Google AI Tops CDC Flu Hospital Forecast Ranking
A flu forecasting model built with Google AI was the best at forecasting flu-related hospital admissions during the 2025-26 season, according to the Centers for Disease Control. The CDC's end-of-season analysis, published this week, shows that of 39 eligible models, Google's best matched the season's observed hospital admissions.
Transcript
Google's model ranked best for forecasting flu hospital admissions. The CDC announced it this week.
The CDC compared thirty nine eligible models. Google's best matched last season's observed admissions.
Every October through May, FluSight collects weekly hospital admission forecasts three weeks ahead. The CDC uses the combined forecast to signal expected demand in each state.
Google built the forecasts with Empirical Research Assistance, a tool generating optimization algorithms. Research on it was published in Nature, and its technology is available only to trusted testers.
Google publishes no accuracy figure and no margin over rival models. The evaluation is the CDC's, but the framing is Google's.
Inboxsmith builds AI receptionists for small business. Please like and subscribe for more news.
Sources
Every claim in this video comes from the top ranking coverage of this topic. The claims and where each one came from:
- "A flu forecasting model built with Google AI was the best at forecasting flu-related hospital admissions during the 2025-26 season, the Centers for Disease Control announced this week." Backs hook sentence one, that Google's model ranked best for forecasting flu hospital admissions. Note on corroboration: YouTube transcript research returned zero usable transcripts this run, so this first party Google post is the sole source for every narrated claim in this script.(Google (blog.google), first party blog post titled "Google's science AI ranks #1 in CDC evaluation", dated September 30, 2026, byline Zahra Shamsi of Google Research, alert time snapshot captured 2026-09-30T21:41:52.577Z and live page refetched identical this run)
- "A flu forecasting model built with Google AI was the best at forecasting flu-related hospital admissions during the 2025-26 season, the Centers for Disease Control announced this week." Backs hook sentence two, that the CDC announced it this week. The article also says "The end-of-season analysis published this week shows", confirming the this week framing rather than a specific narrated date.(Google (blog.google), first party blog post titled "Google's science AI ranks #1 in CDC evaluation", dated September 30, 2026, byline Zahra Shamsi of Google Research, alert time snapshot captured 2026-09-30T21:41:52.577Z and live page refetched identical this run)
- "The end-of-season analysis published this week shows that of 39 eligible models, Google's best matched the season's observed hospital admissions." Backs scene 1 sentence one, that the CDC compared thirty nine eligible models.(Google (blog.google), first party blog post titled "Google's science AI ranks #1 in CDC evaluation", dated September 30, 2026, byline Zahra Shamsi of Google Research, alert time snapshot captured 2026-09-30T21:41:52.577Z and live page refetched identical this run)
- "The end-of-season analysis published this week shows that of 39 eligible models, Google's best matched the season's observed hospital admissions." Backs scene 1 sentence two, that Google's best matched last season's observed admissions. The season is identified elsewhere in the post as the 2025-26 season, which is why narration says last season rather than naming a year.(Google (blog.google), first party blog post titled "Google's science AI ranks #1 in CDC evaluation", dated September 30, 2026, byline Zahra Shamsi of Google Research, alert time snapshot captured 2026-09-30T21:41:52.577Z and live page refetched identical this run)
- "From October through May, the CDC's FluSight combines weekly submissions from government, industry, and academic teams predicting U.S. hospital admissions for the current week and three weeks into the future." Backs scene 2 sentence one, that every October through May FluSight collects weekly hospital admission forecasts three weeks ahead. The list of submitting sectors in that sentence was deliberately not enumerated in narration because spoken lists break the word to duration ratio.(Google (blog.google), first party blog post titled "Google's science AI ranks #1 in CDC evaluation", dated September 30, 2026, byline Zahra Shamsi of Google Research, alert time snapshot captured 2026-09-30T21:41:52.577Z and live page refetched identical this run)
- "The CDC uses the combined forecast each week to communicate the anticipated state-level demand for medical services." Backs scene 2 sentence two, that the CDC uses the combined forecast to signal expected demand in each state. The phrase in each state renders the article's state-level wording without a hyphenated compound.(Google (blog.google), first party blog post titled "Google's science AI ranks #1 in CDC evaluation", dated September 30, 2026, byline Zahra Shamsi of Google Research, alert time snapshot captured 2026-09-30T21:41:52.577Z and live page refetched identical this run)
- "Google forecasts were developed using Empirical Research Assistance, an AI tool that generates optimization algorithms across a range of scientific fields." Backs scene 3 sentence one, that Google built the forecasts with Empirical Research Assistance, a tool generating optimization algorithms. The ERA acronym is avoided in narration because text to speech reads it as the English word era.(Google (blog.google), first party blog post titled "Google's science AI ranks #1 in CDC evaluation", dated September 30, 2026, byline Zahra Shamsi of Google Research, alert time snapshot captured 2026-09-30T21:41:52.577Z and live page refetched identical this run)
- "Research on ERA was recently published in the journal 'Nature.'" and "ERA's underlying technology is now available to trusted testers as part of our experimental science tools." Together these back scene 3 sentence two, that research on it was published in Nature and its technology is available only to trusted testers. Nothing in the post describes a general release, which is why narration never says launched or released.(Google (blog.google), first party blog post titled "Google's science AI ranks #1 in CDC evaluation", dated September 30, 2026, byline Zahra Shamsi of Google Research, alert time snapshot captured 2026-09-30T21:41:52.577Z and live page refetched identical this run)
- Negative citation, verified against the full text of the post, which is four paragraphs long. The only performance detail given anywhere is "of 39 eligible models, Google's best matched the season's observed hospital admissions." There is no accuracy percentage, no error rate, no score, no margin over any other model, and no competitor is named. Backs scene 4 sentence one, that Google publishes no accuracy figure and no margin over rival models.(Google (blog.google), first party blog post titled "Google's science AI ranks #1 in CDC evaluation", dated September 30, 2026, byline Zahra Shamsi of Google Research, alert time snapshot captured 2026-09-30T21:41:52.577Z and live page refetched identical this run)
- The evaluation is attributed to the CDC, "the Centers for Disease Control announced this week" and "The end-of-season analysis published this week shows", but the post itself is first party Google publishing under a Google Research byline and frames the result in Google's own voice: "This performance validates our confidence that AI combined with human ingenuity will improve our ability to forecast diseases worldwide." Backs scene 4 sentence two, that the evaluation is the CDC's but the framing is Google's.(Google (blog.google), first party blog post titled "Google's science AI ranks #1 in CDC evaluation", dated September 30, 2026, byline Zahra Shamsi of Google Research, alert time snapshot captured 2026-09-30T21:41:52.577Z and live page refetched identical this run)
- Context not narrated, recorded for the hub page: "This performance validates our confidence that AI combined with human ingenuity will improve our ability to forecast diseases worldwide." This is a Google opinion statement about future disease forecasting, not a reported result, so it was cut from narration. Also cut for length: the detail that FluSight submissions come from government, industry, and academic teams, that forecasts cover the current week as well as three weeks into the future, that the admissions forecast is specifically for the United States, and that the tool's technology is offered "as part of our experimental science tools."(Google (blog.google), first party blog post titled "Google's science AI ranks #1 in CDC evaluation", dated September 30, 2026, byline Zahra Shamsi of Google Research, alert time snapshot captured 2026-09-30T21:41:52.577Z and live page refetched identical this run)