Claude and ChatGPT for healthcare teams: admin work, privacy and what not to use them for
Claude and ChatGPT can speed up healthcare paperwork, but neither replaces a clinician, and patient data needs the right plan and agreement.
Use Claude or ChatGPT for healthcare admin work such as prior authorization checks, patient message drafts and policy summaries. Do not use either to decide a diagnosis or treatment. Claude's own healthcare page says answers can be traced to their sources so you can verify before you act.
Keep patient information out of free and personal plans. Anthropic's pricing page lists a HIPAA-ready offering under Claude Enterprise, so ask your compliance team and the vendor what agreement covers your use.
Good for
- Practice managers and administrators drafting policies and staff notices
- Billing and authorization staff reviewing documentation
- Front-desk and care coordination teams sorting patient messages
- Compliance and IT leads deciding what can touch patient data
ChatGPT
ChatGPT works for policy summaries and message drafts built from your own approved text. It is not a clinical decision tool.
Plan: A business plan for any staff use. Ask OpenAI for its current HIPAA and business associate terms for your plan before you share patient data. ChatGPT Business is $20 per user per month billed annually or $25 monthly.
What it can do
- Summarize a policy PDF you upload into a staff checklist
- Draft patient-facing instructions from text you provide and your clinic approves
- Keep a clinic's templates and instructions together in a Project
Limits
- On Free, Plus and Pro your chats can be used for training unless you turn off Improve the model for everyone in Settings > Data controls
- Business and Enterprise do not train on your business data by default
- Do not paste patient identifiers into any plan until your compliance team has approved it
Claude
Anthropic publishes a healthcare page with named connectors for coverage, coding and provider lookups. Our advice: for anything near patient data, use Enterprise and confirm the terms with Anthropic first.
Plan: Claude Enterprise if patient data is involved: Anthropic's pricing page lists a HIPAA-ready offering there. Self-serve Enterprise is $20 per seat per month billed annually plus usage, 20 seat minimum, as of November 2026.
What it can do
- Review a prior authorization request: validate codes, check coverage rules and flag missing documentation
- Help build claims appeals from the records you give it
- Sort patient messages by urgency, draft replies to routine questions and route the rest to staff
- Connect to the CMS Coverage Database, ICD-10, the NPI Registry and PubMed
- Connect to FHIR-compliant systems with an Anthropic sample skill, to read and work with your data
Limits
- Anthropic says Claude is built on HIPAA-ready infrastructure; that is not a substitute for your own compliance review
- Ambient scribing and chart review are products built on the Claude Developer Platform, not features of the Claude app
- Anthropic's pricing page lists a HIPAA-ready offering under Enterprise. Ask Anthropic what it covers before you share patient data
- A clinician must make clinical decisions; use Claude's sources to verify, not to decide
Plugs to turn on
- Medplum MCP: Search patient records and appointments stored in a Medplum FHIR server.
- BambooHR MCP Server: Check who is out and file time off for clinic staff.
Plugs worth turning on
Only two plugs here are health-specific: Medplum for FHIR records and AWS HealthOmics for genomics research. The rest are general tools for staff and documents.
- 1
Medplum MCPOfficialHostedOAuthMedplum's hosted server for its open-source FHIR platform: search patients, read lab results and create appointments. It can change records, so limit who connects and what it can do.
HostedRuns - 2
Microsoft Work IQOfficialHosted or LocalOAuthMicrosoft's official way to reach Microsoft 365 mail, calendar, files and Teams, after your IT admin approves it. For scheduling and policy documents, not patient charts.
BothRuns - 3BambooHR MCP ServerOfficialHostedOAuth
BambooHR's official server: see who is out, file or approve time off, and run HR reports. In beta, and an admin has to turn it on.
HostedRuns - 4pdfOfficialSkill
Anthropic's skill for reading PDFs, filling forms, and splitting or merging files, including OCR on scanned pages.
SkillRuns - 5docxOfficialSkill
Anthropic's skill for Word documents with tracked changes, for policies and patient handouts that a clinician reviews.
SkillRuns - 6
google-calendar-mcpLocalOAuthCommunity server for Google Calendar: read, create and change events. Runs on your computer and needs your own Google Cloud setup.
LocalRuns - 7
aws-healthomics-mcp-serverOfficialLocalAWS Labs server for genomics workflows in AWS HealthOmics, for research teams. It can start runs that cost money and has no read-only mode.
LocalRuns
Review a prior authorization request with Claude
Claude
- 1
Use an approved plan and remove identifiers
Work on a plan and agreement your compliance team has approved, or strip names, dates of birth and member IDs first.
- 2
Turn on the healthcare connectors
In Claude, go to Customize > Connectors and add the CMS Coverage Database, ICD-10 and NPI Registry connectors that Anthropic lists.
- 3
Ask for a documentation check
Asking for evidence per criterion, and no decision, keeps the final call with your reviewer.
