Virtual Receptionist for Doctors: A Worked Call Flow and Script Set

A virtual receptionist for a doctor's office should greet the caller with a fixed line, work out the reason for the call from a short list, handle bookings and messages from approved information, read fixed emergency wording when needed, and pass anything clinical or uncertain to staff. It never gives medical advice. That holds for a human service and an AI one.

Most doctors' offices lose trust on the awkward calls, not the easy ones: a worried parent, a caller who half-mentions chest pain while asking about a prescription, an angry patient on hold. A script that only covers bookings fails on exactly those calls, so the flow below puts the safety check before the reason for the call, and every branch ends in a defined hand-off.

Follow me on Instagram@sagnikteaches

The call flow on one page

Draw the flow before writing scripts. Every branch should end in one of four outcomes: handled and logged, booked, message taken for staff, or transferred now. Here is a flow for a small general practice, in text so you can paste it into your own tool.

Connect on LinkedInSagnik Bhattacharya

1. ANSWER   -> opening line + "How can I help?"
2. SAFETY   -> any urgent words in the first 30 seconds?
               YES -> emergency wording, then transfer to staff / end per rule
               NO  -> continue
3. REASON   -> classify into one of:
               a) book / change / cancel appointment
               b) prescription or repeat medication
               c) test results or letters
               d) symptoms or "I don't feel well"
               e) forms, referrals, sick notes
               f) billing, insurance, records
               g) anything else / unsure
4. ACT      -> a: book from approved slots
               b, c, e: take details, promise a call-back window
               d: ask the safety question once, then message or transfer
               f: answer from the approved FAQ or take a message
               g: take a message, never guess
5. CONFIRM  -> read back name, number, reason, next step
6. CLOSE    -> "Is there anything else?" + approved closing line
7. LOG      -> note, tag, urgency flag, time

Step 2 is the one most scripts forget to place early. Callers do not announce emergencies; they mention them in passing. The safety check therefore runs continuously, not once. A receptionist (human or AI) is told to stop the flow whenever an urgent phrase appears, whatever step it is on.

Subscribe on YouTube@sagnikteaches

The opening script

The opening does four jobs in about ten seconds: names the practice, says who is speaking, tells the caller how to get help in an emergency, and asks for the reason. Because the emergency line is the longest, keep it to one sentence.

"Thank you for calling [practice name]. This is [name], the practice's virtual receptionist. If this is an emergency, please hang up and call the local emergency number now. How can I help you today?"

If a caller is speaking to an AI, say so. Rules on disclosing automated callers differ from place to place and are changing, so ask an adviser what applies where your patients are; plain honesty is the safer default either way: "This is [name], [practice name]'s automated receptionist". Do not name the AI tool in the greeting; do tell callers how to reach a person. A script line that is useful for both types of service: "I can take a message or book you in, or I can pass you to a member of staff if that's easier."

Why the emergency line sits in the opening

A caller in trouble should not have to sit through a menu. Putting the line first costs three seconds on every call and gives the one caller who needs it the shortest possible path. Your clinical lead should approve the wording, and you should test it on every route, including the after-hours message and the transfer-failed message.

Emergency wording and the rules behind it

Two parts: the trigger list and the words. Your clinicians write the list; the receptionist does not judge, it matches. Typical triggers to review with your clinical lead (the list below is a starting point, not clinical guidance): chest pain, trouble breathing, signs of a stroke, severe bleeding, loss of consciousness, a child who is unresponsive, thoughts of self-harm, a possible overdose, and anything the caller calls an emergency.

Emergency wording (template; your clinical lead must approve the exact words):
"This sounds like it may be urgent. Please hang up now and call the local emergency number. If you are not able to, ask someone near you to call. I'm also alerting the practice."

Then, per your rule: (A) transfer to the on-call line, or (B) end the call after the wording and send an urgent alert to staff.

Notice what is missing: no advice, no "try lying down", no "it's probably nothing". The receptionist gives one instruction and acts on the alert. Rule A or B is your decision; both are better than a reassuring chat.

Scripts for the six common call types

Book, change or cancel

"I can help with that. Is this for you or someone else? Can I take the patient's full name and date of birth? ... What's the appointment for, in a few words? ... I have [day, time] with [clinician type] or [day, time]. Which suits you? ... Thank you. You're booked for [day, time]. We'll send a confirmation to [number]. Anything else?"

