In this article
Kooky
Builder of Shaka, the payment router that pays every agent their commission on closing date.
About Kooky and Shaka →A buyer who hands a contract to a law firm sees letters, a searches list, a settlement statement and, at the end, a bill. The buyer does not see how each of those was produced. In Queensland, where conveyancing is carried out by solicitors and their staff, the question of whether a text generator helped draft a letter or summarise a body corporate record is now a practical one, and two sets of published guidance answer most of it.
The first comes from the courts. Queensland's courts and tribunals have issued two guidelines on generative artificial intelligence, one for people who are not lawyers and one for judicial officers, and three courts have added practice directions on the accuracy of what lawyers put before them. The second comes from the profession: the Queensland Law Society's Guidance Statement No. 37, Artificial Intelligence in Legal Practice. This guide reads those documents in turn, then sets out what they mean for someone whose sale or purchase is being handled by a firm that uses such tools, and which duties remain with the solicitor whatever the software does.
Queensland Courts guidelines for non-lawyers and for judicial officers; Queensland Law Society Guidance Statement No. 37.
Why a property client should care
Most conveyancing never goes near a courtroom, so court guidelines can look remote. They matter for two reasons. They are the plainest official description in Queensland of what these tools do and where they fail, written for a general reader. And they show the standard a document must meet if a transaction does end in a dispute, whether over a deposit, a termination or a defect.
Related readAI-drafted condition reports: what Queensland tenancy law still asksThe Law Society's statement matters more directly. It is addressed to the people running the file, and it deals with the things a client actually has at stake: the confidentiality of the documents handed over, the accuracy of the advice that comes back, who checked it, and what appears on the invoice.
Neither document bans generative AI. Both describe conditions for using it. The reader's interest is in knowing what those conditions are.
Two court guidelines issued on the same day
The guidelines carry the titles "The Use of Generative Artificial Intelligence (AI): Guidelines for Responsible Use by Non-Lawyers" and "The Use of Generative AI: Guidelines for Judicial Officers". Both state that they are a revised version issued on 15 September 2025 and that the original was issued on 13 May 2024. Both are published on the Queensland Courts website, on a page titled "Using Generative AI".
Each names the same nine bodies: the Supreme Court, the District Court, the Planning and Environment Court, the Magistrates Courts, the Land Court, the Children's Court, the Industrial Court, the Queensland Industrial Relations Commission and the Queensland Civil and Administrative Tribunal. That list covers every forum in which a Queensland property dispute is likely to be heard, from a tribunal claim to a Supreme Court application.
The table shows how the main Queensland documents divide the subject between them.
| Document | Written for | Date | Main subject |
|---|---|---|---|
| Guidelines for responsible use by non-lawyers | People running their own case and lay helpers | Revised 15 September 2025 | Limits of chatbots, confidentiality, checking |
| Guidelines for judicial officers | Judges, magistrates, tribunal members and their staff | Revised 15 September 2025 | Permitted uses, and material received from court users |
| Practice directions on accuracy of references | Lawyers and parties filing submissions | Issued in 2025 | A named person answers for every reference |
| Guidance Statement No. 37 | Solicitors and law practices | 31 May 2024 | Competence, confidentiality, supervision, costs |
What the guideline for non-lawyers explains
The guideline for non-lawyers is aimed at self-represented litigants, McKenzie friends, lay advocates and employment advocates. Its opening warning is that generative AI is not a substitute for a qualified lawyer and cannot give advice tailored to a person's situation.
Related readAI suburb forecasts and growth scores: what an agent may pass onIts explanation of the technology is the part worth reading even for someone with no case on foot. A large language model, it says, analyses a large amount of text to predict the most probable next word, in the way an autocomplete function does. The output is the most likely combination of words, which is not the same thing as the correct answer. The guideline adds that such a model has no understanding of what its words mean and no conception of truth.
From that follows a list of things a chatbot can reasonably help with and things it cannot do. On the helpful side: identifying and explaining laws that may be relevant, organising facts into a clearer structure, suggesting headings and layout, and commenting on grammar and style. On the other side, the guideline says a chatbot cannot understand a person's particular facts, cannot predict the outcome of a case, and cannot guarantee that what it produces is relevant, accurate, complete, current and unbiased.
Two warnings stand out. The first concerns invented material. The guideline says chatbots can make up cases, citations and quotes, or refer to legislation and texts that do not exist, and that the invention can look as though it came from a real source. It uses the usual term for this, "hallucinating". The second concerns training data: publicly available chatbots draw on internet sources and have limited training on current Australian law and court procedure, so an answer may reflect the law of another country or an earlier year.
Related readAI-written listing descriptions: who answers for the errors?The guideline names the products it has in mind, including ChatGPT, Claude, Google Gemini and Microsoft Copilot, and points readers to authoritative sources to check against: the Australasian Legal Information Institute, Queensland Judgments and the Queensland Legislation website. It also refers readers to Legal Aid Queensland.
Three of the guideline's numbered parts carry direct instructions.
On confidentiality the instruction is absolute: private, confidential, suppressed or legally privileged information is not to be entered into a generative AI chatbot. The reason given is that some chatbots retain what users type, and that the material could then be repeated in answers to other users. For a property client the point is easy to translate. A contract of sale, a bank statement supplied for a finance approval and a copy of a passport are all private documents.
On accuracy the guideline places responsibility on the person who files the document. Whatever a chatbot produced, the person relying on it is responsible for ensuring that the information given to the court or tribunal is accurate, and must check it first, with a lawyer where possible or against public legal resources.
