Industry Guide: Healthcare
How to Use Claude in Healthcare: A Guide for Clinical and Operations Teams
Less time on paperwork, more time with patients.
This guide is for clinicians, health system operations staff, revenue cycle teams, and payer staff who want to put Anthropic's Claude to work on clinical documentation, prior authorization, coding support, patient communication, and administrative work. It covers what Anthropic announced for healthcare in January 2026 and what named providers and health technology companies said on Anthropic's own pages.
Learning how to use Claude in healthcare starts with administrative and documentation work around care: drafting and structuring clinical notes, assembling prior authorization requests against payer criteria, suggesting ICD-10 and CPT codes for coder review, abstracting data from charts, and writing plain-language patient letters. Anthropic describes HIPAA-ready offerings for providers and payers. Clinicians remain responsible for diagnosis, treatment, and every note they sign, and protected health information stays inside approved, compliant deployments.
Published
Key Takeaways
- The strongest healthcare uses are clinical documentation, prior authorization, coding support, chart abstraction, and patient communication.
- In January 2026 Anthropic announced HIPAA-ready offerings for providers and payers, with connectors to the CMS Coverage Database, ICD-10, the NPI Registry, and PubMed.
- Anthropic's pages describe named deployments at Banner Health, Carta Healthcare, Commure, Premier, and Qualified Health with UTMB.
- Clinical judgment and responsibility for signed documentation stay with licensed clinicians.
- Protected health information belongs only in deployments your organization has approved under HIPAA, including a business associate agreement where required.
Copy and Paste
Prompts You Can Use Today
Structure a clinical note
You are assisting a primary care physician. Turn the visit transcript and my shorthand notes below into a SOAP note. Use only information stated in the transcript or my notes, put anything uncertain in brackets for me to confirm, keep medications and doses exactly as stated, and do not add diagnoses I did not mention. Keep the assessment and plan as a numbered problem list. Transcript and notes: [paste].
Bracketed uncertainties make the physician's review fast and keep the signed note accurate.
Prior authorization packet
Assemble a prior authorization request for the procedure below. Compare the chart excerpts with the payer's medical necessity criteria pasted here, and for each criterion quote the chart evidence that meets it or state that evidence is missing. Then draft the letter of medical necessity using only that evidence. Do not state any clinical fact not in the chart. Chart excerpts and criteria: [paste].
The criterion-by-criterion table shows the clinician exactly what is missing before submission.
Coding suggestions for review
Read the clinical note below and suggest ICD-10-CM diagnosis codes and CPT procedure codes for a certified coder to review. For each code, quote the documentation that supports it, note any documentation gap that prevents a more specific code, and list questions for the provider. Do not suggest codes without documented support. Note: [paste].
Suggestions with quoted support help coders; final code assignment stays with them.
Plain-language patient letter
Rewrite the discharge instructions below as a letter a patient with no medical background can follow, at about a sixth-grade reading level. Keep every medication, dose, time, warning sign, and follow-up appointment exactly as written. Use short sentences and a checklist for daily tasks. Add a line telling the patient who to call with questions. Do not add any new advice. Instructions: [paste].
Keeping every medication and warning exact is the non-negotiable part of this prompt.
Chart abstraction for a registry
Abstract the data elements listed below from the chart excerpts for a quality registry submission. For each element give the value, the source document and date, and a confidence note if the chart is ambiguous. Leave the value blank rather than guess when the chart does not state it. Elements and chart excerpts: [paste].
Blank-rather-than-guess is what keeps abstraction accurate enough for registries.
From Anthropic
What Anthropic offers for Healthcare
Claude for Healthcare
Announced in January 2026: HIPAA-ready offerings for healthcare providers and payers, connectors to the CMS Coverage Database, ICD-10, the National Provider Identifier Registry, and PubMed, Agent Skills for FHIR development and a sample for prior authorization review, and beta consumer health integrations for US Pro and Max users.
· Source
Anthropic and PwC expanded partnership
A May 2026 expansion of Anthropic's partnership with PwC for regulated industries, including healthcare, in which Anthropic reports that Advocate Health is building toward full-scale deployment across its workforce.
· Source
In Practice
How Healthcare Teams Use Claude
According to Anthropic's published customer story
Banner Health
According to Anthropic's customer story, Banner Health built an internal AI platform on Claude through Amazon Bedrock, available to tens of thousands of employees, with most users reporting time savings and improved accuracy.
