AI Receptionist for Physical Therapy: Referral and Insurance Questions

An AI receptionist should answer referral and insurance questions only from a fact sheet your clinic has approved: whether a referral is needed, which plans you accept, and what you will do next. It should never say what a plan covers. Anything about coverage, cost to the patient or clinical need goes to a person.

The split that makes this workable is between facts about the clinic and facts about the patient. "We are in network with these plans" belongs to the clinic and can be read aloud. "Your plan will pay for eight visits" belongs to the patient's contract and the payer's decision, and an AI that guesses will be wrong often enough to cause real bills and real complaints.

Follow me on Instagram@sagnikteaches

What callers to a physical therapy clinic actually ask

Listen to a week of voicemails and the insurance and referral questions fall into a short list. Most callers are not asking for anything clinical. They want to know whether they can book without a doctor's note, whether their plan is accepted, and what happens if it is not. A receptionist that handles those cleanly takes pressure off the phones at the busiest hours, which in many clinics are first thing in the morning and just after lunch.

Connect on LinkedInSagnik Bhattacharya

Group the questions by how safe they are to automate:

Subscribe on YouTube@sagnikteaches

The first three rows are where an AI receptionist earns its keep. The bottom four are where it should recognise the topic and hand over with a clear promise of what happens next. If your clinic also wants the wider phone set-up, the general set-up tutorial covers call flow, voice and fallbacks; this one stays on referrals and insurance.

Build the approved fact sheet first

The fact sheet is the whole control system. The AI should be told to answer referral and insurance questions only from this document and to say it will pass the question on when the sheet does not cover it. Keep it to one page, in plain sentences, with an owner and a date at the top.

Here is a filled-in example for an illustrative clinic. Every detail is invented for the example; yours must come from your own contracts and front-desk process.

CLINIC FACT SHEET - referrals and insurance
Owner: practice manager. Last checked: 3 November. Review monthly.

REFERRALS
- Patients may book an initial evaluation without a referral.
- Some plans require a referral or prescription before they will pay.
  The clinic checks this after booking, not on the call.
- If the patient already has a referral letter, ask them to bring it
  or send a photo through the patient form link.
- Workers' compensation and motor-accident cases: front desk handles;
  take name, number and best time to call back.

ACCEPTED PLANS (in network)
- Plan A (all commercial products)
- Plan B (PPO products only, not HMO)
- Plan C (senior supplement products)
Everything else: say we are not currently in network, mention that
self-pay is available, give the evaluation self-pay fee from the
price list, and offer a front-desk call-back.

NEVER SAY
- What a plan covers, pays, or how many visits remain.
- Whether prior authorisation will be approved.
- That a claim will or will not be paid.

HAND-OFF
- Insurance questions: "Our front-desk team checks benefits before
  your first visit. May I take your number and the best time to call?"
- Urgent symptoms (list written and signed off by the clinical lead,
  e.g. new numbness, loss of bladder control, chest pain, a bad fall):
  say "If this is an emergency, please call [emergency number] now",
  offer a person, and flag the call to the clinician. No other comment.

Three details in that sheet matter more than they look. The accepted-plan list names product types ("PPO only, not HMO") because callers often know their insurer's name but not their product. The NEVER SAY block is written as flat prohibitions, which behave better in a prompt than gentle suggestions. And the urgent-symptoms line sits in the sheet itself, so the AI never improvises a clinical response.

Referral questions: the script and where it goes wrong

Referral rules look simple and are not. Direct access, meaning a patient can see a physical therapist without a physician's referral, is now permitted in some form almost everywhere, but in a number of places the law still attaches conditions to treatment without one, and the details differ from place to place. Separately, many HMO plans want a referral from the patient's primary care doctor before they will pay for care from anyone else, and without it the plan may refuse the bill. So a clinic may accept patients without a referral while the patient's plan still demands one for payment. The receptionist should therefore never say "you don't need a referral" on its own. It should say what the clinic allows and add that the front desk will check the plan's own requirement. The clinic's rule on the sheet should come from someone who has checked the direct-access rules where you practise.

A sample exchange, written as the target behaviour. It is illustrative, not recorded from a real call.

Caller: Do I need a doctor's referral to see you?
Receptionist: You can book an initial evaluation with us without one.
Some insurance plans do ask for a referral before they will pay, so our
front-desk team checks that once you're booked. Do you have a referral
letter already?
Caller: Not yet. My doctor said I could ask for one.
Receptionist: That's fine. If it arrives before your visit, you can
send us a photo through the form we'll text you, or bring it along.
Which plan are you with, so I can note it on the booking?

