AI Receptionist for Clinics: What It Handles
Jaseem Abbas
- Published

An AI receptionist for clinics is a voice agent that answers the clinic's phone, books, moves and cancels appointments, answers routine questions such as timings and directions, and hands anything medical to a person. It does the phone work of a front desk and none of the clinical work.
Here is the problem it is built for. A two-doctor dental clinic in Kochi has one receptionist. At 11:40 she is billing a patient at the counter while the phone rings for the third time. At 9:15 that night, a patient calls to move tomorrow's root canal to Thursday, and nobody answers. By morning the slot is empty, and the patient has rung another clinic. The clinic in this story is made up, though the problem is not.
This guide is for clinic owners and practice managers in India. It covers what the agent can handle, what it must never do, how to handle after-hours calls, languages, the rules on patient data, how to set one up and test it, and what to measure. We found no verified public statistic on missed clinic calls, so there is no scary percentage here. Every rule that comes from a source is linked.
What is an AI receptionist?
An AI receptionist is software that talks on the phone the way a front desk does: it picks up, understands what the caller wants, does the task or passes it on, and records the result. Searches for "ai receptionist" and "ai voice receptionist" mean the same thing.
Three things are often mixed up, so here is the difference.
- IVR menu. The "press 1 for appointments" system. It sorts calls by key press and cannot hold a conversation. Our guide to IVR vs voicebot explains why callers dislike it.
- Virtual receptionist. A human who answers your calls from somewhere else, often for several businesses. Useful, but it is a person on a shift.
- AI receptionist. A voice agent that listens and speaks, and works from your calendar and your written instructions.
Small clinics are small businesses, so the same idea applies to any "ai receptionist for small business". The clinic version differs in what it must refuse to do, and that is most of this guide. For the wider split between incoming and outgoing calls, see inbound vs outbound calls.
What can an AI receptionist handle for an Indian clinic?
It can handle the routine calls that make up most of a front desk's phone time: new bookings, reschedules, cancellations, timings, directions and callbacks. Below is the list that Indian vendor pages and our own healthcare page describe, and what each call needs from the clinic.
Caller says: I want to see Dr. Nair this week
- What the agent does
- Offers free slots and books one
- What it needs from the clinic
- A calendar the agent can read and write, connected through an integration
Caller says: Can I move tomorrow's visit to Thursday?
- What the agent does
- Finds a free slot, moves the booking
- What it needs from the clinic
- The same calendar, plus a rule on how late a change is allowed
Caller says: I need to cancel
- What the agent does
- Cancels and records it, so the slot can be offered again
- What it needs from the clinic
- A place to log the cancellation
Caller says: What are your timings on Sunday?
- What the agent does
- Answers from your document
- What it needs from the clinic
- A short document with timings and holidays
Caller says: Where are you, is there parking?
- What the agent does
- Gives directions and landmarks
- What it needs from the clinic
- The same document
Caller says: What should I bring?
- What the agent does
- Lists what to carry for the visit
- What it needs from the clinic
- A line per visit type in the document
Caller says: What is the consultation fee?
- What the agent does
- Gives the fee only if you wrote it down
- What it needs from the clinic
- The fee in the document, or an instruction to pass the question to staff
Caller says: Please ask the doctor to call me
- What the agent does
- Takes a name, number and reason, and books a callback
- What it needs from the clinic
- Someone who owns the callback list
Two notes on that table. First, the agent fills slots; it does not own them. Your calendar or booking system is still the source of truth, and the agent reads and writes to it. In our own product, the agent books a new appointment into your calendar through an integration, and on an existing booking it can confirm it, move it to a free slot, cancel and record it, or book a callback when the person is busy. Second, the agent answers routine questions from documents you upload. Bananaflow accepts PDF, DOCX, TXT and Markdown files, so a one-page "clinic facts" document is enough to start. Our healthcare page shows both the booking and question flows.
What about reminder calls?
Reminder calls are the other direction: the clinic phones the patient the evening before. That is a separate job with its own script and measures, and we cover it in full in our guide to doctor appointment reminder calls. This post stays on the line that rings at the clinic.
What must an AI receptionist never do?
