Does a privacy policy have to mention an AI phone receptionist?

Only a lawyer can answer for a particular business, and two tests decide it. The first is whether the business is covered by the Privacy Act at all: businesses with turnover of $3 million or less are generally exempt, but health service providers such as dentists and GPs are covered regardless of size. The second, from 10 December 2026, is whether a computer program makes, or substantially helps make, a decision that could significantly affect a person. A receptionist that writes a patient's chosen time into a free slot is a weaker candidate than one that decides who is turned away.

What counts as an automated decision under Australian privacy law?

Australian Privacy Principle 1.7, which applies from 10 December 2026, asks whether an organisation has arranged for a computer program to make a decision, or do something substantially and directly related to making one, where the decision could reasonably be expected to significantly affect a person's rights or interests and personal information is used in running the program. The regulator reads computer program broadly, including generative AI and chatbots, and says a person reviewing the output does not automatically take a system out of scope. Refusing or failing to decide counts as deciding.

Is a dental clinic exempt from the Privacy Act as a small business?

No. The small business exemption in the Privacy Act generally covers businesses with an annual turnover of $3 million or less, but the Office of the Australian Information Commissioner lists providers of health services as covered regardless of turnover. Dentists, general practitioners, pharmacists, physiotherapists, psychologists and complementary medicine therapists are all named, along with gyms and weight loss clinics that collect health information. A single-practitioner clinic therefore carries the same privacy policy obligations as a large group, including the automated decision-making disclosure that applies from 10 December 2026. Confirm the position for a specific business with a lawyer.

If a staff member checks everything the AI receptionist does, is it still an automated decision?

Not necessarily. The Office of the Australian Information Commissioner's guidance, published on 30 September 2026, says a program can be substantially and directly related to a decision even where a person reviews its output, and that generative AI outputs used in significant decisions generally fall within scope unless there is extensive human oversight and control. A receptionist whose bookings a nurse glances over the next morning is a different case from a system whose recommendation a nurse must document reasons for overriding. Where a clinic sits is a legal judgement, so write down what the system does and what the person does with it.

Does the Privacy Act's 10 December 2026 Automated Decision Rule Cover Your Clinic's AI Receptionist?

On 10 December 2026, a new line in the Privacy Act starts applying to every organisation the Act covers. From that day, if a computer program makes a decision that could significantly affect a person, or does something substantially and directly related to making one, the organisation's privacy policy has to say so, and say what kinds of personal information the program uses.

A practice manager at a dental or medical clinic in Sydney and Greater NSW with an AI receptionist on the phones will hear two confident answers. The vendor says the rule is for banks and recruiters and has nothing to do with answering calls. A compliance newsletter says every chatbot is now an automated decision-maker and the policy needs rewriting by next month. Neither is worth much, because the rule turns on what the particular system decides, and few of the people giving the answers have asked.

Durbacti is not a law firm, and nothing here is legal advice. This post does what Call Recording and Consent in NSW: What to Sort Out Before You Automate Your Phones did for recording: describe the rule from its primary sources, describe what an AI receptionist does and does not decide, and set out what to hand a lawyer so the question can be answered in an afternoon.

What changes on 10 December 2026?

The change is three new subclauses in Australian Privacy Principle 1, inserted by the Privacy and Other Legislation Amendment Act 2024. The Act received Royal Assent on 10 December 2024, and the text of the Act on the Federal Register of Legislation sets these provisions to commence the day after 24 months from that date, which is 10 December 2026.

APP 1.7 says an organisation's privacy policy must contain extra information if three things are all true: the organisation "has arranged for a computer program to make, or do a thing that is substantially and directly related to making, a decision"; the decision "could reasonably be expected to significantly affect the rights or interests of an individual"; and personal information about that individual is used in running the program. APP 1.8 says what the extra information is: the kinds of personal information the programs use, the kinds of decisions they make on their own, and the kinds of decisions they substantially help with. APP 1.9 adds that making a decision includes refusing or failing to make one, and that a decision counts whether it helps or harms the person.

The Office of the Australian Information Commissioner published its guidance on the new obligation on 30 September 2026: an updated Chapter 1 of the APP Guidelines and an APP 1.7 to 1.9 transparency obligation fact sheet, both read on 9 November 2026. Three points in them matter for anyone running an AI receptionist. "Computer program" is read broadly and expressly includes generative AI and chatbots. "Significantly affect" means an effect that is more than trivial and could "considerably influence the circumstances or outcomes" for the person. And a person checking the output does not automatically take a program out of scope; the guidance says machine learning and generative AI outputs used in significant decisions "generally fall within scope of the transparency obligation unless subject to extensive human oversight and control".

