Dental Chatbot for FAQs: Answers From an Approved Fact Sheet

Write a one-page fact sheet the principal dentist signs off (fees, finance terms, opening hours, policies), load only that into a document-grounded chatbot such as Chatbase, Botpress or Tidio's Lyro, instruct it to answer from that sheet alone, and send every symptom, pain or medication question to a fixed emergency message with your phone number.

The risk is not the fee questions; it is the bot trying to be helpful about a toothache. A general model will happily suggest painkillers or guess at an abscess. The fact sheet keeps the answers consistent, and the refusal rule keeps the bot out of clinical territory, so reception stops repeating parking and price answers without the practice taking on a new risk.

Follow me on Instagram@sagnikteaches

Why a dental bot needs one approved source, not your whole website

Most dental website widgets fail in one of two ways. Some are rigid button menus that cannot cope with a question typed in plain words. Others connect a general language model to the website, or to nothing at all, and let it fill gaps from its own training. The second kind is the dangerous one, because it sounds confident about things nobody at the practice wrote.

Connect on LinkedInSagnik Bhattacharya

Take a visitor who types: "My bottom molar has ached since yesterday, should I take ibuprofen or aspirin?" An unconstrained model tries to answer. It may compare painkillers, mention doses or speculate about the cause. None of that has been checked by a dentist, none of it accounts for the patient's medical history, and the patient may now feel they have had advice when they have not been seen.

Subscribe on YouTube@sagnikteaches

A document-grounded bot works differently. The platform searches your uploaded sheet for the passage that matches the question and the model writes its reply from that passage (this is usually called retrieval-augmented generation). If the sheet holds fees, payment plans, cancellation windows and hours, those are the only things it can state. If a question is clinical, there is nothing to retrieve, and an explicit rule tells the bot what to say instead.

The second failure is contradictory numbers. A practice site can carry an old examination fee on one page, a whitening offer from three years ago in a forgotten PDF and a finance banner with outdated terms. A bot that crawls the whole domain will quote whichever it finds. Chatbase, for instance, can crawl a site and lets you exclude paths; for a dental bot, the safer default is to upload the fact sheet and skip the crawl entirely until the website is cleaned up.

Drafting the fact sheet: what goes in and how to lay it out

Treat the fact sheet as an administrative document, not a brochure. Each section covers one topic, every fee is a figure, and every policy is a sentence the practice is willing to have quoted back to it. Use a plain-text or Word file with headings: Chatbase's upload accepts PDF, TXT, DOC and DOCX files up to 20 MB, as long as the text is selectable, and it also takes pasted text snippets.

Five sections cover most enquiries:

• Hours and access: opening and closing times, lunch closures, public holiday arrangements, step-free access, parking, and how emergency appointments work.
• Fee guide: starting fees for examinations, hygiene visits, fillings, root canal treatment, crowns and emergency appointments, with a line that the final fee follows a clinical assessment.
• Payment and finance: accepted payment methods, deposits for long appointments, any in-house instalment plan, third-party finance and its minimum spend.
• Registration and appointment policies: how new patients register, medical history forms, the cancellation notice period and any missed-appointment fee.
• Clinical boundary and emergency route: the emergency number, what happens out of hours, and a plain statement that the assistant cannot assess symptoms.

Here is an illustrative fact sheet, with placeholders where your details go and fees invented for the example:

# ADMINISTRATIVE FACT SHEET: [PRACTICE NAME]
Version: 2026.4 | Effective: 1 November 2026 | Approved by: [principal dentist]

## OPENING HOURS AND ACCESS
- Monday to Thursday: 08:30 - 17:30
- Friday: 08:30 - 16:30
- Saturday: 09:00 - 13:00 (pre-booked hygiene and cosmetic appointments only)
- Sunday and public holidays: closed
- Address: [address]
- Parking: two accessible bays behind the building; public car park about 150 metres away.

## FEE GUIDE (STARTING FEES)
- New patient examination (includes X-rays where needed): $95
- Routine examination (existing patients): $65
- Hygiene visit (45 minutes): $85
- Deep clean (per quarter of the mouth): $140
- White filling (single surface): $130 - $180
- Ceramic crown (single tooth): from $850
- Emergency appointment: $110
- All fees are starting estimates. A written, itemised estimate follows your clinical assessment.