Rules to attach: book only from slots staff have released; never promise a named doctor unless the system allows it; if the reason for the visit contains an urgent phrase, switch to the symptom branch.

Repeat prescription

"I can take the details for the prescribing team. Can I have the patient's full name and date of birth, the name of the medicine, and the pharmacy you use? ... The team usually reviews requests within [your timeframe], and they will contact you if they need anything. Is [number] the best number?"

The receptionist does not confirm doses, advise on missed tablets or promise a time that staff cannot keep. Use your own timeframe; do not borrow one from this page.

Test results and letters

"I can't give out results on this line, but I can pass a message to [the team]. Can I take the patient's full name and date of birth and the best number to call you on?"

Verify identity as your own policy requires before saying anything about a patient. For an AI service, set the rule so it never reads out clinical information, even to someone who gives the right name and date.

Symptoms and "I don't feel well"

"I'm sorry you're not feeling well. I can't give medical advice, but I can get a nurse or doctor to call you. First, is this an emergency, or is anyone having trouble breathing, severe pain, or heavy bleeding? ... [If no] Can I take the patient's name, date of birth, a short description in the caller's own words, and the best number? A member of the team will contact you within [your timeframe]."

Ask the safety question once, then stop asking questions. A receptionist who interrogates symptoms is triaging without training.

Forms, referrals and sick notes

Take the patient's name, the form type, the deadline and how the patient wants to receive it. Do not promise that a clinician will sign anything. Say the team will review the request.

Billing, insurance and records

Answer from an approved FAQ only: accepted payment types, opening hours, how to request records. For anything about a specific bill or coverage, take a message. Do not guess at amounts.

Hand-off rules: when the call must go to staff

Two points about transfers. First, decide where a transfer lands when nobody picks up: a second number, a voicemail with an urgent flag, or a text to the on-call clinician. A transfer that rings out is worse than no transfer. Second, a service may charge for transfers or limit the number of destinations; check your plan. For example, Smith.ai's published human plans include one, two or ten free transfer destinations depending on tier, and list extra destinations at $15 a month each, so count the routes your flow needs (on-call line, nurse line, practice manager) before you choose a tier.

The message you want staff to receive

A good message saves a call-back. Specify the format once and require it from every call, whether it is typed by a person or generated by an AI service.

Caller: [name]  | Patient: [name, date of birth] | Number: [..] | Best time: [..]
Reason (caller's words): [..]
Type: appointment / prescription / results / symptoms / form / billing / other
Urgency flag: none / call back today / URGENT
Safety question asked: yes - answer [..] | not applicable
What I told the caller: [..]

Before and after

Weak message: "Mr [surname] called about his tablets, please call back." Strong message: "Caller: [first name] [surname]; patient: same, date of birth [..]; number: [..]; best time: mornings. Reason: asks for repeat of [medicine], pharmacy [name]. Type: prescription. Urgency: call back today because he says he has two days left. Safety question not applicable. Told him the team reviews requests within [your timeframe]." The second can be actioned without a call to the patient to ask what he wants.

Writing the script set with AI

If you use an AI service, you will need to give it the rules in writing. Even if you use a human service, the same document is the brief they work from. Ask a general AI assistant to tidy your rules into a script, but give it all the facts and tell it what not to add.

Rewrite my receptionist rules as a script for a virtual receptionist. Keep each caller line under 25 words. Do not add medical advice, reassurance, timeframes, opening hours or phone numbers. Where I have written [ ], leave the brackets. Output: (1) opening line, (2) one block per call type, (3) hand-off table.
My rules:
[paste your call types, emergency wording, hand-off rules]

Check the result line by line. AI tends to add "don't worry" or "it's probably nothing", which you must remove. It may also invent opening hours; fill those in from your own sheet. Keep the final script in one document with a version number and a date, so you can show which version was live on a given day.

Check calls: prove it follows the script

A script is only as good as the last time someone tested it. Make a set of scripted check calls, run by a member of staff who is not on the front desk, and score each on the same sheet. Keep calls short and use made-up details, not a real patient's.

