NewCall a live VoiFlow agent in your region. It picks up on the first ring.Call it now
VoiFlow
All resources

Regions 7 min read

HIPAA and AI receptionists: handling patient calls safely

Patient calls carry health information. Here is how to verify callers, limit what is said, and what to ask an AI receptionist vendor before go-live.

Regional team
VoiFlow

A bright family medicine clinic hallway where a nurse in scrubs walks past exam room doors carrying a clipboard

A patient call is rarely just a booking. The caller may name a clinic, a doctor and a time, and that alone reveals they are a patient. The reason for the visit can be more personal still. Before an AI receptionist answers those calls, plan how it will protect patient information. A HIPAA compliant AI receptionist is set up that way from the first call, and your clinic's policies and vendor contract decide whether it is.

No product is compliant on its own. An AI receptionist can verify callers, share only what they need, and limit who can open the records, but the clinic still owns the policies, the contract and the decisions about what the agent may say. This article covers the practical steps. Whether HIPAA applies to your clinic, and what it requires, is a question for your counsel.

What the HIPAA Privacy Rule says about vendors and minimum necessary

Two parts of the Privacy Rule matter most when an outside service answers patient calls. Under 45 CFR 164.502, a covered entity may share protected health information with a business associate only after getting satisfactory assurance that the vendor will safeguard it, and that assurance must be documented in a written contract or agreement. The same section sets the minimum necessary standard: reasonable efforts to limit the information used or disclosed to what the purpose needs.

For an AI receptionist, that means a signed agreement with the vendor before patient calls go live, and an agent set up to ask for, say and store no more than the call requires.

What a patient call contains

Calls to a clinic carry information many people treat as private: that someone is a patient at all, the name of the doctor they see, the time of a scan, and sometimes the reason for the visit.

For example, a caller asks whether "Mrs Hassan" is booked for a scan on Thursday. Even confirming that she is a patient reveals something. The agent should not answer until the caller has been verified, and it should not hint at the answer while it checks.

The healthcare pillar covers how clinics use an agent across the front desk, and AI receptionist for clinics looks at the booking side of the same calls.

Verify the caller before any details

Ask for two details the clinic already holds before the agent reads out any appointment information. Good options include the date of birth and the name on the booking, or a reference number sent by text to the number on file. A name alone is not enough, because anyone can give a name.

The verification use case sets out the same approach for other sectors, and it is a useful model for a clinic's own process. Give the caller a clear route if verification fails. The agent can send a secure link by text, offer a call back to the number on file, or pass the caller to reception. It should never guess, relax the check because the caller sounds certain, or read out details to someone who has not passed.

Share only what the caller needs

A useful rule is to share only what the caller needs for the task in hand. A caller who asks when their next appointment is needs a date and a time. They do not need the name of the doctor, the reason for the visit or notes from the last consultation, unless the task truly requires them.

The same rule applies to what the agent says on a voicemail or a text. A message that says "your scan at the imaging centre is on Thursday" reveals more than "please call the clinic about your appointment". Choose the wording with the least detail that still does the job.

Recordings, transcripts and retention

Decide first whether calls are recorded at all. If they are, tell callers at the start, set a retention period you can explain, and limit who can open the recordings and transcripts.

Ask the vendor whether the summary is stored separately from the recording, and how each one is deleted. A summary can hold more detail than the recording it came from, because it is written to be useful. Treat it with the same care as the call.

Emergencies and clinical questions

The agent should never give clinical advice. A caller who describes chest pain, heavy bleeding, difficulty breathing or a sudden loss of consciousness should be told to call emergency services at once, and the call should be routed to the right place without delay.

A question about a medicine, a test result or a symptom goes to a clinician, with a message to the team and a call-back time. The agent books, informs and passes on. It does not diagnose, reassure or advise.

Questions to ask the vendor

Ask every vendor the same questions, and keep the answers in writing:

  • Will you sign a written agreement covering patient information if we need one?
  • Who at your company can open recordings and transcripts, and why?
  • Where are recordings, transcripts, summaries and backups stored?
  • Which other companies process call data for you, and in which countries?
  • How do you delete a recording or summary when we ask you to?
  • What logs do you keep of who opened a record, and for how long?
  • How do you handle a suspected leak, and how quickly would we hear about it?

A vendor that answers "it depends" to several of these has not finished the homework. Ask for the answer for your plan, and for the set-up you will actually run. Our AI voice agent security checklist has a wider list of questions that apply to any vendor, not only to clinics.

Calls the agent should not take alone

Some calls belong with a clinician or a member of the administrative team from the start. A request for copies of medical records should go through the clinic's formal process, with identity checks and a written request. A complaint about treatment, a dispute about a bill that involves clinical detail, and a request to change a prescription all need a person.

When the agent cannot handle a call, it should say so plainly, take the minimum details needed to route it, and pass a summary to staff. It should never try to settle the matter itself, and it should not promise an outcome that only a clinician can give.

What to set up first

  1. Decide which call types the agent will handle, and which it will pass to staff at once.
  2. Write the verification steps for each call type, including what happens when they fail.
  3. List the information the agent may say and record for each call type.
  4. Agree the retention period and the list of people who can open records, with the vendor in writing.
  5. Test the set-up with staff playing patients, then with a small number of real calls before going live.

What the clinic's own policies must cover

Your clinic has its own policies for patient information, and the AI receptionist must follow them. Staff training, access reviews and the response to an incident all still apply. The agent is one more part of the system, not a replacement for those policies, and it should be listed in them.

Review the policies when the agent changes. A new call type, a new integration or a new vendor each change what the agent can see, so each one is a reason to check the policy again.

How VoiFlow handles this

VoiFlow can be set up to verify callers before it discusses bookings. It books and moves appointments in the clinic's calendar, and hands over to staff with a summary when a person is needed. Every call is recorded, transcribed and scored for review, and permissions, redaction and the audit trail are enforced outside the language model.

For what VoiFlow has in place, read the trust page and ask for the documents you need in writing. This article makes no compliance claims for VoiFlow, and your counsel should review any claim before you rely on it.

What to do next

Write down your call types, the verification steps for each one and the information each call may include. Take that list to your clinic's compliance lead and to counsel. When you are ready to plan a pilot, try a call to hear how a verification step sounds, or contact us to talk through the set-up for your clinic.

Frequently asked questions

Does HIPAA apply to every clinic?

Not necessarily. Whether HIPAA applies depends on the business and how it handles patient information. Ask your counsel for your position before you change how calls are handled.

Does an AI receptionist need a business associate agreement?

Ask your counsel whether one is needed. Ask the vendor to confirm in writing whether it will sign one, before any patient information flows through the system.

Should the agent ask why the patient is calling?

Only when the booking needs it. Ask for the purpose of a visit only if your clinic's process requires it, and never make a caller describe symptoms in order to book an appointment.

What if someone calls about another person?

The agent should not share details unless the caller has passed verification and is recorded in your systems as allowed to receive them. It can explain politely that it cannot share that information, offer to take a message, and pass the message to the clinic for review.

What if a caller describes symptoms?

The agent sends emergencies to emergency services at once and passes other clinical questions to a clinician, with a call-back time. It does not give medical advice.

This is general information, not legal advice.

Call it now · United States line+1 (831) 282-9226

Hear it run one of your calls.

Call the agent yourself, start a free trial, or bring us your hardest workflow. Most teams are live in days.