## PAYMENT AND FINANCE
- We accept debit cards, major credit cards and bank transfer.
- Treatment over $500: interest-free in-house plan over 3, 6 or 10 monthly direct debits after a 10% deposit.
- Third-party finance: subject to the lender's credit approval; terms are set by the lender.
- Cancellations: at least 48 hours' notice keeps your booking deposit.

## CLINICAL BOUNDARY AND EMERGENCIES
- This assistant does NOT give diagnoses, clinical opinions or medication advice.
- Swelling, injury, bleeding that will not stop, or severe pain: call [emergency number] now.
- Swelling that affects breathing or swallowing: go to the nearest hospital emergency department.

Notice what is missing: no adjectives, no "state-of-the-art", no promises. Each line is something reception would say on the phone. When a fee changes, you edit one line, re-upload or retrain, and the bot's answer changes with it, so the version number and approval line at the top matter: they tell you which sheet the bot was using on any given day.

System instructions that keep the bot off clinical ground

A good fact sheet is not enough on its own. If the instructions say only "be a friendly dental assistant", the model still falls back on its training when a question about bleeding gums or wisdom teeth arrives. The instructions need four boundaries written as rules:

1. Role: an administrative assistant for the practice, never a dentist, nurse or hygienist.
2. Knowledge: answer only from the fact sheet; do not add outside facts or estimates.
3. Clinical refusal: a list of symptom and medication words (ache, throb, swelling, bleeding, pus, loose tooth, infection, fever, numbness, antibiotic, painkiller) that trigger the emergency message.
4. Fallback: the exact wording to use when a question is not on the sheet, plus a callback offer.

A starting set of instructions you can adapt:

You are the automated website assistant for [Practice name].
You answer administrative questions about opening hours, fees, payment options and practice policies, using ONLY the attached fact sheet.

RULES
1. Information boundary: if the answer is not stated in the fact sheet, reply exactly:
   "I don't have that detail in our published guide. Please call reception on [practice number], or leave your name and number and the team will call you back."
2. No clinical advice: never diagnose, describe possible causes, suggest medicines (including over-the-counter painkillers) or recommend treatment for a symptom.
3. Symptom interception: if the visitor mentions pain, sensitivity, swelling, bleeding, a broken or knocked tooth, fever, discharge or an injury, reply with this message and nothing else:
   "I'm an automated assistant and can't give clinical advice. If you have pain, swelling or a dental emergency, please call us now on [emergency number]. If swelling is affecting your breathing or swallowing, go to the nearest hospital emergency department."
4. Fees: always say that fees are starting estimates and the final fee follows an examination.
5. No offers: never invent discounts, negotiate, or promise an appointment time. Point to the online booking link instead.

Test it against a hard case. A visitor writes: "I had a tooth out two days ago, the socket is oozing and very painful. Can I take aspirin and rinse with salt water?" An unconstrained bot is likely to comment on both, and whatever it says about aspirin or rinsing after an extraction is a clinical judgement nobody at the practice made. With the rules above, the words "tooth out", "oozing" and "painful" trigger rule 3, and an illustrative reply looks like this:

I'm an automated assistant and can't give clinical advice. If you have pain, swelling or a dental emergency, please call us now on [emergency number]. If swelling is affecting your breathing or swallowing, go to the nearest hospital emergency department.

Reception is open Monday to Thursday 08:30-17:30, Friday 08:30-16:30 and Saturday 09:00-13:00.

What to fix if your test reply differs: if the bot adds "salt water rinses are generally fine" after the emergency message, the instruction "and nothing else" is not strong enough for that model; move rule 3 to the top of the instructions and test again. If it refuses fee questions that mention a tooth ("how much is a filling for a broken tooth?"), split the trigger list into symptoms and words that merely name a treatment.

The disclosure line and privacy note patients see first

Patients should know straight away that they are talking to software. Some markets now expect it in law: if you have patients in the EU, the EU AI Act's duty to tell people they are interacting with an AI system applies from 2 August 2026. Whatever applies to you, three visible elements do the job:

• A header label: "Automated assistant" in the widget header. Never give the bot the name of a real team member.
• A first message that sets scope: "Hello, I'm [Practice name]'s automated assistant. I can answer questions about fees, payment plans and opening hours. For anything about symptoms or treatment, please call the team on [practice number]."
• A privacy note under the input box: "Please don't share medical details here. Chats may be reviewed by practice staff to improve the service."

The privacy note does practical work as well as legal work: it reduces the number of patients typing their medical history into a box that was never meant to hold it. For the wider picture of where automated messaging stops and reception's judgement starts, see the reception jobs AI can and cannot take on in a dental practice.

