Dental AI Receptionist vs Human Reception: What AI Should and Should Not Do

What an AI receptionist is for

In a dental practice, an AI receptionist usually sits on website chat, and often on WhatsApp, SMS, email, and Facebook messages. It acknowledges the enquiry, collects what a coordinator needs, answers from facts the practice has published, and offers a next step such as a consultation booking or a callback.

If a vendor demo blurs that line into examining someone, planning treatment, or choosing a clinically appropriate diary slot, treat it as a warning. The test is simple. Could a well-briefed receptionist say this from a script card, without opening the clinical record and without guessing? If yes, software can often say it too. If the answer depends on radiographs, medical history, or how something looks, it belongs with a person.

What AI should do

Reply when reception cannot

The phone queue at 9am, the lunch hour, and the evening after work are when a lot of intent shows up, and when a small team is least able to answer everything at once. An AI receptionist should take the channels that would otherwise sit in a voicemail or an unread inbox. On denbot.co.uk we already use an illustrative industry figure: around 73% of patient enquiries arrive outside business hours. Treat that as a planning assumption for coverage, not as a measurement of your own phone system. Your own timestamps will tell you the truth within a month.

A fast reply is an acknowledgement, not a pretence that a dentist is awake. Say you have the enquiry, explain how booking works, and promise that a person will confirm in opening hours.

Collect a usable enquiry

A name on its own is not a lead. The software should ask for a mobile number, an email address, whether the person is a new or existing patient, what they want to talk about, and a sense of when they can come in. It should ask one thing at a time, in plain language, and it should cope with people who answer in a sentence rather than a neat form.

Answer from published facts only

Opening hours, the address, how to find the practice, which treatments you offer, and how to request a consultation are fair game when they match what the practice has published and is willing to stand behind. The moment the answer would be "it depends on a consultation", the software should say that, and stop.

Fees are the sharp edge. If the practice has not given the software a current, approved figure, the correct reply is that the team will confirm the fee. Inventing a price, or quoting a figure from an old page, creates a promise the practice then has to unwind.

Offer one next step

The conversation should end in a concrete action: a booking link the practice controls, a choice of callback windows, or a clear statement that reception will be in touch at open. People enquire because they want a next step.

Leave a record the team can actually use

Every conversation should land in one place, with the source, the time, the contact details, and the transcript. The morning handover is then a list of owned enquiries. If the software cannot show a person what was already promised, it is not ready.

What AI should not do

Diagnose, assess urgency from symptoms, or promise an outcome

Toothache, swelling, bleeding, and trauma need the practice's written emergency route, not a chat assessment. The software should point people to those instructions, and to the urgent care route the practice has approved for patients in that nation (for example NHS 111 in England and Wales, or NHS 24 in Scotland). If someone may be in immediate danger, the message is to call 999. The software must not decide how serious a problem is from a sentence in a widget.

The same boundary covers guarantees and suitability. An AI receptionist can explain that a consultation is how the practice assesses suitability. It cannot issue the assessment.

Pretend to be a named clinician, or imply a professional status

Patients should know they are talking to the practice's automated receptionist. Do not give the bot a dentist's name, a registration number, or language that implies a registered clinician. If a patient asks whether they are speaking to the dentist, the honest answer is no, followed by how to reach the team.

Handle complaints, safeguarding, or consent for treatment

A complaint needs a person, a proper record, and the practice's complaints process. Safeguarding concerns need a person. Consent for treatment is a conversation with the clinician, recorded in the clinical record, not a tick in a chat. The software should recognise these moments and hand them over, not try to close them.

Edit the clinical record or the day list unsupervised

Practice management software holds the chart, the medical history, and the appointment book for patients you are caring for. An enquiry tool should not rewrite that record because a chat sounded confident. Offering a new consultation through a route the practice has explicitly set up is reasonable. Moving an existing patient's visit, deleting an appointment, or writing into the notes is not.

Keep contacting someone who has asked to stop

If a person says stop, or asks not to be messaged, the sequence ends. That preference has to be stored on the lead, visible to the whole team, and respected on WhatsApp, SMS, email, and Facebook alike. An automated chase that ignores a stop request is not follow-up. It is a compliance problem.

