all, and when the same model was asked the same question two days later, its answer was consistent only about half the time.
Those results are the argument for the whole project. The UN’s statistics are only useful if the systems that people increasingly use to find information can actually retrieve them. If the models get the numbers wrong, or make them up, or change them between days, then the statistics might as well not exist for the people who rely on AI to find them.
The scale of the rollout is already substantial. Twenty-six UN entities have signed on, and nearly twenty of them were connected at launch. The organization has set a target of covering 80 percent of its statistical datasets by 2027, a goal that turns the launch from a pilot into the first step of a migration.
Google’s role is the structural choice at the center of the project. The company’s Data Commons is open source, which means the UN is not handing its data to a vendor but building on a shared platform it can inspect and control. The Model Context Protocol is the same kind of open standard, which is why the UN can present the project as a public good rather than a private arrangement.
The choice of Google still carries weight. A major multilateral organization building its statistical infrastructure on an American technology company’s platform is a decision with political dimensions, and the UN has been careful to frame it as a commitment to open standards rather than to a vendor. The data remains the UN’s; the plumbing is borrowed.
The practical effect is a quieter one. A researcher who used to spend an afternoon digging through UNData can now get the figure in a query, and an AI model can get it in a call. The statistics become more available, which is the point of publishing them in the first place. The test results showed what happens when they are not available; the new platform is meant to fix exactly that.
The UNICEF test deserves more than a passing mention, because it explains why an international bureaucracy moved this quickly. A model that answers 21 percent of development questions correctly is worse than useless; it is a source of confident misinformation, and the people who rely on it, journalists, policy analysts, aid workers, would have no way to tell the good answers from the bad. When the same model contradicts itself within two days, the problem compounds, because the user has no way to know which of the two answers is the error.
The Model Context Protocol is the piece that makes the project durable. By building the data commons to speak that protocol, the UN is making the statistics available to any AI system that knows how to ask, not simply publishing a website. That is a bet on a future where people find numbers through models rather than through portals, and the UN has decided to meet that future rather than resist it.
The 80 percent target is the metric that will define success or failure. Covering most of the system’s statistical datasets means the data commons becomes the default source for the UN’s numbers, which is what would actually change the accuracy of the models the UNICEF test measured. If the target is hit, the test’s numbers should improve; if it is missed, the models will keep guessing, and the statistics will keep sitting where the models cannot reliably reach them.
Google’s involvement will draw scrutiny from the member states that are wary of concentrating UN infrastructure in the hands of a single American company. The open-source nature of the platform is the answer to that scrutiny, and the UN has emphasized it at every turn. But the scrutiny will not disappear because the code is open; it will only be answered when the system proves that the data remains neutral, accessible and under the organization’s control.