Score each call out of three: followed the script, said nothing it should not have, and produced a usable message. A single failure on call 2 (the hidden emergency) is serious: fix it before anything else, and retest the same call until it passes three times running. If you use an AI service with a testing feature, use it too; for example, Smith.ai's AI receptionist plans list monthly test and simulated call allowances (25 test and 50 simulated calls even on its free plan), and your own provider may offer the same. A simulated call is not a substitute for a real phone call to your real number.

The dizzy caller the trigger list missed

An illustration of a typical gap: an AI service handles routine calls well for two weeks. A check call in week three, with a caller who says "I'm a bit dizzy but I just want to rebook", gets a booking and a cheerful goodbye. The script has a trigger list for chest pain and breathing, but "dizzy" is not on it. The fix is to add dizziness and fainting to the list after the clinical lead reviews it, and to add a general line: if the caller describes any physical symptom while doing something else, ask the safety question. A weekly check call finds the gap before a patient does.

One day of calls, sorted through the flow

Illustrative, for a four-doctor practice on a Monday. The phone logs 120 calls. Using the seven call types, a plausible split might be: 38 appointments, 22 prescriptions, 14 test results, 18 symptom calls, 10 forms and referrals, 12 billing and records questions and 6 unclassified. The numbers are not a benchmark; count your own for a week and use those.

Now apply the flow. Appointments (38) can be booked from approved slots without staff if the service has calendar access, saving most of the time. Prescriptions (22) and results (14) become messages: no live conversation with a clinician is needed, but each message must carry the details the team needs. The 18 symptom calls are the important ones. Suppose 3 of them include an urgent trigger. Those three go through the emergency route; the other 15 become flagged messages that a nurse calls back. The 6 unclassified calls become messages or transfers.

The outcome to examine is not how many calls the service "handled" but how many staff calls were removed and how many were created. If 120 calls now produce 75 well-formed messages and 3 urgent transfers, the front desk's morning begins with a clear list rather than a ringing phone. If the same messages are vague, you have swapped ringing for guesswork. Measure the quality of messages the same way you measure the calls.

Callers who are harder to serve

Four groups need a rule written in advance, because they are where a script falls apart.

• Callers who struggle to hear or speak clearly. Offer to take a number and have someone call back, rather than asking for repeats more than twice.
• Callers who speak another language. Check which languages the service covers. Some AI plans list a limited set (Dialzara's plans, for instance, list English and Spanish), and a human service may use interpreters at extra cost. State your fallback: a message in the caller's own words, passed to staff.
• Callers calling for someone else. Take the caller's details and the patient's, note the relationship, and do not discuss the patient's information without your policy's consent check.
• Callers in distress who are not describing a medical emergency. Allow a longer pause, offer a transfer, and do not rush to close. Your clinical lead should supply wording for these situations.

Briefing a human service versus configuring an AI one

With a human service, your script becomes a training document. Send it with a one-page list of the ten things staff must never say, your trigger list, your transfer numbers and how to reach you. Expect to review calls in the first fortnight and to answer questions from the provider's supervisor; a good provider will ask for the awkward cases. Request that call recordings or summaries are available to you for checking, and ask how quickly the receptionist team can be updated when you change a rule.

With an AI service, the script becomes configuration. The same content goes into the prompt, the knowledge base and the routing settings, and it is tested with simulated calls. The advantage is consistency; the risk is confident wrongness on cases the script did not foresee. Give the AI a firm default: when unsure, take a message. In both cases, name one person on your team as the owner of the script and its version history.

Keeping the script alive

Review the script monthly in the first quarter, then quarterly. Each review asks three questions: which call types are increasing, which messages needed a follow-up call to clarify, and which check call failed. Update opening hours, holiday messages and any new service line the moment they change. A script that still says "we close at five" when you now close at six will quietly annoy patients for months.

What a good service should tell you, in numbers

Ask the provider for a monthly report, or build one yourself from call logs: number of calls, number handled without staff, messages by type, urgent flags raised, transfers attempted and answered, and average time to first answer. Compare against your staff's experience. If urgent flags are zero for a month, ask whether the service is detecting them; if they are very high, the trigger list may be too broad.

Costs differ widely between a human service and an AI one, and by volume. The comparison, with price bands for each model, is in virtual receptionist for a small clinic: live, AI or hybrid. For the broader basics of how an AI receptionist books, see what an AI receptionist is and how it handles bookings.