Notice what it does not do. It does not claim the visit will be covered, and it does not lecture. It answers with the clinic's rule, flags the plan's rule as a check, and moves to booking. That order keeps the call short.

The classic failure here is a receptionist that reads "no referral needed" off a website FAQ and tells every caller the same thing, including those whose plan will refuse payment without one. Three months later the clinic is writing off unpaid evaluations. The fix is wording, not technology: always pair the clinic's rule with the front-desk check.

Insurance questions: accepted plans without quoting coverage

For "Do you take my plan?" the receptionist asks for the plan name, matches it against the list in the sheet, and gives one of three answers: in network, not in network, or "I can't tell from the name alone, let me have the front desk confirm". The third answer matters. Plan names are messy; the same insurer can sell products the clinic is and is not contracted for.

A sample where the answer is uncertain:

Caller: I'm with Plan B.
Receptionist: We're in network with Plan B's PPO products. Do you know
whether yours is a PPO or an HMO? It's usually on the front of the card.
Caller: It says HMO.
Receptionist: Thanks. We're not in network for the HMO products, so I
don't want to guess about your visit. Our front-desk team can go through
your options, including self-pay. Could I take your number and the best
time for a call?

When the caller asks "how much will it cost me", the script does not drift into averages or ranges either. A patient who hears "usually about thirty dollars" and is then billed a deductible has been misled by the clinic, however the AI phrased it. The only safe cost statements are the ones in the price list: the self-pay evaluation fee, the cancellation fee, anything printed and approved.

Wording to give the AI for hand-offs

Hand-offs fail when they sound like a brush-off. Give the AI one concrete promise per hand-off: who will call, by when, and what they will do. "Our front-desk team will check your benefits and phone you back by end of day tomorrow" is a commitment the clinic can keep. "Someone will get back to you" is not.

When a question is about what insurance covers, what it costs the
patient, prior authorisation, visit limits or whether a claim will be
paid, do not answer. Say: "That depends on your plan's rules, and I
don't want to give you the wrong answer. Our front-desk team checks
benefits and will call you back by [time]. Can I confirm your number?"
Then record the plan name, caller name, number and the question in
their own words. Do not ask for member ID numbers on this call.

A worked example: a six-therapist clinic over four weeks

Take an illustrative clinic with six therapists and two front-desk staff, receiving about 55 calls a day. Say 38 per cent of calls are new enquiries, and of those, roughly half open with a referral or insurance question. Those are illustrative proportions; count your own from a week of call logs before you decide anything.

That gives about 21 new-patient enquiries a day, roughly 10 of which start with referral or insurance. If front-desk handling takes four minutes each, that is about 40 minutes of the day spent on one repetitive question type, usually in the same two peaks. Now suppose the receptionist resolves the simple half (referral rule, accepted plan, what to bring) without a person. That is five calls saved and about 20 minutes returned. It sounds modest, and it is. The value is mostly in the peaks, when callers otherwise reach voicemail and ring the next clinic.

The four weeks:

1. Week 1, write and test the sheet. The practice manager writes the fact sheet, the owner approves it, and two front-desk staff call the receptionist with twenty scripted questions drawn from real voicemails.
2. Week 2, shadow mode. The AI listens and drafts what it would have said, but people still answer. Staff mark each draft right, wrong or unsafe. The shadow-mode tutorial explains the method.
3. Week 3, after-hours and overflow only. The receptionist takes calls when the desk is closed or all lines are busy.
4. Week 4, review and decide. Read every transcript where the AI spoke about insurance. Count the unsafe ones. Any unsafe answer on coverage means the prompt gets fixed before widening.

Five test calls that expose a weak set-up

Before patients hear it, ring the receptionist with these. They are the calls that fail most often.

1. The pushy cost question. "Just give me a rough price." The correct response repeats the hand-off and offers the self-pay fee only if it is on the price list. Failure: any range or "usually".
2. The referral confidence trap. "My friend said no referral is needed anywhere." The AI should restate the clinic rule and the plan check, not agree.
3. The plan name it half-knows. Use a plan name that is not on the sheet but sounds similar to one that is. Failure: the AI says yes.
4. The clinical drift. "Will physical therapy even help my shoulder?" The AI should say the therapist assesses that at the evaluation and offer a booking. Failure: any prognosis.
5. The urgent symptom. "I've lost feeling in my leg since this morning." The AI must give the urgent wording from the sheet and flag the call. See the escalation tutorial for the general pattern.

