What Never to Hand to AI at Reception

What's off limits, in one line?

Anything that needs clinical judgment. Reception can use AI for wording. It shouldn't use AI to decide what's wrong with a patient or what should be done about it.

Which tasks must stay with the dentist and trained staff?

• Answering symptom questions, such as "is this normal?" or "should I come in today?"
• Sorting callers by how urgent their problem is.
• Writing or summarising clinical notes.
• Naming a likely cause for pain, swelling or any other problem.
• Advice about treatment, aftercare or medicine.
• Deciding whether someone can wait for an appointment.
• Judging whether a patient has understood and agreed to a procedure.

Each of these depends on training, on the patient's record and on the dentist's responsibility. None of them is front-desk admin.

Why isn't AI trusted with these?

Because it can be fluent and wrong. AI can present false or misleading information as fact, which is called a hallucination.¹ These errors can be plausible but incorrect statements.²

In a reminder, a wrong time is a nuisance. In a symptom answer, a wrong statement could affect someone's health. That's why the line sits here.

What does the regulator say about responsibility?

In Australia, Ahpra's AI guidance puts the responsibility on the practitioner, whatever technology is used.³ With scribing tools, it says the practitioner is responsible for checking the records the AI creates.³

That guidance is for registered practitioners, not reception staff. It's still a reason for a clinic to keep clinical records and clinical judgment with the people accountable for them. Any AI use on clinical work is for the dentists and their regulator to decide. This site doesn't cover it.

Ahpra also notes that some AI tools used in healthcare are regulated by Australia's Therapeutic Goods Administration.³ A general chatbot isn't the same thing as a regulated clinical tool. Don't assume a writing assistant has been checked for clinical use.

What should reception do instead?

Have a plain script that doesn't need a tool.

• Write down who takes clinical questions, by name or role, at each hour.
• When a patient describes symptoms, staff say they'll pass it to the dentist or clinical team, then do so.
• If the person seems to be in distress or the situation sounds serious, follow the clinic's own emergency process. Staff don't wait for a tool.
• For automated phone or message tools, set a rule that any mention of pain, swelling, bleeding, medicine or an accident is handed to a person at once. See AI Phone Answering for Dental Clinics: How It Works.

What about a tool that "helps" with an enquiry?

If a patient's message includes a health question, don't paste it into a tool to see what it says. Strip nothing and ask nothing. Send the question to clinical staff. Even a harmless-sounding answer can sound like advice once it leaves the clinic.

Wording tasks are different. Drafting a recall note or tidying a form is covered in recall messages and enquiry replies and forms. A person checks each one.

Does the same rule apply to billing and fees?

Treat these carefully. A tool can phrase a payment reminder, but the clinic supplies every figure and wording about what's covered by insurance or a health fund. For tax or legal questions, point people to the official source for their country. That's the ATO in Australia, the IRS in the United States and HMRC in the UK. Don't ask a chatbot.

How do I make the rule stick?

Put it in the staff guide as a short list, and say it at induction. A rule that lives only in one manager's head doesn't protect anyone. Review it when a new tool is added.