The obligation is to describe, not to stop. Nothing in APP 1.7 bans an automated decision. It requires the privacy policy to tell people that one is being made and what information feeds it.

Is a dental or medical clinic covered at all?

Many Sydney small businesses sit outside the Privacy Act altogether. The OAIC's small business page, last updated 5 September 2024, says a small business is one with an annual turnover of $3 million or less, and that most small businesses are not covered. A sole-trader plumber in Sutherland with an AI receptionist is, in the usual case, not an APP entity, and APP 1.7 does not reach the plumber's privacy policy, if there is one.

Health is the exception that matters here. The same page lists providers of health services as covered regardless of turnover, and the OAIC's page on what is a health service provider, last updated 27 October 2023, names general practitioners, dentists, pharmacists, physiotherapists, psychologists, complementary medicine therapists, and gyms and weight loss clinics where they collect health information. A two-chair dental practice in Parramatta is an APP entity on its first day of trading. So is a physiotherapist working alone.

That is why this post is written for the clinic rather than the tradie. The clinic already has a privacy policy because it has to, and on 10 December 2026 that policy may need a paragraph it has never needed before. Whether the exemption applies to a particular business, and what else pulls a business inside the Act, is the first question for a lawyer, not the last.

Is booking an appointment a decision that significantly affects someone?

Here is the distinction the whole question turns on, and the one a vendor's "it is just a receptionist" skips past.

A patient rings at 7 pm. The 24/7 AI phone receptionist answers, asks what they need, offers the times the appointment book shows as free, and books the one the patient picks. The patient made the choice. The program read a calendar and wrote an entry. The receptionist Durbacti builds does not decide who deserves an appointment, does not offer a time the calendar shows as taken, and does not quote a fee that has not been set. On its face that is a booking, not an adjudication.

Now the other case. The OAIC fact sheet's first worked example is a health and aged care provider that uses a spreadsheet to score and rank people's need for a service from their personal and health information, with staff using the ranking to decide whom to contact and schedule. The fact sheet says plainly that "the use of the Echidna spreadsheet constitutes a decision within the scope of the transparency obligation", and that using the ranking to decide whom not to contact is in scope too. The program is not writing down a choice the patient made. It is deciding, or substantially shaping, who gets seen.

Most AI receptionist setups sit nearer the first case than the second. Two things stop that being the end of the discussion. APP 1.9 says doing a thing "includes refusing or failing to do a thing", so a receptionist configured to turn certain callers away, for example to decline new patients, or to refuse to book a treatment type over the phone, is doing something closer to a decision about access than a receptionist that simply records a choice. And the OAIC's list of in-scope decisions includes limiting access to healthcare and prioritising health or disability services, so a clinic whose receptionist also runs a waiting list, deciding who is offered a cancelled slot first, has a system that looks more like the spreadsheet than the diary.

Which side of that line a particular clinic's setup falls on is a legal judgement about that clinic. Durbacti can describe the system. It cannot classify it.

What personal information does an AI receptionist actually use?

Whatever the answer to the classification question, APP 1.8 asks for "the kinds of personal information used in the operation of such computer programs", and a clinic cannot write that paragraph if nobody has told it what the system holds. On a Durbacti build the voice platform is Vapi, and Vapi's own documentation is the place to answer from.

Vapi's call recording documentation, read on 9 November 2026, says that "by default, Vapi stores a complete recording of every call", that recording stays on unless the assistant is configured to turn it off, and that recordings can be sent to a storage bucket the client controls instead of being hosted by Vapi. Its data flow documentation, read the same day, lists what is stored by default: call recordings with configurable retention, full transcripts with timestamps, call logs, and structured outputs from call analysis, which on a Durbacti build is where the summary of what the caller wanted and what was booked lives.

Translate that into the categories a privacy policy speaks in, and a clinic's AI receptionist uses, at minimum: the caller's phone number; the name and contact details they give; the date, time and practitioner they book; and the words they say on the call, which at a clinic routinely include why they want to be seen. The OAIC guidance says more sensitive information, health information in particular, should be clearly identified in the policy. A patient saying "it's the tooth on the left again" into a recorded line is health information whether or not the receptionist does anything with it.