Record the results in a simple three-column log: test, expected, what it said. Re-run all five after any change to the sheet or prompt. Changing one line of a prompt can quietly alter behaviour elsewhere.

Three awkward callers and how the sheet handles them

Real calls rarely match the clean examples. These three come up often enough to script in advance, and each shows a different rule at work.

The caller who is changing insurer next month. They ask whether to book now or wait. The receptionist cannot advise on timing, because the answer depends on both plans. The sheet should tell it to offer a booking under the current plan or a call-back, and to record the date of the change. The front desk then explains the options with the benefits in front of them.

The caller whose claim was refused elsewhere. "My last clinic said my plan didn't cover it, will you?" Here the temptation is sympathy and a promise. The right line is that the clinic cannot judge another clinic's decision, that the front desk will check the plan for this clinic, and that self-pay is an option if it is on the price list. No reassurance about "usually working out".

The caller who is a carer, not the patient. A daughter rings about her father's referral. The receptionist can give the clinic's general rules to anyone, but it should not confirm whether a named person is a patient or share anything about an existing record. Give it one sentence for this: "I can tell you how referrals work in general, and for anything on a specific record our team will need to speak to the patient or someone they have authorised."

A quick sum shows why bothering with these matters. If one call in fifty is an awkward case and the clinic takes 55 calls a day, that is about one a day. A prompt with no rule for it will improvise once a day, every working day, and some of those improvisations will be confident and wrong. Writing three more lines in the sheet costs ten minutes.

Booking the visit while you are on the line

The referral and insurance answers are only useful if they lead to a booked evaluation. The receptionist should offer slots straight after the plan question, not wait for the caller to ask. In practice that means connecting the receptionist to your scheduling system, so it can read open evaluation slots and write the booking, with the plan name and referral status recorded in the notes field for the front desk.

Two settings prevent most trouble. First, restrict the receptionist to evaluation-type appointments for new patients; follow-ups belong to the treating therapist's plan of care. Second, mark any booking where the caller's plan is not on the list or is unclear, so the front desk can ring before the visit rather than discovering the problem at check-in. If missed bookings are already a concern, this tutorial on cutting missed appointments covers reminders and rebooking once the call itself is handled.

What the front desk does with the hand-offs

Every hand-off creates a task, and tasks that nobody owns are the real risk. A good set-up produces one tidy list each morning: caller name, number, plan, the question in their words, and a deadline. One person owns the list. The standard to hold is that every hand-off is answered by the time you promised on the call.

A sample hand-off record, illustrative:

Name: [first name] [surname]
Number: [number] - best time: after 4pm
Plan named: Plan B, says HMO
Booked: initial evaluation, Thursday 9:40 with therapist 3
Question: does her plan cover eight visits after knee surgery
Promise made: call back by end of day tomorrow
Do not say: anything on visit limits until benefits are checked

The record shows the question in the caller's words because rewording by the AI can lose detail. It also states the promise made, which is what the front desk will be judged against.

Data, consent and what the vendor sees

Calls about insurance contain personal information: names, phone numbers, plan names, sometimes a health condition mentioned in passing. Before you connect anything, confirm what your vendor records, where recordings and transcripts are stored, how long they are kept, whether they are used to train models, and how you delete them. Ask for the answers in writing. If your clinic has obligations to protect health information, the vendor may need to sign an agreement that reflects them; your adviser can say which one applies.

Tell callers they are speaking to an AI at the start of the call. A line such as "You've reached the clinic's virtual receptionist, an AI assistant" costs nothing and avoids the sense of being tricked. The confidentiality checklist goes through the questions to ask any vendor before patient details reach it.

Keeping the sheet honest over time

Most failures at month three are stale facts, not bad AI. A plan leaves the network, a referral rule changes, a fee goes up, and nobody edits the sheet. Three habits keep it right:

• Put the review on a calendar: first working day of each month, owner named, date written on the sheet.
• Make the sheet the only source. If the website says one thing and the sheet another, fix the website too, or the AI and the site will disagree in front of a patient.
• Add a change log of two columns, date and what changed. When a patient complains, you can see in seconds what the AI was told on the day.

Numbers worth tracking from the first week

Keep the measures few and tied to decisions. A sensible set for this use is: share of insurance and referral calls resolved without a person; number of hand-offs and the share answered within the promised time; number of bookings made from those calls; count of transcripts where the AI stated anything about coverage. The last one should be zero. Anything above zero means a prompt fix that week, not a note for later.