PromptReview this prior authorization request for a Medicare patient needing a robotic-assisted lung biopsy. Check that the provider NPI, CPT and ICD-10 codes are valid, find the coverage policy that applies, and list every criterion with the page of the document that shows it or a note that it is missing. Do not recommend approval or denial. - 4
A reviewer decides
A qualified person checks each cited page and makes the decision. Claude's answers can be traced to sources so you can verify them.
Draft replies to routine patient messages for staff review
- 1
Write your approved answers first
Put your clinic's own instructions in the prompt, such as fasting rules for a blood test. The assistant should rewrite what you approved, not invent advice.
- 2
Ask for a draft with flags
Leave out the patient's name and record details unless your plan and agreement allow them.
PromptHere are our clinic's instructions for preparing for a fasting blood test. Draft a reply to a patient asking how to prepare. Use plain language at a 6th-grade reading level, use only our instructions, and flag anything a nurse must confirm. - 3
Route urgent messages to a person
Anthropic describes Claude identifying urgent issues in patient messages and routing them. Have staff review every urgent flag and every draft.
- 4
Send from your own system
A staff member reads and sends the final message from your patient portal or email. Do not let an assistant send to patients.
Turn a long policy PDF into a staff checklist
- 1
Upload the policy
Attach the PDF. Claude reads PDFs up to 1,000 pages; ChatGPT reads PDFs up to 512 MB per file.
- 2
Ask for a checklist with page numbers
Page numbers let a reviewer check each line.
PromptTurn this infection-control policy into a one-page checklist for front-desk staff. For each item, give the page number it comes from. Do not add anything that is not in the policy. - 3
Have the policy owner review it
The person who owns the policy compares the checklist to the original before it is used.
- 4
Export a Word file
In Claude, the docx skill can produce a Word file with tracked changes for your reviewers.
Uses docx
Verdict
Our advice: for admin work near patient data, start with Claude Enterprise, where Anthropic's pricing page lists a HIPAA-ready offering, and ask OpenAI for its current ChatGPT terms in writing. Use either on de-identified text. Never use them to decide care.
Both assistants save time on paperwork. The limit is clinical and legal: keep a qualified reviewer in the loop and confirm the privacy agreement before patient data goes in.
Best for
- Prior authorization documentation checks
- Draft replies built from approved clinic text
- Policy summaries and staff checklists
Skip it if
- You would use the answer for a diagnosis or treatment decision
- You would paste identifiable patient data on a plan without an agreement
Questions
Can ChatGPT or Claude give medical advice?
No. Neither replaces a clinician, and neither should make diagnosis or treatment decisions. For your own health, see a clinician and use the assistant only for background, checking any sources it links.
Is Claude HIPAA compliant?
Anthropic says Claude is built on HIPAA-ready infrastructure, and its pricing page lists a HIPAA-ready offering under Enterprise. Ask Anthropic what agreement applies before you share patient data.
Is ChatGPT HIPAA compliant?
Ask OpenAI for its current HIPAA terms for your exact plan. Do not paste patient data into Free, Plus or Pro, where chats can be used for training unless you opt out.
Can I paste patient information into Claude or ChatGPT?
Not unless your compliance team has approved the plan and the agreement. Otherwise remove names, dates of birth, member IDs and anything else that identifies a person.
What can Claude do for healthcare administrators?
Anthropic lists prior authorization review, claims appeals, and patient message triage with drafted replies. It connects to the CMS Coverage Database, ICD-10, the NPI Registry and PubMed.
Can Claude connect to our EHR?
Anthropic offers a sample skill that connects Claude to FHIR-compliant systems. The Medplum server also connects to a Medplum FHIR platform. Have IT review access before you try either.
Can Claude write clinical notes like an AI scribe?
Not as a built-in feature. Anthropic says health startups build scribing and chart review products on its Developer Platform. For clinical notes, use a product your organization has approved.
How much does Claude Enterprise cost?
Self-serve Enterprise is $20 per seat per month billed annually with a 20 seat minimum, plus usage at API rates, as of November 2026. Sales-assisted plans are quoted by Anthropic.
Does ChatGPT train on our data?
Not on Business or Enterprise by default. Free, Plus and Pro chats can be used unless turned off in Settings > Data controls.
Can doctors use ChatGPT to look things up?
Our advice: for background reading, with every source checked. Do not rely on a chatbot answer for a clinical decision, and do not enter patient identifiers on a plan your organization has not approved.
Sources
Docs, issue trackers and discussions this page is built from.
Show all 8Hide
- Claude for healthcare · claude.com
- Claude plans and pricing · claude.com
- Data controls FAQ · help.openai.com
- ChatGPT pricing · chatgpt.com
- Uploading files and audio to ChatGPT · help.openai.com
- Projects in ChatGPT · help.openai.com
- Use connectors · support.claude.com
- Uploading files to Claude · support.claude.com