Gemini for healthcare targets prior authorization, the paperwork where patients have the most at stake

Google Cloud launched Gemini for healthcare in preview. It extends the universal Gemini agent Google introduced at its Gemini at Work event (our coverage) and connects to electronic health records, billing platforms, clinical guidelines and Medicare data. Connectors reach CMS coverage data, the NPI Registry, RxNorm and ICD-10, plus partner tools for eligibility, benefits verification and prescription pricing. Google says it carries out multi-step workflows like records synthesis and prior authorization review within defined guardrails and with human oversight.

Google cites nearly 28 hours a week that clinicians lose to administrative work. Launch partners include HCA Healthcare, Highmark Health and Northwell Health, and Google’s press release adds Blue Cross Blue Shield of Massachusetts, Hartford HealthCare, Optum and Stanford School of Medicine.

Google is aiming at administrative sludge rather than diagnosis or chatbots playing doctor: building a patient timeline from scattered records, preparing charts before visits, assembling authorization packets and flagging documentation gaps before a claim is denied. That is useful AI aimed at a real, measurable problem that burns out clinicians and delays care.

Prior authorization is where an insurer decides whether a patient gets a treatment. Google is selling to both sides: providers assembling requests and health plans checking them against medical policy. If both ends run agents, the process could get faster and cleaner, or become an automated arms race where denials scale as easily as approvals.

Human oversight is the safeguard everyone cites. Highmark told Fortune that a clinician reviews every case before it issues a denial. That should be standard, auditable and resistant to rubber-stamping as volumes grow. A human in the loop only counts if that person has time to disagree.

An agent reasoning across EHRs, billing and payer systems concentrates extremely sensitive information in one workflow. Google says apps, skills and connectors sit under a single governance framework. The proof will come in audits, and in how errors are traced when a summary drops an allergy or misses a medication conflict. Google’s Agent Compute Engine, which isolates each agent in its own microVM on the assumption it will eventually be compromised, is relevant groundwork (our analysis).

The product is promising, appropriately scoped and still in preview. The evidence that matters will be denial rates, error rates and clinician time saved, published as outcomes rather than testimonials. More in our healthcare coverage and Google coverage.