Google Cloud Opens Gemini for Healthcare Preview in U.S.

Google Cloud opened a U.S. preview of Gemini for healthcare on November 10, 2026, aiming agentic AI squarely at the roughly 28 hours of administrative work clinicians say they log every week. The system extends Google's universal Gemini agent with healthcare-specific skills and data connectors, and it ships with named launch customers including HCA Healthcare, Optum and Stanford School of Medicine.

The workload math comes from a 2024 Harris Poll survey Google commissioned. Conducted online between August 26 and September 9, 2024, among 821 healthcare providers, 209 payors and 2,079 U.S. adults, it found medical office staff reporting 34 hours of administrative work weekly and claims staff 36 hours. That's context for the problem, not a performance claim for the product — worth keeping straight when vendors cite it.

What does the product actually do?

Gemini for healthcare is not a standalone app. It layers healthcare "skills" — packaged instructions and context from domain experts — onto the Gemini agent Google Cloud introduced on October 8, 2026, when healthcare was still flagged as "coming soon." Organizations can build custom versions around their own policies and standard operating procedures. During preview, it's available to Gemini Enterprise Standard and Gemini Enterprise Plus customers.

For providers, the system merges EHR data with outside records into a cited patient history across care settings. It separates active problems from resolved ones, lists allergies and flags medication discrepancies. Before a scheduled visit or procedure, a separate workflow checks for intervening events, current medications, and outstanding clearances, lab work or consents.

On the payer side, Gemini checks authorization requests against medical policy and organizes traceable findings for clinical reviewers. Member-service workflows estimate out-of-pocket costs, track deductibles and verify network status using plan documents and eligibility data.

What changes for the EHR?

The short answer: nothing gets written back. The initial Epic and Oracle Health integrations are read-only. Epic connects through open.epic APIs; Oracle Health uses Ignite APIs for FHIR, a clinical-data interoperability standard. Chart entries, orders and final sign-off remain clinician-led, per the product documentation. Google is positioning this as assembly work around care and coverage, not autonomous decision-making.

The connector catalog stretches beyond the two EHRs:

• Keragon connectivity to more than 400 healthcare applications
• Current CMS coverage determinations
• The National Provider Identifier registry
• The RxNorm drug vocabulary and ICD-10-CM diagnosis codes

Availability is uneven. Optum Network Eligibility and Coordination of Benefits sit in limited release and testing, and NextGen Healthcare connectivity remains in development with release expected "in the coming months."

Who's already on board?

Google names Blue Cross Blue Shield of Massachusetts, Hartford HealthCare, HCA Healthcare, Highmark Health, Northwell Health, Optum and Stanford School of Medicine as launch customers, with the American Society of Clinical Oncology listed as a contributor. HCA Healthcare said it is exploring how Gemini could synthesize patient information across multiple EHR systems and care settings.

Stanford is pointing the tool at research rather than the chart. "Our clinical data is only as useful as researchers' ability to ask questions of it," said Todd Ferris, deputy chief information officer at Stanford School of Medicine. He described researchers querying an Observational Medical Outcomes Partnership database in plain language — no SQL or data-model expertise required up front.

Governance, model routing and pricing

Google says the platform is built to support HIPAA compliance. Patient records, clinical data, prompts and outputs stay inside the customer's governed Google Cloud environment and are not used to train or fine-tune Google's foundation models. Connectors inherit existing identity and role-based permissions, and citations plus audit logging let reviewers inspect outputs and activity. Customer-managed encryption keys and VPC Service Controls round out the security stack.

One governance framework covers models from Google, Anthropic and OpenAI. Google's routing sends high-volume administrative tasks to faster, cheaper models and reserves higher-reasoning models for complex analysis. Pricing runs through per-user seat subscriptions or pay-as-you-go, and connectors support global and U.S. locations during preview.

The practical test for buyers is whether these cited, permission-aware workflows save hours without weakening oversight — especially while parts of the connector catalog, like NextGen, are still unfinished.