An AI receptionist must never diagnose, advise on medicines, read reports, or judge how serious a symptom is. India's telemedicine guidelines are clear on the principle. Clause 5.4 says "Technology platforms based on Artificial Intelligence/Machine Learning are not allowed to counsel the patients or prescribe any medicines to a patient. Only a RMP is entitled to counsel or prescribe" (Telemedicine Practice Guidelines, 2020). RMP means registered medical practitioner.
Be exact about what that document is. It governs telemedicine, not front desks, and it dates from March 2020. We did not check whether it has since been replaced, so read it as "as of the 2020 guidelines", and read it as a clear sign of where the line sits, not as a ruling on receptionist software. Ask your lawyer or your medical association if you need a firm answer.
For a clinic owner, the practical list looks like this:
- No diagnosis. "Is this rash serious?" gets "The doctor will need to see it. Shall I book you a slot?" and never an opinion.
- No medicine advice. "Can I take this with my BP tablet?" goes to staff.
- No reading of reports. The agent should not interpret results, even if the caller reads numbers out.
- No triage. The agent should not decide who is urgent. It books, or it hands over.
- No promises on the doctor's behalf. It should not say what the doctor will prescribe, or whether a procedure is needed.
On a clinical question, our product transfers to a person on Twilio, Plivo, Telnyx, Cloudonix and Asterisk, or books a callback on Exotel, Vobiz and Vonage, and leaves the transcript on the call record. The healthcare page describes this. See also our glossary entry on call transfer.
What should it do when a caller describes an emergency?
It should tell the caller to dial 112, or to go to the nearest emergency department, and pass the call to staff where it can. 112 is India's single emergency number, run under the Ministry of Home Affairs' Emergency Response Support System. The same 2020 guidelines say telemedicine "should however be avoided for emergency care when alternative in-person care is available" (clause 1.4.2.3, same document). A phone agent is no better placed than a video call.
This is the most important rule in this post: the emergency line is a rule you write. Write the rule into the agent's instructions, and a workflow can branch on what the caller says. A rule worth starting from:
If the caller describes chest pain, difficulty breathing, heavy bleeding, an accident, a fainting spell, or a child who is limp or not responding, do not book anything. Say: "Please hang up and dial 112 now, or go to the nearest emergency department. I am passing a note to our staff." Then transfer the call to staff if a transfer is possible.
Then test it. Call the agent and say each of those sentences in each language you offer. If it ever says "let me check the next slot" to a person describing chest pain, the rule is not ready. Our prompt guide covers how to write rules like this so the agent holds to them.
What does a sample call look like?
A good call is short, names the clinic, does one job and ends with a clear result. Here is a booking call at a paediatric clinic in Lucknow. The names and times are invented.
Agent: Namaste, this is the Bal Swasthya Clinic in Lucknow. How can I help? Caller: Mujhe Dr. Verma ka appointment chahiye, bachche ke liye. Agent: Zaroor. Aap kis din aana chahenge? Caller: Kal shaam ko. Agent: Kal shaam 5:30 baje ek slot khali hai. Kya main wo book kar doon? Caller: Haan, kar do. Agent: Done. Dr. Verma, kal shaam 5:30 baje. Kya aur kuch madad kar sakti hoon?
Now a call where the agent must hand over:
Caller: My son took two doses of the syrup by mistake. What should I do? Agent: I cannot advise on medicines, and I do not want you to wait. Please call 112 now or go to the nearest emergency department. I am passing your number to our staff right away.
The second call has no booking, no guess and no delay. That is the behaviour to test for. Notice also what the first call leaves out: no question about why the child needs to see the doctor. The reason for a visit is health information, and the agent does not need it to book a slot.
How should an AI receptionist handle calls after hours?
After hours, it should answer, book or move appointments, take a callback request for the morning, and give the emergency number. That is the main reason clinics look at this tool: the 9:15 pm call in Kochi gets answered. Our inbound calls page describes after-hours and overflow answering, and a missed call service explains the older fix.
Be careful about what you promise patients. Do not say "the doctor will call you tonight" unless your own clinic arranges that. A simple honest line is: "Our staff will call you back at 9 am. For an emergency, please dial 112."
Three set-up choices matter after hours:
- Which actions are allowed. Book and move appointments, yes. Give a fee for a procedure, maybe not.
