El widget web y Telegram ya están activos. WhatsApp es lo siguiente.
OpusDone
Iniciar sesiónSolicitar acceso
For dental clinics

The 22:10 toothache message gets a real answer.

An AI chatbot for dental clinic enquiries that answers at 22:10.

OpusDone answers dental enquiries at 22:10, in the language the patient wrote in, using only the emergency and price notes you have approved as live. A patient with toothache messages long after the desk is empty, and by half eight they have found a clinic that replied. It offers appointment times read from your real Google Calendar or Cal.com at the moment of writing, so two patients cannot take one slot. Anything clinical goes straight to you.

  • Answers at 22:10, not tomorrow
  • Books into Google Calendar or Cal.com
  • Anything clinical goes to a person

Reviewed

Appointments in the OpusDone bookings screen
One Tuesday night

The 22:10 toothache

The practice shut at six. The message still arrives.

  1. 22:10

    A patient in pain

    She opens the widget on your website because there is nobody left at the desk to ask.

  2. 22:12

    Logistics, not clinical advice

    It explains how an emergency appointment works from the answers you approved, and says nothing about her tooth.

  3. 22:16

    A slot at nine

    Your hours, your closed days and the calendar itself are all checked before the appointment is written.

  4. 22:19

    She writes something worrying

    The message is treated as sensitive, so the alert tells you a sensitive topic needs you and withholds what she wrote.

  5. 07:45

    You open the diary

    The appointment is already in it, and the alert is waiting with one button.

The evening gap

Why do dental enquiries arrive after your clinic closes?

Most of what reaches a clinic after six is not complicated. It is just unanswered.

01

Pain arrives at night

Toothache does not wait for Monday. The message lands late, nobody is there, and the person spends the gap messaging the next three clinics on the map until one of them replies.

02

The same six questions, forever

Are you taking new patients. What does a check-up and clean cost. Do you see children. Where do I park. Every one of them has been answered before, by someone standing up with gloves on.

03

The hole in Thursday afternoon

A cancellation at four leaves a gap nobody fills, because filling it means working through a list while the next patient is already in the chair.

Practice setup, step by step

From your website to your waiting list

Nothing here needs a supplier visit or a change to how the surgery already runs. You can do the whole of it between two patients.

  1. 01

    It reads the practice website you already have

    The crawl looks for your treatments, prices, FAQ, contact and hours pages by English and Spanish keywords, then reads the meta description, Open Graph tags and JSON-LD for your phone number, address and opening hours. One model pass turns that into structured services, FAQs, policies and four to ten draft knowledge entries. It only fills fields that are still empty, so anything someone at the practice has typed can never be overwritten.
  2. 02

    You approve every word a patient can see

    Only entries you mark live ever reach the assistant. Drafts, imports waiting for approval, archived and rejected entries are invisible to it, so a treatment you stopped offering in March simply is not there. A quality gate throws out navigation menus and cookie boilerplate, anything under 80 or over 4,000 characters, and near-duplicates at 85% overlap.
  3. 03

    You set surgery hours, closures and notice

    Working hours, and closed days set by clicking one day or dragging across a run of bank holidays on the month calendar. Minimum notice is 120 minutes by default, slots 30 minutes, granularity 15, buffers zero, horizon 60 days, and all of it is editable. Closures are mirrored into Google as an all-day “Closed — {your business}” event, and deleting that event does not reopen the day.
  4. 04

    You connect one calendar for the practice

    Google Calendar over OAuth with no API key, or Cal.com with one. It reads every calendar you mark as busy and merges them, so a hygienist’s course or your own commitment elsewhere blocks the slot. It books the practice rather than a named dentist, and the scopes are deliberately narrow: it can read and create events, and cannot delete your calendars.
  5. 05

    You test it against a 22:10 toothache

    The sandbox runs your real instructions and your real tools, then stops before it writes anything. No appointment is made, no lead is created, no alert is sent, no escalation fires. Type the message a frightened patient would type at that hour and read what comes back — the sandbox is free and never gated behind a plan.
  6. 06

    One line on your site, then it runs

    The widget is a single script tag on the site you already have, with no iframe and no rebuild, and Telegram is the second live channel; every channel enters the same function, so the answer does not depend on the door the patient came through. Hours and closed days are checked before a slot is offered and again before it is written, and availability is re-read at the moment of writing so two patients cannot take the same slot. It replies in the language the patient wrote in, and the AI disclosure required by Article 50 of the EU AI Act is added once per conversation after the model writes, and cannot be prompted away.

What happens when a patient starts describing a symptom?

It stops. The assistant handles logistics — hours, approved prices, directions, booking, cancellations — and hands anything clinical to a person. Sensitive is one of five escalation reasons and it is locked on: you cannot switch it off, and for a sensitive escalation both the email and the Telegram card withhold the content entirely. You are told only that a sensitive topic needs you.

