AI Receptionist for Medical Clinics With Two Sites: Routing and Hand-Offs

To route calls between two clinic sites, decide first whether patients dial one number with a short menu or two numbers. Then give the AI receptionist a separate fact sheet for each site, written transfer rules for when a site is closed, and a live check of the shared booking calendar before it offers a time.

Most failures in a two-site clinic are not about the AI at all. They come from rules nobody wrote down: which site handles which service, where calls go at lunch, what happens when a site is closed for a staff meeting. An AI receptionist exposes every one of those gaps on day one, so the writing-down is the real work.

Follow me on Instagram@sagnikteaches

One number or two: the decision that shapes everything else

This choice affects menus, scripts, reports and what you tell patients. Use the table to decide, then stick to it for the first three months.

Connect on LinkedInSagnik Bhattacharya

Two numbers is simpler to run and simpler to measure. The receptionist at each number only needs that site's facts, and a wrong-site mistake is rare because the caller chose the number. One number is better for new patients and for services that move between sites, but the AI has to ask "which site?" and handle callers who do not know. It also shifts risk onto the menu: a caller who gives up during a menu is a lost booking.

Subscribe on YouTube@sagnikteaches

A middle route that often works

Keep both public numbers and forward them to the same AI receptionist configuration, with each number telling the AI which site the caller dialled. The assistant then greets with the right site name, and can still hand over to the other site when asked. That preserves patient habits and gives you one place to manage rules. Check with your phone provider and the AI vendor that the dialled number is passed through, because not every setup does that.

Write a fact sheet for each site, not one for both

A shared document with "Site A does X, Site B does Y" invites mix-ups. Write two separate sheets with the same headings, so the AI can answer from the right one. Illustrative extract for one site:

SITE: North Street Clinic
Address and entrance: 12 North Street, side door for step-free access
Parking: six spaces behind building, no charge for 90 minutes
Opening hours: Mon-Fri 8:00-18:00, Sat 9:00-12:00
Closed: public holidays; staff training last Wednesday of each month 12:30-14:00
Services here: doctor appointments, nurse clinic, blood tests (mornings only), travel vaccines
Not here: physiotherapy, minor surgery
Phone for staff transfer: ext. 201 (reception), ext. 205 (nurse line)
Urgent-call instruction: [exact wording the clinic approves]
Booking rules: new patients book 20 minutes; follow-ups 10 minutes; blood tests need no doctor slot

The "Not here" line is as useful as the services line. Callers ask "do you do physio?" far more often than the clinic expects, and the right answer is "not at this site, but our other site does, shall I book there?". Write both sheets in the same order, then test the assistant on questions that cross sites.

Keep the sheets current

Name an owner per site, and put a review date on each sheet. The commonest cause of embarrassing errors is a changed opening time or a moved service that never reached the sheet. A fortnightly five-minute check by each site manager costs almost nothing. The tutorial on setting up an AI receptionist shows where the fact sheet sits in a first build.

Routing rules, written as plain sentences

Before touching any tool, write the routing rules in sentences a new reception hire could follow. Illustrative set for a clinic with a main site (Site A) and a smaller site (Site B):

1. If the caller says they want an appointment, ask which site. If they do not know, offer the nearest by name and the services each offers.
2. Routine bookings go to the booking flow for that site; check the calendar before offering a time.
3. Prescription and results queries are logged as a message for that site's team, never answered by the AI.
4. Anything the caller describes as urgent, severe or involving chest pain, breathing, heavy bleeding or a child who is very unwell goes to the urgent instruction immediately.
5. If the caller asks for a named clinician, check which site that clinician works at that day.
6. If the caller asks for a person, transfer to the site's reception line; if no one answers in 20 seconds, offer a callback.

The urgent rule must be written by a clinician, not the practice manager. It should state exactly what the receptionist says, such as telling the caller to hang up and call the emergency number. Do not let the AI improvise in that situation. If you want a fuller list of call types and escalation triggers, the tutorial on call scripts and escalation rules covers the wording.

What happens when a site is closed

This is the hardest case to get right, and the one patients remember. A site can be closed for a planned reason (training, holiday, building work) or an unplanned one (staff sickness, power cut). The AI needs rules for both.

Planned closures

Put planned closures in the site's fact sheet with start and end times, and have the receptionist state them: "North Street is closed this Wednesday afternoon for staff training. I can book you at our other site or for Thursday here." Ask staff to update the sheet at least a week ahead, and set a standing reminder to check it on the last working day before a public holiday.

Unplanned closures

Give a named person a way to flip a site to "closed" within a minute, ideally from a phone, with a message such as "This site is temporarily closed. I can book you at the other site, take a message, or, if it's urgent, give you the right number to call." Test the flip before you need it. The worst version is a closed site whose AI cheerfully offers slots because nobody turned it off.

For both types, the receptionist should still run the urgent check first, then offer the other site and only then take a message. Never leave a patient with "we are closed" and no next step. If your clinic also closes for lunch, treat that as a planned closure every day, because that is where a lot of the first-month confusion comes from.