- Who owns the callback list. If nobody reads it at 9 am, the patient is worse off than before.
- What the opening line says. Say the clinic name and offer help. If you record calls, plan how you tell patients, which we come back to below.
Which languages should an AI receptionist speak?
It should speak the languages your patients use on the phone, which in most Indian cities means more than one. A patient in Hyderabad may start in English, move to Telugu and finish in Hindi in the same call. Bananaflow supports English plus 10 Indian languages (Hindi, Bengali, Gujarati, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil and Telugu), follows the caller's language from their first clear sentence, and can switch mid-call. Not every voice speaks every language, so check the voice you pick.
One honest limit. With two or more Indian languages switched on, a short first word like "haan" can be misheard. Test the opening seconds of a call in each language, and keep your language list to the ones you really need. Many patients also mix Hindi and English in one sentence, which we cover in our guide to Hinglish voice agents.
Names are the other trap. Doctor names and area names are spoken very differently in Kerala and in Uttar Pradesh. A speech-to-text word list for names and clinic jargon is part of the product, so add yours and listen to the result.
What rules apply to patient calls in India?
The main one is the Digital Personal Data Protection Act, 2023, together with its 2025 rules. This is not legal advice, so check details with your lawyer. As of November 2026, three points matter for a front desk.
- A patient's name, phone number and booking are personal data. The DPDP Rules, 2025 were notified in November 2025, with most duties starting after an 18-month phase-in, per the government's press release. The same release mentions clear consent notices, breach notices and verifiable consent for children's data, with some exemptions including healthcare. If you treat children, read that part.
- The Act has no separate "sensitive" category. The word "sensitive" does not appear in the DPDP Act, 2023. That does not make health details casual. Treat them as the most private thing the agent could hear, and keep them off the call.
- There is a medical-emergency rule. Section 7(f) lets a data fiduciary process personal data without consent "for responding to a medical emergency involving a threat to the life or immediate threat to the health of the Data Principal or any other individual". It is a narrow rule for emergencies, not a general permission for booking calls.
The 2020 telemedicine guidelines also bind doctors to the IT Act, data protection and privacy laws for patient confidentiality (clause 3.7.1). For the rest, see our guide to DPDP rules for call centers and our post on whether call recording is legal in India.
In practice, three house rules cover most of the risk:
- Ask for a name, a phone number and a slot, and nothing about the condition.
- Never confirm details to a caller who is not the patient or the patient's family.
- If you record, tell the caller, and decide how long you keep recordings.
On our side, call audio, recordings and transcripts are not used to train or fine-tune models, and recordings are stored in India (Mumbai). That covers recordings only. DPDP duties sit with the clinic.
How do you set up an AI receptionist for your clinic?
You set it up in seven steps, and the first five are writing, not technology. A multi-doctor clinic in Hyderabad can do this in a few days if one person owns it.
- Write one clinic facts document. Timings by day, holidays, doctors and their days, visit types, fees you are happy to state, directions, parking and what to bring. Keep it to a page or two. PDF, DOCX, TXT and Markdown all work.
- Connect the calendar. The agent needs to read free slots and write bookings, through an integration or an HTTP tool. Decide the rules: how far ahead, how late a change is allowed, which doctor takes which visit type.
- Write the "never do" list. Use the list above: no diagnosis, no medicines, no reports, no triage.
- Write the emergency rule and the sentence it says. Put it first in the instructions, not buried at the end.
- Decide who gets handovers. Name the person who takes a transfer in clinic hours, and the person who reads callbacks at 9 am. Transfer runs on Twilio, Plivo, Telnyx, Cloudonix and Asterisk, so check that your carrier is one of them.
- Test it yourself. Call it from your own phone, in each language, as a rude caller, a confused caller and a caller in distress. In our product you can test the agent in the browser, by voice or text, before it takes a single real call.
- Go live on the real number, then read transcripts. Read every transcript for the first week. Fix the instructions, not the person.
Bananaflow works on your own numbers and carrier. Our guide to writing a voice agent prompt and our walkthrough of how to build an AI calling agent cover steps 3 and 4 in detail.
