Short answer: answer insurance questions online by publishing what is true for everyone — which insurers you work with, whether you bill directly, what you need from the patient, what you charge self-pay — and let an AI chat widget state those facts instantly. Never let any automated system confirm an individual's coverage, benefits or out-of-pocket cost. Those depend on the patient's specific plan, and getting them wrong is worse than not answering.
Why this one question is so expensive
Do you take my insurance is probably the most-asked question in outpatient healthcare, and it is the worst one to handle by phone. It is short, it feels like it should have a yes or no answer, and it almost never does.
Reception picks up. The patient names an insurer. Reception knows the clinic works with that insurer but not whether this particular plan covers this particular service, whether the patient has met their deductible, whether pre-authorisation is needed, or whether the specific clinician is in network. So the honest answer is a paragraph of qualifications, and the call runs five minutes. Multiply that across a practice and it is one of the largest single consumers of front desk time in any clinic.
The second cost is worse. When a staff member gives a confident answer that turns out to be wrong, the patient arrives expecting to pay nothing and leaves with a bill. That is a complaint, a refund conversation and a lost patient.
Separate the two kinds of insurance question
Almost every insurance enquiry is one of two types, and the whole approach depends on telling them apart.
Clinic-level facts are true regardless of who is asking. Which insurers you have agreements with. Whether you bill the insurer directly or the patient claims reimbursement. Which of your clinicians are in network with which plans. What documents you need. What you charge patients paying themselves. Whether you provide itemised receipts and procedure codes.
Patient-level facts depend entirely on someone's individual policy. Is this service covered. How much of the deductible remains. What the copay will be. How many sessions are allowed this year. Will the claim be paid.
Clinic-level facts should be published on your website and stated instantly by your chat widget. Patient-level facts should never be answered by an automated system, and honestly should be answered carefully by staff too — the reliable source is the patient's insurer.
What a good online answer sounds like
The pattern that works has three parts: the clinic-level fact, the honest limit, and the next step.
Something like: yes, we work with that insurer and we bill them directly for most treatments. What your plan covers and what you would pay depends on your individual policy, so we would recommend confirming with your insurer — and it is worth asking about your deductible and whether pre-authorisation is needed. If you would like, I can take your details and we will get you booked in and send you the procedure codes you will need.
That answer is genuinely useful, it does not overpromise, and it ends with a captured booking request rather than a dead end. That is the whole design goal.
Making a chat widget do it reliably
An AI chatbot is a good fit for this because the clinic-level layer is stable and the boundary is easy to define.
Train it on your own content. Clerkzo crawls your website, so if you have an insurance page listing your accepted providers and your billing approach, that becomes the source. If you do not have that page, write it — it is worth doing regardless of whether you run a chatbot, and it is the single highest-value page most clinic sites are missing.
Then upload the rest. A knowledge file with your current insurer list, which clinicians are in network where, your self-pay prices, your documentation requirements and your billing process covers what the website does not say. Update it when your contracts change, because a stale insurer list is the fastest way to generate an angry patient.
Then set the guardrails. The assistant must not confirm coverage, estimate a patient's cost, quote a copay or deductible, or predict whether a claim will be paid. When asked, it states the clinic-level fact, explains that the specifics depend on their plan, and offers the booking request. Clerkzo's guardrails plus human handoff make that the default, and the corrections loop lets you tighten any answer that comes out too confident.
Test it deliberately before launch. Ask it about an insurer you do not take. Ask it whether a specific procedure is covered. Ask it what you will pay. If any answer sounds more certain than it should, fix it.
Capture, then follow up
Because insurance questions arrive right at the point of decision, they are high-intent. The person asking is close to booking. So the conversation should end in a captured request, not a suggestion to phone during opening hours.
Clerkzo can collect this conversationally or as an in-chat form with custom fields: name, phone, insurer, policy number if you want it, service requested, and preferred times. Requests land in a leads inbox with a New to Contacted to Qualified to Won or Lost pipeline, with email, webhook or Telegram notifications.
Note that it captures the booking request and hands it to your team rather than writing into your calendar. That is correct here — an insurance-dependent appointment often needs verification before it is confirmed, and you want a human to place it. This is the same model described on appointment booking.
Where clinics see the difference
Two places. First, evenings and weekends: insurance questions are a classic out-of-hours enquiry because people check their benefits when they get home. Answered instantly, they convert; left in a form queue, they cool. That is the core of after-hours answering.
Second, the front desk. The clinic-level questions stop reaching the phone, so the calls that do come through are the ones needing a person. Practices across dental and physiotherapy report the same shape of change: less repetition, more real work.
A note on data and compliance
You are handling patient contact details and possibly policy information, so be deliberate. Collect the minimum you need for the callback, do not ask for clinical detail, and set your retention policy consciously. Clerkzo is SOC 2 Type II examined, GDPR compliant and HIPAA assessed.
Frequently asked questions
Can a chatbot check someone's benefits?
No, and it should not attempt to. It states which insurers you work with and directs benefit questions to the insurer, while capturing the booking request.
What if we take an insurer but not for every service?
Say so explicitly in your knowledge file. The assistant should state the exception rather than giving a blanket yes.
How do we keep the insurer list current?
Update the knowledge file whenever contracts change and review the transcripts monthly for insurers people ask about that you do not list.
Should we publish self-pay prices too?
Yes. Many enquiries end with the patient choosing to self-pay, and a published price converts far better than an invitation to call for a quote.