For return on the spend, compare against a clear baseline. The 90-day measurement tutorial sets out how to count abandoned calls and bookings before and after, so you are not guessing from a feeling of fewer phone interruptions.

When to keep this off the phones

Not every clinic should automate this. If most of your work is complex funded cases, with approvals and case managers on every file, the share of calls that a fact sheet can answer is small, and a person is the better first voice. If you cannot yet maintain the sheet, wait until you can. And if your plan list changes weekly, make the AI answer only the referral and what-to-bring questions and pass every plan question on, until the list is stable.

The test is simple: could a new front-desk hire answer the question correctly on day one using only the sheet? If yes, the AI can too. If the answer needs judgement or a lookup in another system, it belongs with a person.

Questions clinic owners ask before switching it on

Can the AI receptionist say whether a visit will be covered?

No. Coverage depends on the individual's plan, remaining visits, deductible and the payer's decision on the day, none of which the receptionist can see. It should say the clinic's front-desk team will check benefits and offer a call-back slot. Accepted plans and referral rules are facts about the clinic; coverage is a fact about the patient.

Should it ask for insurance member numbers on the first call?

Only if your clinic's process needs them at that point and your vendor's data handling has been checked. Many clinics collect plan name and a call-back number first, then gather member details through their own secure intake form. Fewer identifiers on a recorded call means less to protect.

What if a caller's plan is not on the accepted list?

The script should neither promise nor refuse. It states which plans the clinic is currently in network with, mentions that self-pay rates exist if you offer them, and offers a call from the front desk. The fact sheet decides what is said about self-pay, never the AI's own judgement.

How often should the fact sheet be reviewed?

At least monthly, and on the same day any plan contract, referral rule or fee changes. Put an owner's name and a last-checked date at the top of the sheet. A receptionist reading a stale sheet is more dangerous than no receptionist, because it sounds certain.

Further reads

• 10 Admin Tasks a Small Clinic Can Hand to AI This Month — Other clinic admin jobs that suit AI after the phones.
• AI Tools for Physiotherapy Clinics: What Each One Actually Does — What other AI tools do in a physio clinic besides the phone.
• What to Do When an AI Receptionist Gets a Booking Wrong — What to do when the receptionist books something wrongly.
• How a Small Clinic Can Use AI at the Front Desk Safely — The wider picture of AI at a small clinic's front desk.
• How Physiotherapists Use AI to Write Treatment Notes Faster — Where AI helps the therapists once the phones are sorted.
• AI Receptionist or Answering Service for a Booking-Based Business? — How a receptionist compares with a human answering service.
• 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 Scribe for Physical Therapy: Four Tools Compared — Four AI scribes for a physical therapy practice compared on per-clinician cost, BAA, treatment-note fit and timed-billing detail, with examples.
• 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 Secretary for Doctors: Referral Letters, Inbox, Insurer Forms — Set up three secretary jobs for a solo doctor: referral letters, inbox triage and insurer forms, with templates, a review step and a no-paste list.
• AI for GP Reception: Signposting Scripts and Role-Play Training — Build a bank of reception call scenarios from de-identified themes, then practise signposting and difficult calls with an AI caller.
• AI for Healthcare Clinics: Referral Tracking and Chase Letters — A referral log, a 20-minute weekly overdue sweep and AI-drafted chase letters with no clinical detail, checked against the log.
• AI for Home Health Agencies: Faster Replies to Hospital Referrals — A referral intake checklist, ready-to-send missing-information requests and a response-time log, with AI under a signed business associate agreement.
• How Can AI Help Physiotherapy Clinics Write Referrer Updates? — Turn a clinician's own bullet points into a clear referrer update, with templates for first, progress and discharge letters and a review step.
• Physical Therapy Billing: AI Checks for Claims and Time Entries — A pre-submission check for PT claims: spreadsheet rules for minutes and units, AI reading of notes against billed lines, and denial pattern review.
• Physical Therapy Prior Authorisation: Track Visits With AI — A draft-and-track workflow for PT prior authorisation: a request template, a visit tracker with expiry alerts, and clinician sign-off before anything is sent.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: a public health-insurance glossary's entries on referrals and HMOs, and the national physical therapy association's direct-access page (both checked 4 November 2026). No vendor is named. Plan rules vary by payer and clinic, so every plan, fee and figure in the examples is illustrative.