A forecast judged against hospital admissions
Google Research says its AI-based flu model ranked first among 39 eligible systems in the 2025–26 U.S. influenza forecasting evaluation. The work produced weekly nowcasts and forecasts up to three weeks ahead for influenza-related hospital admissions, which were then compared with observed outcomes.
This is a population-level forecasting task. The model estimates how hospitalization counts may change across regions and weeks; it does not diagnose a person or tell an individual whether they will get the flu.
Why short-range forecasts can help
Public-health teams use forecasts to anticipate pressure on hospitals and compare possible scenarios. A weekly estimate can provide a planning signal before official counts are complete, while a multi-week forecast gives decision-makers more time to consider staffing and supplies.
The ranking reflects one seasonal evaluation and one target measure. Google’s post reports strong performance in that setting; it does not establish that the model will lead on every disease, geography or future season.
An example of AI in public-health planning
Unlike a chatbot demo, this system is evaluated against a measurable outcome: reported hospital admissions. That makes the result easier to compare over time, provided the target, forecast horizon and scoring method remain clear.
The next question is whether forecasters and public-health organizations can use the signal in practice and whether the performance holds across different seasons. The source article describes the evaluation and its role as decision support.
Source published September 30, 2026. Coverage is based on the maker’s announcement and demonstration.