On the front desk

What can OpusDone do while the clinic is dark?

It offers times you can actually keep

Working hours and every day you have closed are both checked before a single slot is shown, and availability is read again at the moment of writing, so two patients cannot take the same time. At most six options, and at most two from any one day, so nobody is reading a wall of times on a phone.

Google Calendar or Cal.com. No other provider connects.

Appointments in the OpusDone bookings screen

The questions it takes off your desk

Five it will answer today, from knowledge you approved: “Are you taking on new patients?” · “How much is a check-up and hygienist clean?” · “Do you see children, and from what age?” · “Is there parking near the clinic?” · “Can I pay for treatment in instalments?”.

A customer conversation in the OpusDone inbox

You open the dashboard to a short list, not a transcript

Each person becomes one lead with a plain score out of 100 — a phone number is worth 40, an email 40, a booking 40 — and you are alerted at 50, because one way to reach somebody is enough. People who showed real intent and left no contact go on their own list so you can see what you nearly had.

Filter, search, sort, export to CSV.

A scored lead in the OpusDone leads workspace
The settings behind the diary

Chosen once, then left alone

120 minminimum notice before slots
6options offered, never more
3closure checks before writing
60 dayshow far ahead it books
Guardrails

What will OpusDone never do in your clinic?

The limits are the product. A front desk that guesses is worse than one that is closed.

It will not give clinical advice

No triage, no opinion on whether that pain is an abscess, no “it sounds like”. It says a person will help and fetches one.

It will not invent a price or a waiting time

If the fee for a crown is not in the knowledge you marked live, it says it does not have that and passes you the enquiry. Drafts, website imports awaiting approval and archived entries are invisible to it.

It will not look at a photo of a tooth

There is no image or document understanding. An attachment it cannot read skips the model entirely: it says so plainly and brings in a person.

It will not pass itself off as your receptionist

The AI disclosure required by EU AI Act Article 50 is added after the model writes, once per conversation, in the patient’s language, and it cannot be prompted away.

Dental clinic AI chatbot questions, answered

Can it book a patient with a particular dentist?

No. There is one calendar for the business, not one per clinician, and it does not route by practitioner. It writes the appointment into that calendar and you assign it internally the way you always have.

A patient writes in Catalan, or German. What does it reply in?

It replies in the language the patient wrote in, and only falls back to your default when it genuinely cannot tell. The Article 50 disclosure appears in their language too. You write your knowledge once, in whatever language suits you.

How do we stop it saying something the practice has not agreed?

Only knowledge you approved as live ever reaches it. Setup reads your existing website — prices, services, FAQ, contact and hours — and turns it into drafts for you to approve; it only fills fields that are empty and can never overwrite what a human typed. The sandbox runs your real instructions and your real tools and stops before writing anything, so you can push it hard before a patient ever sees it.

What happens when a patient writes something distressing at midnight?

You get an in-app bell always and an email always — escalation email is never held by quiet hours, which default to 22:00–08:00 and hold only the phone-noisy channels. Telegram is included free if you connect it. For a sensitive escalation both the email and the Telegram card withhold the content entirely: you are told only that a sensitive topic needs a person, so nothing about that patient sits in your inbox waiting to be read. Alerts are capped at one per patient per 15 minutes.

We keep appointments in our practice software. Does this still work?

Booking connects to Google Calendar or Cal.com and to nothing else, so the diary it writes into has to be one of those. Clinics usually keep a single business calendar there for online bookings and carry them across at the desk. If that is not workable for you, the assistant can still answer and hand every booking request to a person.

Someone in pain asks to speak to a human at eleven at night. What happens?

They are told a person is taking over, and you get an in-app bell and an email immediately — the email is never held by quiet hours. If you have connected Telegram you also get a card with the last five messages and a Take over button, and taking over pauses the assistant in two taps. Quiet hours default to 22:00–08:00 and hold only the phone-noisy channels.

Could it end up giving clinical advice?

It answers from the entries you marked live and from nothing else, so it holds no clinical knowledge you did not approve. Symptoms, pain and anything else read as sensitive trigger an escalation, and the sensitive reason is locked on and cannot be switched off. If a patient asks for a person and the model does not escalate, the runtime escalates anyway. A photo of a tooth is accepted up to 25 MB, but the assistant confirms it arrived, says plainly that it cannot open it, and hands the conversation to you — there is no image understanding, and a file always creates an escalation.

At 22:10 it answers instead of collecting emails

Website widget and Telegram, on Google Calendar or Cal.com. It handles logistics and hands anything clinical to a person. Every captured lead is written to a Google Sheet in your own Drive within a minute, free on every plan — and the conversation summary is off by default, so you can send a patient’s contact details without ever transmitting what they said.

Solicitar acceso
AI Chatbot for Dental Clinic Front Desks | OpusDone