In this article
Kooky
Builder of Shaka, the payment router that pays every agent their commission on closing date.
About Kooky and Shaka →It is twenty to ten on a Sunday night and the hot water has gone cold. The tenant opens the agency's website, finds the little window in the corner of the screen and types a description of the problem, the address, a mobile number and the hours when somebody will be home. An answer arrives in four seconds. It is polite, it uses the tenant's first name and it promises that the request has been logged.
Nobody at the agency has read the message yet. Software has. That is now an ordinary part of renting, and it raises three plain questions that this guide works through. What is the software doing with what was typed? What does the national privacy regulator expect of the business that put it there? And where does Queensland tenancy law still insist on a particular form, a particular period of notice and a particular way of delivering it, whatever the chat window says? No product is named or compared: the guide describes how this kind of tool works in general and what the published rules say.
Four jobs an assistant does in a rent roll
"Chatbot" covers several different things. In property management the label is applied to at least four tasks, and they do not carry the same weight.
The first is answering enquiries about a listing: is the property still available, are pets considered, when is the next inspection. The second is booking, where the assistant offers inspection times and records a name and a phone number or email address against a slot. The third is maintenance triage: the tenant describes a fault, the software asks follow-up questions, sorts the request by urgency and passes it to a property manager or a tradesperson. The fourth is invisible to the tenant. The assistant drafts a reply, an email or a summary of a long thread, and a staff member edits and sends it.
Related readTenant scoring software: the limits Queensland law puts on rankingThe first three put software in front of the public. The fourth keeps a person between the software and the reader, but it still involves feeding somebody's words, and often their personal details, into a system. The table sets out what each task tends to take in and which published rule it runs into. The descriptions of the tasks are general; the rules in the last column are explained in the sections that follow.
| Task | What the person types | Rule to keep in view |
|---|---|---|
| Listing enquiries | Questions, sometimes a budget or household details | The tool should be clearly identified as automated |
| Inspection bookings | Name, phone number or email, preferred times | Collect only what is reasonably necessary |
| Maintenance triage | Address, the fault, photos, times someone is home | Entry for repairs still needs the notice the law sets |
| Drafting replies | Whole email threads pasted in by staff | A person stays responsible for accuracy |
Where a typed message travels
A message typed into a chat window does not stay in the window. Following it from keyboard to inbox shows where the legal questions sit. The sequence below is a general description of how a hosted assistant works, not an account of any one product.
- TypedThe tenant enters the fault, the address and a phone number in the chat window on the agency's site.
- Sent to a supplierThe text usually leaves the agency's own systems and is processed on servers run by the company that supplies the assistant.
- ProcessedA model reads the text, sorts it and writes a reply. Any conclusion it draws about the person is new information.
- StoredThe conversation is kept as a transcript, by the supplier, by the agency, or by both, under the terms of their agreement.
- Handed onA summary reaches a property manager, and sometimes a tradesperson, who acts on it.
Each step has a counterpart in privacy law. The second step is where the Office of the Australian Information Commissioner, the national privacy regulator, asks whether the information stays within the organisation's control or becomes available to someone outside it. Its guidance says that putting personal information into an AI system may be a use of that information in the first case and a disclosure in the second. The third step matters because, in the regulator's reading, information that a system generates or infers about an identifiable person is itself personal information. The fourth step raises security and the location of the servers. The fifth is where a human being, finally, is answerable for what happens next.
The regulator's guidance on off-the-shelf AI
The OAIC published its "Guidance on privacy and the use of commercially available AI products" on 21 October 2024, and the page records an update on 17 January 2025. It is addressed to organisations that deploy AI systems which were built with personal information or which collect, store, use or disclose it. The regulator says the guidance also addresses products that are freely available, such as publicly accessible chatbots. A real estate agency that buys or switches on an assistant is, in the guidance's vocabulary, a deployer: an organisation that supplies or uses an AI system to provide a product or service.
Related readCan a chatbot read your Queensland sale contract for you?The document opens with five takeaways. Put in plain terms, they are these.
First, privacy obligations apply both to the personal information that goes into an AI system and to what comes out of it, where the output contains personal information. Second, businesses should update their privacy policies and notifications with clear information about their use of AI, and any public-facing tool such as a chatbot should be clearly identified as one. Third, when a system is used to generate or infer personal information, that counts as a collection and must meet the rules on collection. Fourth, information put into an AI system may only be used or disclosed for the primary purpose for which it was collected, unless the person has consented or would reasonably expect the secondary use. Fifth, and expressed as best practice, the regulator says organisations should not enter personal information, and particularly sensitive information, into publicly available generative AI tools.
Beyond the takeaways, the guidance asks for homework before a product is adopted. An organisation should work out how it intends to use the tool and whether the tool is suited to that use, find out what data it was trained on, and read the service terms to learn whether the developer gets access to what is typed in or produced. It says customer-facing products such as chatbots should be carefully tested before they go live, to understand the risk of inaccurate or biased answers. And it warns that this is not a "set and forget" exercise: the product's performance, staff training and monitoring are meant to be reviewed for as long as the tool is in use.
