A Google AI-powered model outperformed 38 other eligible forecasting models to become the top-performing system for predicting flu-related hospital admissions during the 2025–2026 influenza season, according to an evaluation published by the Centers for Disease Control and Prevention (CDC) on September 30, 2026.
The CDC’s FluSight initiative collects weekly forecasts from government agencies, universities, and private companies to predict influenza-related hospital admissions across the United States. These forecasts help public health officials and hospitals prepare for potential increases in patients, allocate medical resources, and respond to outbreaks before they become overwhelming.
This season, Google’s AI-powered model, known as Google_SAI-FluEns, ranked first among 39 eligible models, demonstrating the growing potential of artificial intelligence in public health research.
How Google’s AI Predicts Flu Hospitalizations
Traditional disease forecasting relies heavily on statistical models, historical infection patterns, and epidemiological expertise. Researchers analyze past outbreaks, seasonal trends, and hospital admission data to estimate how a disease might spread.
Google’s approach introduces another dimension: artificial intelligence that can help create and improve the forecasting algorithms themselves.
Google developed its forecasting system using technology associated with Empirical Research Assistance (ERA), an AI research tool capable of generating and optimizing algorithms for scientific problems.
Rather than relying on a single forecasting approach designed entirely by human researchers, the system uses a large language model to help write, test, and refine computer programs. It explores different approaches and evaluates their performance to develop stronger predictions.
This represents an interesting evolution in AI technology. Instead of simply answering questions or summarizing existing information, AI can increasingly participate in the scientific discovery process.
Research describing ERA has been published in the scientific journal Nature, and Google has begun making the underlying technology available to selected researchers through its experimental science tools.
Why Accurate Flu Forecasting Matters
Predicting influenza outbreaks is more important than many people realize.
During the 2025–2026 flu season, weekly influenza-related hospital admissions in the United States exceeded 40,000 at their peak. Hospitalizations began increasing in November 2025 and reached their national peak during the week ending December 27.
When hospitals experience a sudden increase in patients, they may face shortages of available beds, healthcare workers, and medical supplies.
More accurate forecasts can help healthcare systems prepare in advance.
For example, if a forecasting model predicts that flu hospitalizations will increase significantly in a particular state over the next two weeks, healthcare administrators can consider adjusting staffing schedules, preparing additional beds, and ensuring adequate supplies of antiviral medications.
Public health agencies can also use these forecasts to help determine when additional public health messaging or vaccination campaigns may be useful.
In other words, better predictions have the potential to improve how healthcare resources are distributed and how communities prepare for seasonal disease outbreaks.
AI Is Powerful, but It Isn’t Perfect
Although Google’s model achieved the highest overall ranking among individual submissions, the CDC’s evaluation also revealed that forecasting infectious diseases remains challenging.
During the late-December surge in hospitalizations, many forecasting models struggled to anticipate how quickly admissions would rise. They also had difficulty predicting the subsequent decline in January.
The CDC reported that around the peak of the flu season, fewer than 25% of the two-week forecast prediction intervals across jurisdictions captured the actual observed values.
This highlights an important limitation: even sophisticated AI systems cannot perfectly predict the future.
Disease transmission is influenced by numerous factors, including human behavior, population immunity, vaccination rates, viral changes, and environmental conditions.
AI models can identify patterns and improve predictions, but uncertainty will always remain.
It is also important to recognize that Google’s first-place ranking does not mean AI has replaced traditional forecasting methods. The CDC’s own FluSight ensemble, which combines predictions from multiple models, ranked seventh overall and remains an important part of the agency’s forecasting efforts.
Beyond Influenza: AI’s Potential in Medical Research
The implications of Google’s achievement extend beyond seasonal influenza.
Researchers involved in the project also developed forecasting systems for COVID-19 and respiratory syncytial virus (RSV), two other respiratory illnesses that can place significant pressure on healthcare systems.
According to the researchers, their automated system was able to generate useful forecasts across multiple diseases, including situations where relatively little historical data was available.
This capability could become particularly valuable when scientists encounter emerging infectious diseases for which established forecasting models do not yet exist.
If AI can help researchers develop useful forecasting systems more quickly, public health agencies may eventually be able to respond more effectively to emerging threats.
However, additional testing and real-world validation will be necessary before such systems can be relied upon in new settings.
A Positive Example of Artificial Intelligence
Artificial intelligence has become one of the most debated technologies of our time.
Its rapid adoption has raised concerns about job displacement, the future of creative industries, misinformation, and the growing influence of technology companies.
These concerns deserve serious consideration, particularly as AI systems become more capable and increasingly integrated into everyday life.
However, applications such as disease forecasting demonstrate another side of the technology.
When used responsibly, AI can help scientists analyze complicated information, test new ideas, and potentially solve problems that would otherwise require substantial time and resources.
Google’s success in the CDC’s FluSight evaluation is a promising example of how artificial intelligence can complement human expertise rather than simply replace it.
The achievement does not mean AI will eliminate influenza outbreaks or prevent every hospitalization. But it suggests that AI-assisted scientific research could become an increasingly valuable tool for protecting public health.
Ultimately, the significance of this achievement goes beyond which company developed the best forecasting model. It demonstrates how artificial intelligence, combined with human scientific expertise, may help society better anticipate and prepare for real-world challenges.
Peter Cheng
Editor-In-Chief
YouMobile, Inc