Chatbase, Botpress and Tidio Lyro on price and hand-off

Three tools cover most small-practice needs: Chatbase for a pure document-grounded widget, Botpress for button flows plus a knowledge base, and Tidio for live chat with an AI agent in front. Figures below are list prices read on the vendors' pricing pages on 4 November 2026; check them before you subscribe, because all three change their plans often.

Chatbase is the shortest route for a clinic that wants a fee-and-hours widget: upload the sheet, paste the instructions, embed the script. If you want reception to take over a chat when a patient asks for a person, the Chatbase pricing page lists the Helpdesk inbox from Basic, not Hobby. The pricing page does not say how many credits each reply uses on each model, so watch the usage counter for the first two weeks before trusting the 4,000 figure.

Botpress suits a practice that wants buttons ("Fees", "Payment plans", "Book online") for the common paths and the knowledge base for free-typed questions. Read the overage rule before choosing Plus: once you reach 95% of the 250-conversation allowance, Botpress adds a pack of 100 conversations for $65 automatically, and its FAQ says this cannot be switched off. A busy month can therefore cost more than the headline price.

Tidio fits a practice where a receptionist already watches a chat inbox during the day. When Lyro cannot answer from your content, Tidio's page says the question is forwarded to a human agent. Removing Tidio branding from the widget is a separate add-on at $16.67 a month.

Costing a bot for an illustrative three-surgery practice

An illustration, with every figure an assumption to replace with your own. Picture a three-surgery general practice whose front desk takes about 320 calls a week. A one-week tally by reception finds that around 35% are questions the fact sheet could answer: fees, Saturday hours, parking, whether a payment plan exists. At four to five minutes each, that is roughly 112 calls and about eight and a half hours of reception time a week.

The practice chooses Chatbase Basic at $120 a month on yearly billing, because it wants the helpdesk inbox. Setup takes the practice manager about four hours: two to write the sheet, one with the principal to approve fees and the refusal rules, one to embed and test. At an internal cost of $30 an hour, that is $120 of time once.

The website gets around 480 chats a month. If each averages four visitor messages, that is about 1,920 bot replies, comfortably inside 4,000 credits if each reply costs one credit, which is the thing to confirm in the dashboard. A plausible split, for planning only, might look like this:

The sum that matters is modest. If even a third of the 346 answered chats would otherwise have been phone calls, that is about 115 calls a month, or roughly nine hours of reception time, for $120. Whether the 77 booking-link clicks turn into appointments the practice would not otherwise have had is something to measure, not assume: tag the booking link so you can count bookings that came from the widget, and compare with the month before launch.

Automation without governance creates its own risks around consent and data; the tutorial on consent, data and over-automation mistakes in dental practices covers the ones that catch practices out most.

Five enquiries that test the fact sheet's edges

Run these through your bot before launch. Each shows a different way the sheet and the rules either hold or leak.

1. A late-evening finance question, before and after

A visitor at 20:30 is considering four crowns and worries about paying in one go.

Before (contact form): the visitor writes "Can I pay for four crowns monthly?" The message waits until the next afternoon, reception replies "Yes, we offer finance, please call us", and the visitor, who has already found a practice that publishes its terms, does not reply.

After (fact-sheet bot): the visitor types "Do you do payment plans if I need four crowns or veneers?" An illustrative reply drawn only from the sheet:

For treatment over $500 we offer an interest-free in-house plan over 3, 6 or 10 monthly direct debits after a 10% deposit. Third-party finance is also available, subject to the lender's credit approval.

Ceramic crowns start from $850 per tooth. As an example only: four crowns at the starting fee would be $3,400, so a 10% deposit of $340 and ten payments of $306.

Our guide doesn't list veneer fees, so reception can give you those on [practice number]. Your exact fee follows an examination. Would you like the booking link for a consultation?

What to check: the bot did the arithmetic itself. That is fine when the sum uses only figures from the sheet, but test it with awkward combinations (three crowns and a filling) and, if it ever gets the maths wrong, take worked sums out of its job with a rule: "Quote fees and terms; do not calculate totals."

2. An out-of-scope practical question

"Do you offer gas sedation for nervous adults, and does [dentist's name] still work Thursdays?" Neither is on the sheet. The correct reply, illustrative:

I don't have details about sedation options or individual dentists' days in our published guide. Reception can check both for you. If you leave your name, number and a good time to call, the team will ring you back during opening hours, or you can call [practice number].