According to Anthropic's published customer story
Carta Healthcare
Anthropic's customer story on Carta Healthcare describes clinical data abstraction with Claude through Amazon Bedrock, cutting both routine and complex case times while keeping high inter-rater reliability.
According to Anthropic's published customer story
Commure
On Anthropic's healthcare page, Commure's CTO describes using Claude to automate clinical documentation at scale.
According to Anthropic's published customer story
Premier
Premier, which serves thousands of healthcare organizations, is quoted on Anthropic's healthcare page on Claude handling complex healthcare workflows.
According to Anthropic's published customer story
Qualified Health
Anthropic's customer story describes Qualified Health using Claude to screen a large patient population at the University of Texas Medical Branch, surfacing medication optimization opportunities for heart failure patients.
Illustrative walkthrough
A prior authorization request, evidence first
- A clinic's authorization coordinator loads the chart excerpts and the payer's criteria into the approved workspace.
- Claude builds a criterion-by-criterion table quoting chart evidence and marks one criterion as unsupported.
- The coordinator asks the ordering clinician, who confirms a prior therapy trial documented in an outside record.
- Claude drafts the letter of medical necessity using only the confirmed evidence.
- The clinician reviews and signs the letter, and the coordinator submits the request.
Healthcare and Claude: the Published Numbers
| Figure | Context | Source |
|---|---|---|
| 85% of users report time savings | Banner Health's internal AI platform built on Claude, per Anthropic's customer story | Source |
| 55,000+ employees | Banner Health staff with access to its Claude-based platform, per Anthropic's customer story | Source |
| complex cases from 5-6 hours to 90 minutes | Carta Healthcare clinical data abstraction, per Anthropic's customer story | Source |
| 98-99% inter-rater reliability | Carta Healthcare abstraction quality, per Anthropic's customer story | Source |
Core Workflows
AI in Healthcare Use Cases That Return Time to Care
These are the Claude healthcare use cases where AI for healthcare teams saves the most time without touching clinical judgment.
Clinical documentation
AI for clinical documentation structures transcripts and shorthand into notes, including ambient clinical documentation workflows, for the clinician to review and sign.
Prior authorization
AI for prior authorization compares chart evidence with payer criteria, lists gaps, and drafts letters of medical necessity.
Coding support
AI for medical coding suggests ICD-10 and CPT codes with quoted documentation, leaving final assignment to certified coders.
Chart abstraction
Registry and quality measure abstraction from charts, with sources and blanks where the record is silent.
Patient intake and communication
AI for patient intake summarizes forms and history for the care team, and turns instructions into plain-language letters.
Health IT and FHIR
Developers use Claude to build against FHIR APIs, and Anthropic has published an Agent Skill for FHIR development.
What Anthropic Released
What Anthropic Announced for Healthcare
In January 2026 Anthropic announced HIPAA-ready products for healthcare providers and payers. According to the announcement, organizations operating under HIPAA can use Anthropic's enterprise plan and access healthcare connectors including the CMS Coverage Database, ICD-10, the National Provider Identifier Registry, and PubMed.
The same announcement added Agent Skills for FHIR development and a sample skill for prior authorization review. For consumers in the United States on Pro and Max plans, Anthropic announced beta integrations with HealthEx, Function, Apple Health on iOS, and Android Health Connect, and said health data from these integrations is not used for training.
For hospitals, this means the workflows on this page can run with connectors to coverage rules and code sets rather than copied text, where your organization has enabled them. Without connectors, the same prompts work on text your policies allow you to share with an approved deployment.
What Providers Report
How Hospitals Use AI Today
How do hospitals use AI? The named examples on Anthropic's pages center on documentation and data. Banner Health's customer story describes an internal AI platform built on Claude through Amazon Bedrock and available to tens of thousands of employees, with most users reporting time savings. Carta Healthcare's story describes clinical data abstraction where complex cases dropped from hours to about an hour and a half.
Commure's CTO describes using Claude to automate clinical documentation at scale. Premier, which serves thousands of healthcare organizations, says Claude handles complex healthcare workflows accurately and securely. Qualified Health describes screening a large patient population at the University of Texas Medical Branch to find medication optimization opportunities.
The pattern is consistent: AI prepares, structures, and screens, and clinicians and staff decide. That division is what makes these deployments defensible to patients, regulators, and the clinicians who use them.