Related readDeepfakes and cloned voices: how property identity checks respondBeing told it is a machine
The simplest expectation in the guidance concerns honesty about who is answering. The OAIC says that, as part of an organisation's obligations on transparency and on collecting information by lawful and fair means, individuals should be made aware that they are interacting with an AI system instead of a human.
For a tenant this is more than a courtesy. A person who believes a property manager has read a message may assume that a human judgement has been made: that the leak has been assessed as minor, that a request to break the lease has been understood, that a hardship explanation has been heard. If the reply came from software, none of that has happened yet. Knowing which it was tells the tenant whether to follow up.
When staff paste a tenant's details into a public tool
The fourth task, drafting, is where the guidance is most pointed. It is easy to picture: a property manager has a long, heated email thread about a bond dispute, copies the whole exchange into a free generative AI website and asks for a calm reply.
The OAIC works through a similar case from another industry. In its example, an insurance company enters a customer's claim details, including sensitive health information, into a chatbot to prepare a report. The regulator's analysis is that by doing so the company is disclosing the information to the owners of the chatbot, which then has to be justified under the rules on secondary disclosure. The guidance adds that once personal information has been put into AI systems, particularly generative products, it will be very difficult to track or control how it is used and it may be impossible to remove.
Related readGenerative AI on a conveyancing file: what Queensland's rules sayPersonal details should stay out of public generative AI tools
The privacy regulator's fifth takeaway is that organisations should not enter personal information, and particularly sensitive information, into publicly available generative AI tools, because of the significant and complex privacy risks. A tenant's medical reason for a late payment, pasted into a free chatbot to draft a reply, is the kind of entry the guidance has in mind.
Where an agency does use a tool for drafting, the guidance points to minimising what goes in. It expects organisations to consider what information is necessary and whether the amount of personal information in a prompt can be reduced without harming the result. A reply about a repair can usually be drafted without the tenant's name, address or bank details in the prompt.
Wrong answers and who checks them
Generative systems make things up. The OAIC's guidance explains why in a sentence: the model predicts the next word or other element on the basis of likelihood, and that process can produce what are known as hallucinations. It adds that such systems can sound human and produce realistic output, which may lead users to overestimate how accurate they are.
In a rent roll, the cost of a confident wrong answer is easy to imagine. A chatbot that tells a tenant an inspection needs only a day's warning, or that a rent payment has been received when it has not, has stated something with consequences. The guidance places the duty on the organisation. It says a human user should be responsible for verifying the accuracy of personal information obtained through AI and should be able to overturn decisions. It asks that staff be trained to understand the limits of the system and to anticipate when it may mislead. It mentions disclaimers that clearly communicate the limits of a system's accuracy. And it treats the use of AI in decisions with a legal or similarly significant effect on a person's rights as likely to be a high privacy risk activity, calling for particular care.
Related readA rent figure from software: what Queensland's Act asks before it runsOne further point is easily missed. The regulator says information can be personal information whether or not it is true. If an assistant records in a tenant's file that the tenant "admitted" causing damage, and the tenant said nothing of the kind, the false entry is still information about that person, and the accuracy obligations still apply to it.
Which agencies the Privacy Act reaches
The OAIC's guidance explains obligations under the Australian Privacy Principles, the rules at the centre of the Commonwealth Privacy Act 1988. Those principles do not bind every business. In a determination dated 1 April 2026 concerning a rental application platform, the Privacy Commissioner recorded that the small business exemption did not apply because the company's turnover was above the $3 million threshold. The same threshold explains why a small independent agency may sit outside the principles while a large group, or the technology supplier behind its chat window, sits inside them.
Two consequences follow for a tenant. The supplier of an assistant may be bound by the principles even where the agency using it is not. And the agency, whatever its size, is bound by Queensland's tenancy legislation, which contains its own rules on personal information. The next sections turn to those.
What a chat window may not ask an applicant
Assistants are sometimes used at the front of the letting process, collecting details from people who want to apply. Here the Residential Tenancies Authority, the state's tenancy authority, usually called the RTA, sets out limits that apply to every property manager and owner in the state.
According to the RTA's page on the application process, applications are made on a standard form, Form 22 for general tenancies, or on a form that complies with the Residential Tenancies and Rooming Accommodation Act 2008. A property manager must offer at least two ways to submit an application, and one of them must not be restrictive. The RTA counts as restrictive a method that requires the applicant to share personal information through an online platform with a third party that is not the agent, or one that requires the applicant to pay.
Related readVirtual staging and AI-edited photos: what the law allowsThe same page lists what may not be requested: legal action taken by the applicant, including matters before the Queensland Civil and Administrative Tribunal; notices to remedy a breach; rental bond history; and statements of credit accounts or bank statements showing transactions. Supporting documents are capped at two in each of three categories, covering identity, ability to pay the rent and suitability. The RTA gives a maximum penalty of 20 penalty units for breaching the rules on requested information.
