Do You Have to Tell Callers They're Speaking to AI in Australia?

Does an AI receptionist have to identify itself in Australia? There is no single statutory script for every call and every business. As a practical operating rule, though, say it is AI in the first sentence.

That operating rule draws on official privacy and responsible-AI guidance. General consumer-law risk reinforces it without creating a universal wording that every business must use.

This is practical business guidance, not legal advice. If your calls involve health, financial, legal or other sensitive information, get advice for your specific setting before launch.

What the Australian guidance actually says

The clearest wording comes from the Office of the Australian Information Commissioner. Its guidance says businesses should give clear and transparent information about their use of AI. It also says public-facing AI tools, such as chatbots, should be clearly identified to customers.

That matters for phone systems because a caller may share personal information before they know what is handling the conversation. The OAIC says privacy obligations apply to personal information entered into an AI system and to AI output that contains personal information.

Australia’s AI Ethics Principles add a broader expectation. People should be able to find out when an AI system is engaging with them and obtain reasonable disclosure about the AI system. The current Guidance for AI Adoption is guidance rather than a blanket disclosure statute, but its six practices are designed to support responsible adoption and customer trust.

Consumer law is the third part. The ACCC’s guidance on false or misleading claims says a false impression can break the law even when misleading someone was not the business’s intention. It also says silence can mislead when important details relevant to a consumer’s decision are omitted.

You do not need a legal lecture in the greeting. You do need an opening that plainly identifies the voice as AI.

Put the disclosure in the first sentence

A customer should not have to work out whether the friendly voice belongs to a person. Disclose before asking for a name, phone number, booking details or the reason for the call.

A useful opening does four jobs:

1. identifies the business
2. says the caller is speaking with an AI assistant
3. explains the limited job it can perform
4. offers access to a person

Here is an illustrative script, not a record of an actual client call:

“Thanks for calling [business name]. You’re speaking with our AI receptionist. I can take a message, answer basic service questions or help with a booking. You can ask for a person at any time.”

That wording is deliberately plain. Avoid softened labels such as “virtual team member” or “digital concierge”. Neither identifies the system plainly as AI. Say “AI receptionist” or “AI assistant”. A natural voice is useful. A concealed identity is not.

The same rule belongs in a customer-facing voice-agent setup, whether it answers every call or only handles overflow and after-hours traffic.

Explain what happens to the caller’s details

Identification is only the first layer. The caller also needs a practical understanding of what the system will do next.

Keep the explanation tied to the task. If the system is taking a message, say the details will be sent to the team. If it is checking a calendar, say it can offer available times. If the call is transcribed or recorded, explain that before collecting the information and make sure the process meets the rules that apply to your business and location.

The AI call transcription and voicemail workflow behind the conversation matters here. Decide where the summary lands, who can read it, how long it is kept and what happens when the caller shares something the system should not handle.

Your website privacy information should match the real call flow. The OAIC specifically recommends updating privacy policies and notifications with clear information about AI use. A vague statement that says “we may use technology” does not give a caller much help.

Give the caller a real human exit

“Ask for a person at any time” only works if the system has somewhere useful to send the request.

A sensible handover design has three paths:

• Transfer now: connect the caller to an available person.
• Arrange a callback: capture the minimum details and create a clear follow-up task.
• Stop and escalate: end the automated path when the call involves a complaint, distress, an unusual request or information outside the approved scope.

Those are operating choices, not claims about what every business is legally required to automate. The important point is ownership. Someone needs to be responsible for reviewing escalations and checking that callbacks happened.

This is also why an AI readiness review should cover data, access, human ownership and failure handling, not just whether the voice sounds convincing.

A launch checklist for an AI receptionist

Before a customer can reach the system, test these seven items:

1. Identity: The first sentence uses the words “AI assistant” or “AI receptionist”.
2. Purpose: The greeting says what the system can do.
3. Choice: The caller can request a person without fighting through a script.
4. Data notice: Recording, transcription or personal-information handling is explained where relevant.
5. Privacy alignment: The website privacy information describes the real system rather than a generic possibility.
6. Limits: The agent stops instead of guessing when the request falls outside its approved knowledge.
7. Ownership: A named person owns each escalation through to a completed callback.

Then test the awkward paths, not only the polished demo. Use illustrative scenarios such as a caller refusing to give a name, asking for a human immediately, raising a complaint, sharing sensitive information or asking a question the agent cannot answer.

The correct response is not a clever improvisation. It is a clear limit and a clean handover.

If the system is already live

Start with the greeting, because that is where disclosure belongs. Next, make sure the privacy notice matches the call flow and that human requests reach a real person. Check where call summaries go as a separate step.

Listen to a sample of recent calls only if your policies and permissions allow it. Look for points where callers appear confused about who or what they are speaking with. Rewrite the smallest part of the script that creates the confusion, then test again.

Do not wait for a complaint to show that the disclosure was too vague. Identify the AI before it collects details, and make the route to a person obvious.

If you are planning an AI phone system and want those rules designed before it handles a customer, book a short discovery call. Bring the current greeting and your escalation rules. The useful conversation starts there, not with a voice demo.