The practical consequence is that the recording and retention settings decided during the call recording conversation are the same settings that decide what the privacy policy has to list. Keep recordings for seven days, and the policy says so. Turn recordings off and keep the transcript, and the policy says that instead.

Which other automations look more like the OAIC's examples?

A clinic's AI receptionist is rarely the only program touching personal information, and some of the others sit much closer to the examples in the guidance.

•

Automatic refund or fee decisions. The APP 1 Guidelines give the example of a third-party tool that automatically approves or declines refunds. An AI customer support assistant that is allowed to approve a refund, waive a cancellation fee or decline a request on its own is making a decision that affects a person's rights under a contract, which APP 1.9 lists by name. The Durbacti position is that an assistant answers from the clinic's stated policy and hands anything discretionary to a person, and that choice is made at scoping, not discovered later.

•

Ranking enquiries. Software that sorts new patient enquiries by a score it calculated about the person, and decides who gets a call back first or not at all, is doing what the triage spreadsheet does. Sorting by the order enquiries arrived is not.

•

Waiting lists and cancellation offers. A Smart scheduling system that offers a cancelled slot to whoever responds first is running a queue. One that decides who is offered the slot based on information about the patient is prioritising access to a health service. Same feature on the sales page, different answer under APP 1.7.

•

Pre-screening forms. Any form or chatbot that collects health information and then decides whether the person may book, or with which practitioner, before anyone at the clinic sees it.

None of these is forbidden. Each is a candidate for a sentence in the privacy policy, and each needs a lawyer to say whether it crosses the line.

What does Durbacti hand over so a lawyer can answer in an afternoon?

The OAIC fact sheet says responsibility for the policy stays with the organisation using the information, and that providers of software to those organisations "should provide clear, high-level information about how their software can be used to make decisions". Durbacti reads that as a description of the handover document it already produces, with one more page.

Every build described on How we work ends with documentation the client owns. For a clinic, that documentation includes a plain-language page covering four things: what the system decides on its own, which for a standard receptionist is nothing beyond writing the patient's chosen time into a free slot; what it does that feeds a decision a person makes, such as the call summary the front desk reads in the morning, the flag on a call that mentioned pain, and the transfer rules; the kinds of personal information it uses, with recording and retention as actually configured; and where each of those things is stored and for how long.

That page is not a privacy policy and it is not legal advice. It is the input a lawyer needs to write the APP 1.8 paragraph, or to say with reasons that the clinic does not need one. Writing it takes an hour at scoping. Reconstructing it from a vendor's marketing site in the first week of December takes longer.

What we will not do

•

Tell a clinic whether APP 1.7 applies to it. That is a legal classification of a particular business and a particular system, and the answer belongs to a lawyer who has seen both.

•

Draft the privacy policy paragraph. Durbacti describes what the system does, in the categories the guidance asks for. Someone qualified turns that into policy wording.

•

Let the receptionist make the decisions that sit in the regulator's examples. No approving or declining refunds, no ranking who gets seen, no deciding who is turned away, unless the clinic has asked for exactly that, understood what it means for the policy, and signed off on the rules.

•

Pretend the law is settled. The guidance is under six weeks old as this is written and the obligation has not started. What a regulator or a court makes of a chatbot that books appointments is not yet known, and anyone claiming otherwise is guessing.

What to ask a lawyer, and what to ask the vendor

Take these to a lawyer before 10 December 2026:

•

Are we an APP entity, and on what basis? For a clinic the usual answer is yes, as a health service provider, regardless of turnover.

•

Does our AI receptionist make, or substantially and directly inform, any decision that could significantly affect a patient? Hand over the system description rather than the brochure.

•

Does any rule we have given it about turning callers away or declining bookings change that answer?

•

Which of our other systems, the online booking form, the recall list, the waitlist, look like the OAIC's examples?

And take these to the vendor, Durbacti included:

•

What does the system decide on its own, and what does it hand to a person? A vendor who cannot answer in a sentence has not scoped it.

•

List the personal information it stores, where, and for how long, in writing, before go-live. The answer should come from the platform's documentation, with a link.

A clinic's AI receptionist that books the slot a patient chose is, on its face, a long way from the loan approvals and triage spreadsheets in the regulator's examples. The gap narrows the moment the receptionist is allowed to decide who is turned away or who is seen first, and by 10 December 2026 the privacy policy has to be able to say which kind of system the clinic is running. The Dental and medical clinics page sets out how Durbacti builds a receptionist that records choices rather than making them, and a free strategy call is where the one-page description for your lawyer gets started.