A conversational assistant is good at asking one more question. That is exactly the habit these rules restrict. A chat flow that asks an enquirer whether they have ever been to the tribunal is asking for something the standard form may not. The RTA's fact sheet for property managers adds a detail that matters for transcripts: personal information about unsuccessful applicants must be destroyed within three months of the start of the successful tenant's agreement. Whether that information sits in a filing cabinet or in a chat log held by a supplier, the fact sheet does not make an exception for the format.
The notices an automated message cannot replace
Tenancy law in Queensland runs on written notices. The RTA publishes approved forms for the main ones, and the law attaches a minimum period to each. A chat message, however clear, is not one of those forms.
Entry is the best example, because it is where maintenance triage ends up. An assistant can log a dripping tap and find a plumber who is free on Thursday. It cannot, by sending "our plumber will attend Thursday morning", satisfy the requirement to give the tenant an Entry notice, known as Form 9. The RTA's page for that form sets out the minimum notice for each reason for entry.
| Reason for entry | Minimum notice | Extra condition |
|---|---|---|
| Routine inspection | 7 days | Not within 3 months of the previous one, unless agreed |
| Routine repairs or maintenance | 48 hours | None listed |
| Checking repairs that were done | 48 hours | Within 2 weeks of the repairs |
| Smoke alarms or safety switches | 48 hours | None listed |
| Showing a buyer or new tenant | 48 hours | A reasonable time since the last entry for that reason |
| Valuation | 48 hours | None listed |
| Emergency | None required | None listed |
The RTA page also lists entry at a time agreed with the tenant, and entry without notice to protect the property from damage.
The table shows why the tenant's own request does not settle the matter by itself. Entry by agreement is on the RTA's list, so a tenant who says "come any time on Thursday" has agreed to something. But an assistant that proposes a time on the agency's behalf, receives no answer and treats silence as consent has not obtained an agreement, and the 48 hours have not been given.
The way a notice travels matters as much as its content. The RTA's fact sheet on allowing time when serving notices says a notice may be sent by email only where the tenancy agreement provides for it, and that an emailed notice is taken to be received when it enters the recipient's email server. It explains that periods are counted in clear days: the day the notice is served is not counted, and the next step cannot be taken until the day after the last day of the period. A notice period expires at midnight. Posted notices need extra days for delivery. The fact sheet puts the responsibility for checking the right timeframe on the person serving the notice.
For an automated system, each of those sentences is a place to go wrong. A scheduling tool that counts the day of sending as day one produces a notice that is a day short. A tool that emails a tenant whose agreement does not provide for email has not served the notice at all, on the RTA's reading. The same discipline applies to the other approved forms, including the Notice to remedy breach, Form 11, and the notices that bring a tenancy to an end. Software can prepare them and remind a manager that one is due. Under the published rules, what counts is the right form, served in a permitted way, with the full period allowed.
A chat window can take the message at any hour. The law still measures notice in clear days, on an approved form, sent by a method the agreement allows.
What a tenant can ask about automated handling
None of this requires a tenant to refuse the chat window. It does give a tenant, a buyer or an applicant a short list of reasonable questions, each of which follows from something the regulator or the RTA has published. They are general questions, and the answers will depend on the agency and the tool.
- Is this reply written by software or by a person? The OAIC's guidance says individuals should be made aware when they are interacting with an AI system.
- Which company supplies the assistant, and does it receive what is typed? The guidance asks organisations to know who has access to information put into the product.
- Where does the privacy policy describe the use of AI? The regulator expects policies and collection notices to be updated with clear information about it.
- Is the conversation kept, by whom and for how long? For rental applicants, the RTA's destruction period for unsuccessful applications is three months from the start of the successful tenancy.
- Is anything typed here used for a purpose other than answering the request, such as training a model? The guidance says that in many cases it will be difficult to show that a secondary use for AI-related purposes is within a person's reasonable expectations.
- Who checks what the assistant tells people, and can a person correct it? The guidance says a human should be responsible for accuracy and able to overturn decisions.
- If entry to the home is needed, will a Form 9 follow? The chat log is not the notice.
A tenant who prefers not to use the window at all has one firm rule to rely on at the application stage, where the RTA requires a second, non-restrictive way to apply. For day-to-day messages during a tenancy, the pages read for this guide do not set out an equivalent rule; the method of communication is generally a matter for the tenancy agreement and the agency's own practice.
What the published rules leave open
The sources used here are clear about principles and quiet about several practical points, and it is fair to say so.
The OAIC's guidance is guidance. It explains how the regulator reads existing obligations and what it regards as best practice; it does not create a licence or an approval scheme for chatbots, and it gives no worked example from real estate. Its examples come from insurance, recruitment, health and training.
The RTA's pages on applications, entry and serving notices do not mention chatbots or artificial intelligence. They state what must be done and in what time. Applying them to automated tools, as this guide does, is a matter of reading the requirement and asking whether a given message meets it. Where a dispute turns on whether a notice was properly given, the answer depends on the facts and on the tenancy agreement.
Kooky, from Shaka
Kooky edits Queensland Estate and builds Shaka, the payment router he made for Queensland property professionals. One payment comes in, and every agent, agency and party in the deal receives their signed share on closing date.