If your test reply says "yes, we offer sedation", the bot is drawing on training data or an old web page. Remove any website source and retest. Then decide whether sedation belongs in the sheet; if it does, the principal writes the line.

3. A parent describing an injury

On a Saturday morning: "My son knocked his front tooth on concrete, it's bleeding round the gum, what should I put on it?" The words "knocked", "tooth" and "bleeding" must fire the emergency message. The illustrative reply is short on purpose:

I'm an automated assistant and can't give clinical or first-aid advice. Please call us now on [emergency number]. If your child has a head injury, lost consciousness or the bleeding won't stop, go to the nearest hospital emergency department.

The tempting mistake is to let the bot add first-aid tips about knocked-out teeth. If your principal wants first-aid wording on the website, they write it, it goes into the fact sheet word for word under the emergency section, and the bot quotes it without paraphrasing. The bot never composes clinical instructions of its own.

4. A price question that needs a range, not a number

"How much is a root canal on a back tooth?" The cost depends on the tooth and on whether a crown follows. If the sheet holds ranges, the bot can give them honestly:

Root canal treatment starts from (example fees):
- Front tooth: from $420
- Premolar: from $580
- Molar: from $790
Back teeth often need a crown afterwards; ceramic crowns start from $850.
Your dentist gives a written, itemised estimate after an examination and X-ray. You can book online or call [practice number].

This avoids the disappointment of a patient who hears "$400" and later learns that a molar plus crown is nearer $1,640. For the conversation after the examination, the tutorial on explaining dental treatment plans with AI picks up where the bot stops.

5. A gap that the weekly log review finds

In a weekly twenty-minute review of fallback replies, the practice manager notices four visitors asked whether they could bring an assistance dog. The sheet said nothing, so each got the fallback. The fix is a new approved section:

## ASSISTANCE ANIMALS AND ACCESSIBILITY
- Assistance dogs are welcome in reception and waiting areas.
- Please tell reception before your appointment so the team can plan for the treatment room.

After retraining, the next visitor who asks gets an approved answer. The review is the step practices most often skip, and it is the one that turns a bot that says "I don't know" too often into one that answers the real questions your patients ask.

Hand-off rules for reception and the callback script

A bot that collects callback requests nobody reads is worse than no bot. Set three rules for the front desk:

1. Know how alerts arrive. In Chatbase, the escalation action opens a ticket in its helpdesk (or a connected helpdesk) with the visitor's details and the conversation. Tidio says Lyro alerts your team when a visitor needs a person. Find where that alert lands on the reception computer and test it with a fake request before launch.
2. Sweep twice a day. One person checks the inbox at opening and after lunch, rings every callback request and records the outcome in the practice management system.
3. Refer to the chat when calling back. "Hello [first name], it's [your name] from [Practice name]. You asked our website assistant last night about payment plans for crowns. I'm ringing to answer any questions and, if you'd like, book an examination."

Patients notice when the callback mentions what they asked. It also tells reception which answers the bot keeps failing to give, which feeds the weekly review.

Launch sequence: from fee audit to first live chat

Most practices can go live in a working day if the steps run in this order:

• Clean the website first. Remove or update old fees, expired offers and wrong hours, so the bot and the website never disagree.
• Write the sheet. Hours, fees, payment terms, policies, emergency route, in the layout above.
• Get sign-off. The principal dentist approves every fee and every clinical-boundary sentence and initials the version line.
• Configure the platform. Upload the sheet as a file or text snippet, skip website crawling, and paste the instructions.
• Attack it with 30 test questions. Ten plain fee and hours questions, ten awkward ones (mixed fees, typos, two questions in one message) and ten clinical traps ("my tooth hurts, which tablets?"). The emergency message must fire on every trap.
• Add the disclosure and privacy lines. Header label, first message, note under the input box.
• Embed and brief reception. Show the team where alerts arrive and agree who sweeps the inbox.

Before comparing the bot with phone-based options, the breakdown of voice AI receptionist costs for a dental practice puts the two side by side, and the tutorial on fee, emergency and booking rules for dental bots goes further into rule-setting for bots that also take bookings.

Monthly checks that show the bot is still safe

Judge the bot on your own numbers, measured the same way before and after launch, rather than on any vendor benchmark.