Privacy and Choice
HIPAA Compliant AI and Claude vs ChatGPT for Healthcare
Is AI in healthcare HIPAA compliant? A tool is not compliant by itself. Compliance depends on the deployment, the contract, including a business associate agreement where required, and how your organization configures and uses it. HIPAA compliant AI use means protected health information goes only into deployments your privacy and security teams have approved, with access controls, audit logging, and minimum necessary data.
The Claude vs ChatGPT for healthcare question should be answered on de-identified or synthetic versions of your own work: a note, a prior authorization case, and a patient letter. Compare accuracy, how each handles uncertainty, and whether each invents details. Then weigh the practical factors: HIPAA-ready deployment options, integration with your EHR and connectors, and support.
Clinicians already carry the legal and professional responsibility for what they sign. Any AI tool has to fit that reality, which is why review steps matter more than which model drafted the text.
Getting Started
A Safe First Month for a Health System Team
Confirm the approved deployment
Work with privacy and security to confirm which deployment may hold PHI and under what agreement.
Start with synthetic cases
Test note, prior authorization, and letter prompts on synthetic or de-identified cases first.
Choose one workflow
Pick one high-volume task, such as prior authorization letters, and measure time and error rates.
Design the review step
Decide who reviews each AI-assisted output and how that review is recorded.
Limits
Where Clinicians Must Stay in Charge
Claude can omit a symptom, carry a dose incorrectly, or phrase an uncertain finding as certain. Every AI-assisted note must be reviewed before signing, and medication details deserve particular care. Do doctors trust AI? Trust is earned case by case, and designs that make errors easy to spot, such as bracketed uncertainties and quoted evidence, earn it faster.
Diagnosis, treatment decisions, and medical necessity judgments belong to licensed clinicians. Claude can organize evidence for those decisions and draft the paperwork afterward.
Patients may also use consumer AI tools to understand their records. Clinicians can help by encouraging patients to bring those summaries and questions to appointments, where they can be checked and put in context.
Learn Safe, Practical AI Habits
Reliable AI use in healthcare comes from precise instructions and disciplined review. The free Nightschool AI curriculum is hands-on from the first lesson.
Frequently Asked Questions
How do hospitals use AI?
Mostly for documentation, prior authorization, coding support, chart abstraction, patient communication, and internal knowledge tools. Anthropic's pages describe Banner Health's internal AI platform, Carta Healthcare's clinical data abstraction, and Commure's clinical documentation work. Clinical decisions remain with clinicians.
Is AI in healthcare HIPAA compliant?
It depends on the deployment and the agreements around it, not the tool alone. Anthropic describes HIPAA-ready offerings for providers and payers. Your organization must confirm the approved deployment, any business associate agreement, and controls before protected health information is used.
Can AI write clinical notes and SOAP notes?
AI can draft and structure notes from transcripts and shorthand, including SOAP format, for the clinician to review and sign. It should only use information from the encounter, mark uncertainties, and never add diagnoses. The signed note is the clinician's responsibility.
How does AI help with prior authorization?
It compares chart evidence with payer criteria, shows what is met or missing, and drafts the letter of medical necessity. Anthropic has published a sample Agent Skill for prior authorization review and a CMS Coverage Database connector. Clinicians confirm medical necessity before submission.
What has Anthropic released for healthcare?
In January 2026 Anthropic announced HIPAA-ready offerings for providers and payers, connectors to the CMS Coverage Database, ICD-10, the NPI Registry, and PubMed, Agent Skills for FHIR development and prior authorization review, and beta consumer health integrations in the United States. Details are linked in the offerings section.
Do doctors trust AI?
Trust varies and is earned through experience with specific tools. Designs that make errors easy to catch, such as quoted evidence and marked uncertainties, help clinicians decide when to rely on output. Organizations should measure accuracy on their own cases before scaling.
How much time does AI save clinicians on documentation?
It depends on the specialty, workflow, and tool, so measure it locally. Anthropic's pages report that most Banner Health users see time savings and that Commure's documentation automation saves clinicians significant time. Review time should be counted too.
Can AI summarize my medical history?
In the United States, Anthropic announced beta integrations that let Pro and Max users connect health records and apps such as HealthEx, Function, Apple Health, and Android Health Connect, and said that health data is not used for training. Bring any summary to your clinician to check.
Keep going
Related Guides and Courses
Claude in Healthcare, by role
Give Time Back to Patient Care
Hands-on lessons for clinical and operations teams.