Checking the shared calendar before it promises a time

The assistant should never offer a slot it has not just checked. Whether it can depends on your booking system and the vendor's integration, and you need to verify this with a real test, not a feature list. Ask the vendor three questions: can it read availability for both sites; can it write a booking back; and what happens if the booking system is slow or down.

If it can read but not write, use it for answers and callbacks only: "I can see there's a Tuesday morning slot at the main site; I'll pass your details to the team to confirm." If it can do neither, it takes messages. A clinic that cannot connect its calendar should say so plainly in the plan, because a receptionist guessing at availability will double-book.

A worked check, illustrative: run twenty test calls across a morning, ten for each site, each asking for the next available appointment for a nurse visit. Compare what the assistant offers with what the diary shows at the same minute. Count mismatches. Zero mismatches in twenty is a reasonable pass for a pilot; more than one means the integration is not ready, and you fall back to callbacks until it is fixed.

The main worked example: a two-site clinic's first week

Illustrative figures for a medical clinic offering doctor, nurse and physiotherapy appointments, with a main site and a smaller satellite. The clinic receives about 220 calls a week at the main site and 90 at the satellite. About a third of calls at both sites are bookings, a quarter are results or prescription queries, and the rest are directions, opening hours and general questions. Say 12 percent of calls at the satellite are meant for the main site, because patients confuse the numbers. These proportions are assumptions for the example.

The clinic chooses the middle route: two public numbers forwarding to one configuration, with the dialled number identifying the site. Week one goes like this.

1. Before launch. The managers write the two fact sheets and the routing sentences. A clinician writes the urgent instruction. The vendor connects the main calendar; the satellite's diary cannot be read, so satellite bookings are callbacks.
2. Days one and two. The AI answers only during lunch and after hours, with staff handling the rest. Staff listen to every call.
3. Days three to five. The AI answers overflow when staff are busy. A manager reviews 30 calls a day, scoring site accuracy, correct hours, correct hand-off and urgent handling.

By Friday the review finds two types of error. Callers to the satellite asking for physiotherapy are told "yes" because the satellite sheet's "Not here" line was missing, and a Wednesday closure was in a manager's head, not on the sheet. Both are fixed in minutes because they are writing errors, not technical ones. The clinic keeps the AI on overflow and after hours for another fortnight before extending it. That staged approach follows the pattern in measuring an AI receptionist's return in 90 days.

Hand-offs to staff that do not frustrate callers

A hand-off fails when the caller has to repeat everything. Design three kinds.

• Warm transfer. The assistant passes the call to a person along with a short summary shown on screen or sent by message: name, site, reason, what has been tried. The caller hears "I'll put you through to [first name] on reception, who has the details".
• Callback. The assistant takes name, number, site, reason and a preferred time, and creates a task for the right site. The promise it makes must match what staff can deliver, such as "someone will ring you by 2 pm today".
• Message to a clinician. For results, repeats and clinical questions, the assistant logs a message with the patient's identifiers and the question. It says nothing about the answer.

Test each by making a call and reading what the staff member receives. A summary that says "caller wants an appointment" is useless; one that says "caller wants a nurse blood test at North Street, mornings, has fasted, call back on this number" saves a minute and a repeat question. For a bigger picture of when calls should go to a person, see live, AI or hybrid reception.

What the receptionist must not do

A clinic assistant touches sensitive territory, so write the boundaries as firmly as the routing.

• It does not give medical advice, interpret symptoms or reassure a worried caller about a condition.
• It does not read out or discuss test results, prescriptions or notes.
• It tells callers it is an automated assistant when asked, and at the start where your rules or local rules require it.
• It does not take card or bank details by voice unless your payment arrangements specifically allow it.
• It does not confirm that a named person is a patient to a caller who has not been verified.

Decide how it verifies callers before it discusses an appointment: usually a name and date of birth match against the booking system. Check the vendor's data terms for call recordings and transcripts, since both contain health information. If you handle patient information, you need written terms covering storage, retention and training use, and the rules that apply to you may require a specific agreement.

Testing before real callers hear it

Run a set of scripted calls and record the result of each in a table. A minimal set for two sites:

1. Book a routine appointment at each site.
2. Ask for a service that is only at one site while calling the other.
3. Call vaguely: "the clinic near the station" or "wherever has a nurse tomorrow".
4. Call about a site that is closed today.
5. Say something urgent in a calm voice, and again in a rush.
6. Ask to speak to a named clinician at the wrong site.
7. Ask for a result or a repeat prescription.
8. Interrupt, mumble, hang up halfway and call back.
9. Ask in a way the sheet does not cover, such as "can I bring my child's friend".
10. Press zero or say "person" at each step.

Score each on right site, right facts, right hand-off and right tone. Repeat after every change to a fact sheet or a rule. If you cannot get through the list without a failure, you are not ready, and a staged launch with staff listening is the safer route.

