Design a medical office AI receptionist for scheduling, reminders, insurance routing and HIPAA-aware escalation with clear Fish Audio speech boundaries.
An AI receptionist for a medical office handles the routine phone work that does not require clinical judgment: appointment requests, confirmations, reminders, preparation instructions, insurance routing and message taking. Fish Audio supplies transcription, streaming text to speech, a consistent voice and API access. Your phone system, EHR or practice-management platform and clinical team own scheduling, patient records, triage and care. Fish Audio is not an EHR and does not claim HIPAA compliance or a BAA.
The boundary is the key design decision. The agent should make the front desk faster and the after-hours line less silent, without pretending to be a clinician.
A safe first scope:
Keep these with clinicians or staff:
For related scheduling mechanics, see AI voice agents for appointment booking. For dental-specific intake, see AI receptionists for dental offices.
Agent: Thanks for calling Cedar Family Medicine. Is this about an appointment, a prescription or something urgent?
Caller: I need to move my appointment. I also have chest tightness.
Agent: Chest tightness can be an emergency. If it is happening now, please hang up and call 911. Is the tightness happening right now?
Caller: It comes and goes.
Agent: I am connecting you to our nurse line now. Please stay on the line.
The agent stops the scheduling flow when a red flag appears. That is the correct behavior even if the caller originally asked for something routine.
Connect the agent to narrow endpoints:
Do not give the agent a general-purpose database connection or the ability to read full charts. Use scoped service accounts and log every write.
Collect the minimum necessary information for the task. A booking call does not need a full medical history, a Social Security number or a card number. Send forms through a secure link.
Disclose recording before collecting patient information and follow the consent rules in the relevant jurisdiction. Keep patient identifiers out of analytics events, debug logs and error messages. Define retention for transcripts and recordings before launch.
HIPAA obligations depend on the practice, the vendors and the data flow. Have counsel review the architecture and any business associate agreement. Fish Audio supplies speech capabilities; the practice and its integrated systems remain responsible for compliance.
Use a short, approved script. For chest pain, difficulty breathing, signs of stroke, severe bleeding, fainting or a severe allergic reaction, tell the caller to contact emergency services immediately. Do not attempt to assess severity or advise on medication.
For mental-health crisis calls, route to the appropriate local crisis line and a trained human. In the US, 988 is the Suicide and Crisis Lifeline. The agent should stay on the line only as long as the approved protocol allows and should not counsel the caller.
Document the red-flag questions, the caller's answers and the transfer outcome.
| Failure | Safe fallback |
|---|---|
| Scheduling API times out | Do not confirm; take a message and offer a callback |
| Patient identity is ambiguous | Ask one clarifying question, then transfer |
| Caller describes a red-flag symptom | Stop scheduling and follow the emergency script |
| Refill request is outside policy | Route to the clinical team; never promise approval |
| Insurance answer is uncertain | Read the approved list or transfer |
| Caller requests test results | Transfer to the clinical team |
| SMS confirmation fails | Keep the appointment and escalate the notification |
Log the tool response and the transfer reason with the call record.
| Metric | What it tells you |
|---|---|
| Booking completion rate | Whether routine calls convert to appointments |
| After-hours capture rate | Whether callers reach a useful next step |
| No-show rate | Whether reminders and preparation copy work |
| Transfer rate | Whether the scope is narrow enough |
| Red-flag routing accuracy | Whether urgent calls are handled safely |
| p95 time to first audio | Whether the agent feels responsive |
| Patient complaint rate | Whether tone and boundaries are appropriate |
Audit urgent-call transcripts separately. A booking metric cannot compensate for a missed emergency.
Natural speech is roughly 150 words per minute, close to 900 characters. One Fish Audio credit covers about 1,000 characters, so agent speech starts near one credit per minute before repetitions or longer instructions.
| Pack | Price | Credits | Approximate agent speech |
|---|---|---|---|
| Starter | $9 | 1,000 | About 16 hours |
| Creator | $29 | 4,000 | About 66 hours, plus commercial rights |
| Studio | $99 | 15,000 | About 250 hours, plus API access |
EHR or practice-management software, telephony, transcription, SMS and staff time remain separate. The TTS API pricing guide converts characters into a full budget.
It can answer routine calls, read approved information, book or reschedule appointments and route urgent calls when connected to your phone and scheduling systems. Fish Audio supplies the speech layer; the practice remains responsible for clinical care and records.
Fish Audio provides speech generation and transcription; it is not a medical practice-management platform and this guide does not claim HIPAA compliance or a business associate agreement. The practice must evaluate its full data flow and vendors with counsel.
No. It should route clinical questions, test results, medication changes and symptoms to a nurse or clinician. It may read approved preparation instructions and office policies.
Use a short red-flag script. For chest pain, trouble breathing, stroke signs or severe bleeding, direct the caller to emergency services immediately. For mental-health crisis calls, use the local crisis line and transfer to a human.
At roughly 150 words per minute, one minute of agent speech is about one Fish Audio credit. EHR, phone service, transcription, SMS and staff time are separate costs.