What an AI chatbot should (and shouldn't) do for a clinic
What an AI chatbot for clinics should do, where it must stop, and the questions to ask a vendor before you put one on your website.
Table of contents
Patients don't keep office hours. They scroll Instagram at midnight, see a treatment, and want to know three things right away: how much, is it right for me, and when can I come in. If nobody answers until 10am, many of them move on.
An AI chatbot on your website can answer some of those questions at any hour. It can also cause real problems if it guesses prices, gives medical opinions or books patients into slots that don't exist. This article sets out what an AI chatbot for clinics should do, what it should never do, and the questions to ask before you choose one.
Answer from your clinic's information, not the internet
The single most important rule: the chatbot should answer from your clinic's own information.
General-purpose AI models know a lot about the world, including a lot of things that aren't true for your clinic. Ask one "how much is laser hair removal?" and it may produce a confident average price from nowhere. That's worse than no answer, because the patient arrives expecting that price.
A clinic chatbot should draw on sources you control:
- Your treatments, with the prices, durations and downtime you've published.
- Opening hours, address, parking and arrival instructions.
- Your practitioners and which treatments each one performs.
- Current offers, with their start and end dates.
- Your FAQs and cancellation policy.
- Documents you upload, such as price lists, aftercare sheets and brochures.
When your information changes, the chatbot's answers should change with it. If a vendor can't explain where answers come from, or how quickly an updated price reaches the chatbot, keep asking.
Test it yourself. Ask about a treatment you don't offer. A well-behaved assistant should say you don't offer it, not invent a price.
Show live availability, but let the patient book
"Do you have anything for a HydraFacial on Saturday?" is one of the most useful questions a chatbot can answer, and one of the easiest to get wrong.
What it should do: check your real schedule and reply with actual free times, for example "Saturday has 11:00, 14:30 and 16:00 with any available practitioner", plus a link to book one. To do that properly it needs to know today's date and your clinic's time zone, so "tomorrow" and "this weekend" mean what the patient means. That matters for clinics in the US, where daylight saving time moves the clocks, and for patients messaging from a different city.
What it shouldn't do: book the appointment on its own. Chat messages are easy to misread. A patient who types "maybe 2:30 sat?" hasn't agreed to anything yet. Sending them to the booking page lets them check the treatment, practitioner, date and time, enter their details and confirm themselves.
Know where to stop: medical questions and emergencies
A clinic chatbot isn't a clinician. It should be built to stay in its lane, and you should test that it does.
Questions the chatbot should hand to your team rather than answer itself:
- Pregnancy and breastfeeding. "I'm pregnant, can I have Botox?" The chatbot should not say yes or no. It should suggest speaking to the clinic directly.
- Medication and medical conditions. "I take blood thinners, is that OK for microneedling?" Same: refer to a practitioner.
- Symptoms or reactions after treatment. "My skin is blistering after my peel." The chatbot should urge the patient to contact the clinic now and, if it sounds serious, local emergency services.
- Emergencies. Anything describing severe pain, breathing problems, chest pain, heavy bleeding or similar should get an immediate, plain message to contact emergency services, not a chat about opening hours.
- "Is this treatment right for me?" The chatbot can share general information your clinic has written and who a treatment isn't suitable for, then recommend a consultation.
Write these boundaries into your chatbot setup and check the wording of its refusals. They should be kind and useful ("That's a question for one of our practitioners. Would you like us to call you?"), not cold.
Hand over to a human and learn
No chatbot will know everything. What matters is what happens next.
Offer a clear handover. When the chatbot can't answer, or the patient asks for a person, it should offer a callback request or contact option and collect a name, phone number and a short note. That enquiry should land somewhere your team actually checks, with a reply-time promise you keep. See our article on clinic lead response time for why speed matters here.
Log unanswered questions. Every "I don't know" is a signal. If ten patients a month ask whether you offer evening appointments, that belongs on your FAQ page and in the chatbot's knowledge. Review the log regularly and add answers.
Read real conversations. Spend a few minutes a week skimming chats. You'll spot confusing wording on your website, treatments patients ask about that you don't offer, and places the chatbot could be friendlier.