Privacy and consent, briefly

Calls to a doctor's office contain health information. Before a service starts answering, check where recordings and transcripts are stored, for how long, who can read them, and whether the provider offers the compliance options your practice needs; some providers list these only for human answering, so ask about the exact service you plan to use. Tell callers when calls are recorded, in the greeting or the privacy notice. Do not send patient names into a general-purpose chatbot to draft scripts: use placeholders as above. Ask an adviser who knows the rules that apply to your patients to review the provider's terms.

A rollout in three weeks

1. Week 1. Draw the flow, write the trigger list with your clinical lead, write the scripts, and approve the message format.
2. Week 2. Set up the service on overflow or after-hours calls only. Run all eight check calls twice. Fix, then retest.
3. Week 3. Extend to more hours, if the checks pass. Keep weekly check calls for a month, then monthly.

If you want help turning your own call types into a flow like this and setting up the check calls, that is what my AI implementation consultation covers, and you can book a 1:1 call. Two-site practices should also read routing and hand-offs for a medical clinic with two sites, because routing rules double in complexity.

Questions doctors ask about virtual receptionists

Should a virtual receptionist ever give medical advice?

No. A receptionist, human or AI, takes details, books, routes and reads approved wording. Anything resembling advice, triage or interpretation of symptoms belongs to a clinician. Your script should say so, and your hand-off rules should send any symptom question to a nurse or doctor.

What should the emergency wording be?

Short, fixed and approved by your clinicians: tell the caller to hang up and contact the local emergency number if the situation is urgent, and give no further advice. Test it on every call route, including after hours. Your own clinical lead should write and sign off the exact words.

Does the same script work for a human and an AI service?

Largely yes, if you write it as rules and short lines rather than a long monologue. Humans handle odd cases better; AI follows the script more consistently. Test both with the same check calls and compare how each deals with the awkward ones.

How often should I test the service?

Weekly in the first month, then monthly. Make a handful of scripted calls, including one emergency and one angry caller, score them on the same sheet and keep the results. Retest whenever you change the script, the provider or the phone system.

Further reads

• AI Receptionist for Physical Therapy: Referral and Insurance Questions — Referral and insurance questions on a therapy line.
• How to Measure an AI Receptionist's Return in the First 90 Days — Measure results over the first 90 days.
• What to Do When an AI Receptionist Gets a Booking Wrong — What to do when a booking goes wrong.
• AI Receptionist for UK Dental Clinics: NHS, Private and Hand-Offs — Hand-offs and private versus public patients at a dental clinic.
• AI Receptionist or Answering Service for a Booking-Based Business? — Compare an AI receptionist with a human answering service.
• Questions to Ask an AI Vendor Before You Sign Anything — Questions to put to any provider before signing.
• AI Answering Service for Doctors: What It Costs and What You Get — Compare per-call and per-minute bills, then specify the on-call hand-off your practice actually needs.
• AI Answering Service for a Doctor's Office: After-Hours Call Rules — Build an after-hours call matrix, approved emergency script and test set before forwarding patient calls to AI.
• AI Chatbot for a Doctor's Website: Scope, Disclosure, Handover — A practice owner's set-up guide: scope sheet, disclosure line, hand-over rules, logging and named chatbot tools with costs.
• AI Platform for Doctors: One Suite or Separate Scribe and Phone Tools? — A buying rule for small practices: when one suite wins, when separate scribe and phone tools win, and the integration test to run first.
• Orthodontic Staff Training: AI Role-Play Scripts for Phone Calls — Role-play prompts, a call-type list, a scoring sheet and a quiz bank for orthodontic front-desk phone training, built with AI and reviewed by a coordinator.
• Patient Intake Automation in a Small Practice: Forms, AI Summaries — How a small practice sets up digital intake forms, consent capture and a clinician-checked AI summary, with retention and privacy rules.
• Patient Participation Group: Agendas, Minutes and Newsletters With AI — A practice manager's AI workflow for patient participation group agendas, minutes drafted from notes and newsletters, with the checks to keep it safe.
• Patient Recall Letters for Reviews and Screenings: AI Drafts — Recall letter templates by type, the wording rules that keep them safe, opt-out handling and a tracking list, all drafted with AI and checked by a person.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Smith.ai human and AI receptionist pricing pages, Abby Connect pricing page and Dialzara pricing page, read November 2026. Clinical wording in the scripts is a template: your own clinicians and a qualified adviser must approve it.