Where Bananaflow fits
Bananaflow is an AI voice agent platform. For a clinic, that means an agent that books, moves and cancels through your calendar, answers routine questions from your documents, speaks English and 10 Indian languages, and passes medical calls to a person or books a callback. It gives no medical advice. After each call you get a transcript, a recording and an outcome code. Our healthcare page has the full list, and appointment confirmation calls covers the booking calls in more detail.
Pay as you go is Rs 4.00 a minute plus 18% GST with no monthly fee, and your carrier bills its own charges separately. Plans can change, so check the pricing page for current figures.
What should you measure on an AI receptionist?
Measure what the front desk used to count by hand, and read the transcripts for what it never could. We give no benchmark figures because we have none from customers. Compare your own numbers before and after.
Measure: Calls answered vs calls received
- How to get it
- Your carrier's call log against the agent's call list
- What it tells you
- Whether anyone is still hitting a busy line
Measure: Bookings made, moved and cancelled
- How to get it
- Outcome code on each call record
- What it tells you
- What the line is actually used for
Measure: Handovers and callbacks
- How to get it
- Call records marked transferred or callback
- What it tells you
- How many calls still need a person, and whether staff return them
Measure: Time to callback
- How to get it
- Callback list against time of call
- What it tells you
- Whether the 9 am promise is kept
Measure: Emergency rule triggers
- How to get it
- Search transcripts for the 112 line
- What it tells you
- Whether the rule fires, and fires only when it should
Measure: Wrong answers
- How to get it
- Weekly transcript read, ten calls
- What it tells you
- Facts missing from your clinic document
Read ten transcripts a week for the first month. Every wrong answer is a missing line in your document or your instructions, and it is cheap to fix. If a patient asks something twice and the agent fails twice, add the answer or send that question to staff.
What should you remember about AI receptionists for clinics?
An AI receptionist answers the clinic's incoming line and does the routine phone work: book, move, cancel, timings, directions and callbacks. It never gives medical advice, because India's 2020 telemedicine guidelines say AI platforms may not counsel patients or prescribe, and it should send emergencies to 112. The emergency rule is something you write and test. Pick the languages your patients speak, keep health details off the call, tell people if you record, and name who takes handovers. Test it by phone before any patient does, then read the first week of transcripts.
FAQ
Frequently asked questions
What does an AI receptionist do for a clinic?
It answers the clinic's phone, books, moves and cancels appointments, answers routine questions such as timings, directions and what to bring, takes a callback request, and hands anything medical to a person. It does the front desk's phone work, not the clinical work. You decide the exact list when you write its instructions.
How much does an AI receptionist cost?
Pricing for voice agents usually has two parts: the voice software, charged by the minute or a monthly plan, and your phone carrier's call charges. On Bananaflow, pay as you go is Rs 4.00 a minute plus 18% GST with no monthly fee, and the carrier bills its own charges separately. Plan prices can change, so read the current figures on the pricing page before you decide.
Can an AI receptionist give medical advice?
It should not. India's 2020 telemedicine guidelines say AI platforms are not allowed to counsel patients or prescribe any medicines, and that only a registered medical practitioner may. A clinic's receptionist agent should answer a symptom or medicine question with a handover to staff or a callback, never with an opinion.
What happens if a patient calls with an emergency?
That depends on what you write into the agent's instructions, because the emergency line is a rule you set. A sensible rule is that, if the caller describes chest pain, heavy bleeding, an accident or a faint, the agent tells them to dial 112 or go to the nearest emergency department, and passes the call to staff where it can. Test this rule before launch.
Can patients still talk to a person?
Yes, if you set it up that way. The agent can transfer the call to a person on Twilio, Plivo, Telnyx, Cloudonix and Asterisk, or book a callback on Exotel, Vobiz and Vonage, and the transcript stays on the call record. Decide who answers a transfer and who returns a callback before you go live.
Is patient call data used to train AI?
No. Bananaflow confirms that call audio, recordings and transcripts are not used to train or fine-tune models, either its own or its providers'. Call recordings are stored in India (Mumbai). That covers recordings only, so ask any vendor where the rest of your data runs.
Will it reach the doctor after hours?
After hours the agent can answer, book or move an appointment, take a callback request for the morning and give the emergency number. For an on-call doctor, set your own on-call rule and tell patients what it is.