Protect it from abuse
A public chatbot is open to anyone, including people trying to misuse it.
- Rate limits stop one visitor from sending hundreds of messages a minute.
- A daily limit caps the total, so a spike in misuse can't run up large AI costs.
- An off switch you control lets you disable the chatbot instantly if something goes wrong.
- Topic limits keep it focused on your clinic, so it doesn't write essays or discuss unrelated subjects when prompted.
Ask how each of these works before you sign up, and who pays if usage spikes.
Think about privacy from day one
Patients sometimes share more in a chat than they would on a form: health conditions, photos, medication. Plan for that.
- Collect only what you need. For a callback, that's usually a name, a phone number and a short note. Don't ask for medical history in chat.
- Tell patients what happens to their messages. Your privacy policy should cover the chatbot: what's stored, for how long, and who can see it.
- Limit who can read conversations. Not every staff member needs access.
- Ask where data goes. Which AI provider processes the messages? Are they used to train other models? Where is data stored?
- Separate marketing consent. Answering a question isn't permission to send promotions later.
For US practices, ask whether the product is designed to handle protected health information and, if your practice needs a HIPAA Business Associate Agreement, discuss that with the vendor before launch. In Pakistan and elsewhere, check the privacy rules that apply to your clinic.
Questions to ask a chatbot vendor
Use this list in your next demo. Vague answers are a warning sign.
| Question | What a good answer sounds like |
|---|---|
| Where do answers come from? | Your clinic's treatments, prices, FAQs and uploaded documents, not general internet knowledge. |
| How fast do changes show up? | Updates to prices or hours are reflected without a rebuild or a support ticket. |
| Can it check real availability? | Yes, from your actual schedule, time zone included. |
| Does it book by itself? | No, it links to booking where the patient confirms. |
| What does it do with medical questions? | Refers them to the clinic or emergency services; no diagnosis or treatment advice. |
| What happens when it doesn't know? | Says so, offers a callback, and logs the question for you. |
| How is misuse limited? | Rate limits, a daily cap and an off switch you control. |
| Where are conversations stored, and who can see them? | A clear answer, with role-based access for your team. |
| Can we test it before patients see it? | Yes, with your own information loaded. |
Then test it. Ask awkward questions: a treatment you don't offer, a price that changed last week, a pregnancy question, a fully booked Saturday. Watch what it does.
How the Velvet Lead assistant works
The rules above are the ones we built our own assistant around. It answers 24/7 from each clinic's own treatments, prices, hours, practitioners, offers, FAQs, cancellation policy and uploaded documents. It checks live availability and links to booking, but never books on its own. It gives no medical advice, points pregnancy, medication and emergency questions to the clinic or emergency services, and offers a callback request when it doesn't know. Unanswered questions are logged, rate and daily limits protect it, and the clinic can switch it off at any time.
You can read more about the AI assistant for clinic websites, or see how it answers typical questions for med spas and dermatology clinics. Whichever tool you choose, hold it to the same standard: your information, clear limits, and a human one tap away.
Frequently asked questions
Can an AI chatbot give patients medical advice?
It shouldn't. A clinic chatbot should answer practical questions from your own information, such as prices, hours and aftercare you have written, and pass anything about symptoms, pregnancy, medication or emergencies to your team or emergency services.
Should a clinic chatbot book appointments by itself?
It's safer for the chatbot to show real free times and link to your booking page, where the patient checks the details and confirms. That keeps the patient in control and avoids bookings made from a misunderstood message.
What happens when the chatbot doesn't know the answer?
A good assistant says so plainly, offers a callback or contact option, and logs the question so your team can add an answer. Guessing is the behavior to avoid.
Is it safe to put an AI chatbot on a medical or aesthetic clinic website?
It can be, if it is limited to your clinic's own information, refuses medical advice, protects patient details and can be switched off. Ask any vendor how they handle each of these before you launch.
Clinic websites, booking and lead management
The Velvet Lead team builds and launches patient-booking websites for clinics in the USA and Pakistan. We write about what we see every day: how patients enquire, why bookings get lost, and the practical fixes that help clinics reply faster and keep their diaries full.