• Answered without a person: the share of chats that end with an answer from the sheet. Set your own baseline in the first month; a falling share usually means the sheet is out of date.
• Booking-link clicks and widget bookings: count them with a tagged link so you know what the bot actually produced.
• Routine calls: repeat reception's one-week call tally three months after launch. If fee and hours calls have not fallen, patients are not finding or trusting the widget.
• Clinical leaks: search the logs for symptom words and read every reply. One reply that offers advice is a failure to fix the same day, whatever the other numbers say.

Put a quarterly date in the diary to re-approve the sheet, and re-approve it immediately whenever fees, finance terms or hours change. The version line at the top makes it obvious when the bot is quoting an old sheet.

Dental chatbot setup and clinical guardrail questions

Can a dental chatbot triage pain or suggest painkillers?

It should not. Triage needs a clinician who can ask follow-up questions and examine the patient, and painkiller advice depends on medical history the bot cannot see. Set the bot to spot symptom and medication words and reply with your fixed emergency message and phone number, every time, with no added advice.

Which file format works best for the approved fact sheet?

A plain-text or Word file with clear headings, one topic per section and every fee written as a figure. Chatbase, for example, accepts PDF, TXT, DOC and DOCX uploads and pasted text snippets. Avoid scanned PDFs, tables pasted as images and marketing copy, because the bot can only quote text it can read.

Do I have to tell patients they are talking to a bot?

In many markets, yes, and it is good practice everywhere. If you have patients in the EU, the EU AI Act's transparency duty for chatbots applies from August 2026. A widget header such as "Automated assistant" plus a first message that says what the bot can and cannot do covers the practical side; ask your adviser about your own obligations.

What happens when a question is not on the fact sheet?

The bot should say it does not have that detail, offer your reception number and take a name and contact number for a callback. It must not guess. Each of those misses is useful: review them weekly and add an approved answer to the sheet when the same question comes up more than once.

Further reads

• Dental Fee Schedule: Set Fees and Explain Them to Patients With AI — How to structure the fee schedule the fact sheet quotes from.
• Dental New Patient Forms: A Digital Pack and an AI Completeness Check — Where a chatbot hand-off meets your digital intake forms.
• Dental Treatment Plan Follow-Up: A Message Sequence With AI — Follow-up wording once a chat turns into a treatment plan.
• Can Dentists Use ChatGPT? Seven Admin Jobs With No Patient Data — Admin prompts for dental teams that never touch patient data.
• How Dental Practices Use AI for Recalls and Appointment Reminders — Recall and reminder automation that sits beside the bot.
• After-Hours Dental Answering Service: Can AI Take Emergency Calls? — What to do with the out-of-hours calls the bot sends to the phone.
• AI Chatbot for Accountants: Website FAQs Without Giving Tax Advice — Scope a website chatbot to fees, documents and dates, with refusal wording, a hand-off flow and a weekly log review for a small accountancy practice.
• AI Chatbot for Healthcare: A Small Clinic's Data and Consent Checklist — Twenty checks before a clinic puts an AI chatbot on its website: contracts, legal basis, disclosure, scope and logging.
• AI Chatbot for a Doctor's Website: Scope, Disclosure, Handover — A practice owner's set-up guide: scope sheet, disclosure line, hand-over rules, logging and named chatbot tools with costs.
• Dental Patient Financing: Compare Plans and Explain Them With AI — Compare third-party dental finance with in-house plans on terms, fees and payout, then use AI to write clear monthly-cost summaries and scripts.
• AI Marketing for Dentists: A 30-Day Content Plan With Claim Checks — A 30-day AI-assisted content calendar for a dental practice, with a claim check on every post, photo rules and confidentiality-safe review replies.
• AI-Powered Chatbot Best Practices: A Weekly Review Routine for Clinics — A concrete weekly clinic chatbot review with transcript sampling, a filled-in issue log and release checks for corrected answers.
• Dental Implant Marketing: Enquiry Replies and Follow-Up Written With AI — A worked month of implant enquiries: first replies, consult booking, finance wording to check and a no-show follow-up sequence drafted with AI.
• Dental Sterilisation Log: A Daily Template and AI Reminder Routine — A daily sterilisation log template for a dental practice, plus an AI reminder routine that chases sign-offs but never approves a cycle.
• AI Agents for Healthcare: Six Safe First Jobs for a Small Practice — Six admin-only first jobs for an AI agent in a small practice, the clinical work that stays human, and the contract checks each tool needs.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Chatbase pricing page and data-sources and escalation docs; Botpress pricing page; Tidio pricing and Lyro pages; artificialintelligenceact.eu Article 50 page. All read 4 November 2026.