What human reception should keep

Reception is the practice's judgement layer and its welcome. People still do the work that depends on tone, context, and accountability.

• Greeting patients in the building, and calming someone who is anxious before they sit down.
• Existing-patient questions that need the record, such as a change to a long visit or a question about a plan they have already been given.
• The live day list, when a cancellation opens a gap that a person should fill with judgement rather than a blunt rule.
• Complaints, upset callers, and anything that sounds clinical or urgent once a human is on shift.
• The final word when a diary slot depends on the clinician, the surgery, or the length of the visit.
• Relationships. A patient who has attended for years should not be trapped in a bot when they ask for the person they know.

The aim is to stop reception retyping overnight enquiries, so the first hour belongs to the people in the building.

A practical split for a normal week

Agree the split in writing. A vendor slide will not survive a busy Monday.

An evening message about aligners or implants. The AI receptionist acknowledges, collects contact details and what the person wants, and offers the consultation route. It does not quote an unpublished fee and it does not say they are a candidate. Reception reviews the lead at open, and a person takes the conversation if the patient is ready to book or has questions the script cannot answer.

The Monday morning phone queue. A person answers the phone. The AI receptionist keeps website and social enquiries moving so they do not sit behind the handset. Reception is not removed. Reception is no longer the only queue.

A message about swelling or severe pain. The software does not assess. It gives the practice's emergency instructions and flags the conversation for a person immediately if someone is working. If nobody is working, the message still has to tell the patient how to get urgent help, in words the practice has approved.

A fee question. If you have published a fee and loaded it as a current fact, the software may repeat it and say what it includes. If the fee depends on an examination, the software says the team will confirm it. A person handles the plan conversation.

How to tell whether the software is behaving

Judge the tool on behaviour you can inspect, not on a demo that always goes well.

• You can read every conversation, including the ones that happened late in the evening.
• A person can take over mid-chat, and the patient is not bounced back to the bot against their will.
• Clinical questions get a refusal and a handover, not a confident paragraph.
• Fees appear only when your team has supplied them.
• Contact details, source, and the next action sit on one pipeline, not in a personal inbox.
• Stop requests are visible and honoured.
• The bot says when it does not know.

Before you point real patients at it, ask it to diagnose pain, to guarantee a result, to pretend it is the principal, and to quote a fee you have not published. If it complies, it is not ready. If it refuses and offers a human next step, you have the right shape.

Where this sits next to the rest of the practice

An AI receptionist is one layer. The clinical record stays in your practice management software. The enquiry, the owner, and the follow-up live in a CRM pipeline, with WhatsApp, SMS, email, and Facebook on the same record. Our comparison of a dental CRM and practice management software sets out that boundary in more detail.

Denbot is built for that enquiry layer: replies, a pipeline, and website chat plus WhatsApp, SMS, email, and Facebook leads, with the team able to take over. It does not sit in the surgery, replace the person who knows your patients, or run advertising. If the problem is the day list inside your practice management system, fix that system rather than asking a chatbot to become it.

Frequently asked questions

Will an AI receptionist replace our reception team?

No. It covers speed and coverage outside staffed hours. Reception still owns judgement, the in-person welcome, complaints, urgent situations once a person is available, and the clinical record.

Can it book appointments on its own?

It can offer a booking route you have set up, and it can collect a preferred time. A person should still own exceptions, clashes, and any visit that depends on the clinical record or the length of the appointment.

Is it allowed to give prices?

Only fees the practice has published and loaded as current. If a fee depends on an examination, the honest reply is that the team will confirm it at a consultation. The software should not guess.

What should it do with out-of-hours toothache?

Follow the practice's written emergency instructions and signpost urgent care. It should not decide how serious the problem is from the chat.

Do patients mind talking to software?

Many people are glad of a clear reply in the evening, provided the practice is honest that a person will follow up where needed, and provided the software does not bluff. Irritation starts when the bot traps them, invents an answer, or pretends to be a dentist.

How is this different from a website chatbot?

A chatbot is the conversation on the site. An AI receptionist is that conversation plus the handover: a complete lead, a next step, and a human who can see what was said. A widget that chats and then emails a partial transcript to a shared inbox is only half the job. Our guide to what a dental practice chatbot should do covers the website side in more depth.