Numbers worth tracking per site

Pick four and review them weekly for the first month, then monthly. Calls answered by the AI, by site. Calls transferred to staff, by reason. Bookings completed against bookings requested. Wrong-site corrections, meaning the number of times staff or patients had to redirect a call. If you have two numbers, the per-site split is automatic; if you have one, tag calls by menu choice. A rise in wrong-site corrections is the earliest sign that a fact sheet is stale.

The cost side depends entirely on the vendor and your call volume, so get quotes for the two-site configuration rather than the single-site price, since pricing can differ per number or per location. The tutorial on what an AI receptionist costs gives the shape of the pricing, and you should confirm every figure on the vendor's page. If you want help mapping call types and rules before you talk to vendors, that is what my AI implementation consultation covers.

Callers who are not sure which site they need

A fair share of calls will not name a site at all. A parent rings to book a child's vaccination, a patient wants "the next free nurse", a carer asks where to take someone for a blood test. The assistant needs a way to decide without sounding like an interrogation.

Write a short decision order and give it to the AI as an instruction. First, ask what the visit is for, because services often decide the site. Second, if both sites offer it, ask which is easier to get to, and offer the earliest available slot at each by name. Third, if the caller still has no preference, offer the nearest slot overall and confirm the address aloud. Fourth, whatever the caller picks, repeat the site name, the day and the time before ending the call, and send a text confirmation that includes the address.

The repeat-back is the cheapest wrong-site safeguard you have. Some callers will still end up at the wrong site if you rely on them to remember, and a confirmation text with the address gives them a second chance to notice before they travel. Count wrong-site arrivals for a month with and without the text to see what it does for your clinic. Add a line to that message for step-free access or parking, since those are the questions that usually follow.

Illustrative case: a caller asks for a "quick blood test" and says she works near the satellite. The assistant checks the satellite's sheet, sees blood tests are mornings only, notes it is now afternoon, and offers the satellite tomorrow at 8:30 or the main site this afternoon. She picks tomorrow, hears the address repeated and gets a text. No staff member touched the call.

What to do when a call goes wrong

It will, sooner or later. A patient is sent to the wrong site, or an urgent call is handled badly. Have a one-page response ready. Someone calls the patient the same day, apologises, fixes the booking and logs the incident. A manager listens to the recording, identifies the missing rule or the incorrect fact, and fixes the sheet. If the call involved an urgent symptom, a clinician reviews it and decides whether the scripts or the staged rollout need to change. The tutorial on what to do when an AI receptionist gets a booking wrong has a longer version.

Keep a simple log of every such incident, with date, site, what happened and what changed. After a month it shows whether errors are random or clustered around one site, one service or one rule. Clustering is good news: it means one fix will clear many errors.

Questions about an AI receptionist across two sites

Should two sites share one phone number or keep two?

Keep two if patients already know each site's number and your booking systems are separate. Use one number with a short menu if patients mostly don't care which site and staff can see both diaries. Either way the receptionist needs a separate fact sheet for each site.

What should happen when one site is closed?

The receptionist should say so, offer the other site and check its calendar before promising a time. Urgent calls go to the on-call route or the emergency instruction you have written, never to a voicemail that nobody checks until morning.

Can the AI see both sites' booking calendars?

Only if your booking system exposes both to the tool and the vendor supports the connection. Test it with real slots before you rely on it. If it cannot, have it take a callback request instead of promising a time.

How do I stop it sending patients to the wrong site?

Ask which site first, repeat it back, and give each site its own fact sheet with address, parking, services and hours. Then test with twenty calls where the caller is vague, such as asking for the nearest clinic or the one with the physio.

Further reads

• What Is an AI Receptionist and How Does It Handle Bookings? — How an AI receptionist handles bookings, in plain English.
• Healthcare AI Call Centre: When a Group Practice Should Centralise Calls — When a group practice should centralise its calls.
• AI Receptionist for UK Dental Clinics: NHS, Private and Hand-Offs — Hand-offs and private versus public bookings in a dental setting.
• AI Receptionist for Physiotherapy Clinics: Cost and Booking Setup — Booking and cost for a similar allied-health clinic.
• AI Receptionist or Answering Service for a Booking-Based Business? — Whether a human answering service suits you better.
• What Is the Cheapest Way to Handle Enquiries Out of Hours? — Options for calls outside opening hours.
• Best AI Receptionist for a Medical Practice: What Each Type Does — Booking AI versus transcription-only tools for a small medical practice, compared on one 20-call test with EHR and cost checks.
• Best AI for Healthcare Providers: What to Buy First by Pain Point — A buying order for a small practice: price your missed calls, late notes and denied claims, then pick the phone agent, scribe or claim check.
• Virtual Receptionist for Doctors: A Worked Call Flow and Script Set — A call flow, opening script, emergency wording, hand-off rules and check-call routine for a human or AI virtual receptionist.
• AI Receptionist Pricing for Medical Offices: Per Minute or Booking — Per-minute, per-call and per-caller AI receptionist pricing compared for a medical office, with worked monthly costs, overage rules and BAA gating.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: general practice of call handling and booking-system integration; no vendor-specific pricing used, so confirm features and prices on your chosen vendor's page.