Clinicians spend nearly 28 hours a week reconciling information by hand, according to Google Cloud. That figure, whatever its precise source, points at the real problem in healthcare AI: not a shortage of clever models, but a shortage of connected data. Google’s answer, announced November 10 by Google Cloud CEO Thomas Kurian, is Gemini for healthcare, a purpose-built version of its Gemini agent aimed at providers and payers.

It is available today in preview. Google says it is built to support HIPAA compliance.

What it is and how it works

Gemini for healthcare extends the universal Gemini agent introduced last month at Gemini at Work. An employee hands the agent an objective. The agent plans the steps, keeps organizational context, and routes work across different models to produce a finished outcome inside the tools a team already uses.

Governance is built into the runtime. Each agent has its own identity, inherits the user’s existing role-based permissions, and enforces need-to-know boundaries across users, sub-agents and workflows. Humans stay in the loop. That matters in a field where an unsupported answer is a clinical risk, not an inconvenience.

Four components

Google developed the product with clinical and health plan leaders. It has four parts:

• Skills: reusable instructions written by domain experts for tasks such as medical records synthesis, pre-appointment preparation, coverage and policy review, and payer benefit determination. Organizations can write their own to match local procedures.
• Connectors: links to EHR, claims, care management and administrative systems. Gemini queries each one while preserving existing access controls, so data does not have to be migrated.
• Partner ecosystem: systems integrators and digital health partners extend the product and connect it to existing infrastructure, with Google promising no lock-in.
• Governed platform with model choice: one place to set policy, with audit logging and verifiable citations so reviewers can check the source of each answer. Organizations can pair frontier models for complex clinical reasoning with cheaper, faster ones for routine administrative steps.

Workflows for providers

At the point of care, the first skill builds a longitudinal patient timeline from EHR and external records. It separates active problems from resolved ones, lists allergies and flags medication discrepancies. A second prepares the chart before a visit or procedure, noting interval events, current medications and outstanding labs, clearances or consents.

The third targets a costly pain point: prior authorization. Gemini pulls chart evidence into an audit-ready packet aligned with payer policy and flags documentation gaps before submission, not after a denial. If a denial arrives and the record supports it, the agent drafts an appeal grounded in the clinical record and the plan’s policy.

Workflows for payers

On the coverage side, Gemini checks incoming requests against a plan’s medical policies and organizes findings for the clinical reviewer, with each finding traceable to the submitted documentation. The idea is that reviewers spend less time assembling files and more time applying judgment.

It also parses benefit booklets to estimate out-of-pocket costs, track deductibles and verify network status, giving member services staff a step-by-step breakdown. For Medicare Advantage members, it applies Medicare’s coverage criteria where a plan’s internal rule is stricter.

Connectors and integrations

The connector library covers major clinical systems. The Epic and Oracle Health connectors give authorized users read-only access to conditions, medications, labs and clinical notes, with no write-back to the chart. That design choice protects the system of record, though it also means Gemini assembles and drafts but does not update records directly.

Other listed integrations include InterSystems’ FHIR MCP Server for standardized FHIR data access, Verato for patient identity matching across sources, and Keragon, which connects to more than 400 healthcare applications through a single MCP integration. On the eligibility and claims side, Optum Coordination of Benefits is among the connectors, used to find additional patient coverage. The source material was cut off before the rest of that list, and Google has not detailed pricing in the announcement.

Why it matters

Most health AI so far has been a point tool: an ambient scribe here, a coding assistant there. Google is betting that the bigger gain sits in the connective work between systems, where staff retype, search and reconcile all day. That is also where denials and delays start.

The caveats are familiar. This is a preview, and Google has published no outcome data on denial rates or time saved. Health systems will want to test accuracy, citation quality and audit trails against their own records before relying on it. But the design, with read-only access, inherited permissions and human review, is the kind buyers in a regulated field have been asking for.