On consequences it is specific. If a document contains non-existent citations or inaccurate AI-generated information and a hearing is delayed as a result, a costs order may be made against the person who filed it.
The guideline gives separate attention to affidavits and witness statements. Particular caution is needed, it says, because those documents must reflect the person's own knowledge and words. The remaining parts deal with bias in training data, copyright and plagiarism, and following ordinary cyber-safety practice.
Related readTenant scoring software: the limits Queensland law puts on rankingWhat judges and tribunal members are told
The guideline for judicial officers repeats much of the same explanation and then goes further in three directions.
It limits what a judicial officer may do with the tools. Generative AI may be a useful secondary tool for research or preparatory work, the guideline says, but AI tools should not be used for decision-making or to develop or prepare reasons for decision. The reasoning has to be the judicial officer's own. The guideline also states that a judicial officer is personally responsible for material produced in his or her name, and that anything typed into a public chatbot should be treated as published to the whole world.
It describes what to expect from court users. Legal representatives, it says, are responsible for the material they put before courts and tribunals and have a professional obligation to ensure it is accurate and appropriate. Where it appears that a chatbot was used to prepare a document, the guideline suggests it may be appropriate to ask about that and about what checks were made. It lists signs to watch for: cases that sound unfamiliar or carry unfamiliar citations, submissions that do not fit the judicial officer's general understanding of the area, and elaborate language that does not match the way the person otherwise writes or speaks.
And it looks past text. The guideline observes that AI tools are now used to produce fake material, including text, images and video, and says judges should be aware of the challenges posed by deepfake technology, adding that courts have always had to deal with forgeries of varying sophistication. It suggests judges also consider whether an expert who used AI in preparing a report should be required to say precisely how.
Related readCan a chatbot read your Queensland sale contract for you?Practice directions: one named person answers
Guidelines advise. Practice directions bind those who appear. In an article dated 13 October 2025, the law firm McCullough Robertson described three practice directions in identical terms, each titled "Accuracy of References in Submissions": Supreme Court Practice Direction 5 of 2025, District Court Practice Direction 12 of 2025 and Planning and Environment Court Practice Direction 7 of 2025.
As the firm sets them out, written submissions must identify at their end a named individual legal practitioner as the responsible person. The name of a firm is not enough. That person verifies the accuracy and relevance of every reference to legislation, authorities and other sources. For oral submissions, the person speaking is the responsible person. Self-represented litigants are asked to endeavour to ensure their references are accurate.
The same article lists what may follow a failure: a hearing may be adjourned, costs may be ordered against the party responsible, a costs order may be made against the practitioner personally, and the practitioner may be referred to the Legal Services Commissioner. The firm also notes an amendment to Supreme Court Practice Direction 14 of 2024, on expert evidence in criminal proceedings, under which an expert who used generative AI is to disclose the program, how it was used and the records of what went in and what came out. Its article cites three Queensland decisions from 2025 in which AI-generated material came before a court or tribunal.
The Law Society's Guidance Statement No. 37
The Queensland Law Society's statement, dated 31 May 2024, is the profession's own document. It is organised around the duties a solicitor already owes and cites rules 4, 5, 9, 17, 19 and 37 of the Australian Solicitors Conduct Rules, with a further reference to rules 10 and 11 on conflicts of interest. Its starting position is that using AI changes how work is produced and leaves those duties where they were.
Related readDeepfakes and cloned voices: how property identity checks respondOn competence, the statement says practitioners must make sure that they and their staff understand the fundamental capabilities, limitations and appropriate application of the tools in use. That includes being able to tell when a product they already subscribe to contains an AI component, keeping up with frequent changes in the technology, and retaining enough legal knowledge to recognise an error when the tool makes one.
On confidentiality, the statement asks for a risk assessment before any client data is submitted to a tool. The questions it lists are concrete: where the data goes, how it is stored, who can access it, and whether the provider gives a confidentiality commitment beyond its standard terms of service. It asks practitioners to consider the effect on legal professional privilege and whether a provider's terms are incompatible with an intention to keep material confidential. It also asks whether the client's own obligations restrict what can be done with the data, naming copyright, the Privacy Act, licences and non-disclosure agreements. Data obtained under compulsion, it says, should not be uploaded unless the owner of the data has been informed and any objection considered.
A property file holds a good deal of that kind of material. A seller's disclosure documents, a tenant's details in a lease, a body corporate's records and a lender's loan offer all arrive from someone else, sometimes under terms of their own.
Supervision, and who checks the output
The statement treats supervision at two levels: the firm's control over which tools are used, and a lawyer's check of what a tool produces.
At the first level, a law practice is expected to make clear that any use of AI in legal work or in the firm's systems happens only with approval, and to state the conditions attached to each approved product. The practical effect is that a clerk or paralegal is not meant to choose a tool unilaterally and paste a client's document into it.
At the second level, the statement says each discrete piece of work produced with an AI system should be checked. How much checking depends on how complex the task is, how long the tool has been in use, and whether it was built for the purpose or adapted from a general product. The person reviewing must know the area of law well enough to detect errors. Because products are revised continually, the statement adds, periodic auditing is still needed after a tool has been approved.
On accuracy the statement uses the same vocabulary as the courts. It identifies hallucination, meaning output in which parts look plausible but have been invented, as a key risk, and says current AI cannot be relied on to validate its own output. One sentence carries the weight of the whole document: an error in a technical system may explain, but does not absolve, negligent service to a client.
Where AI-generated material is used in court proceedings, the statement says reasonable efforts should be made to disclose the AI involvement and to outline the controls used to ensure accuracy.
The court directions do not ask whether a machine was used. They ask which person stands behind the page.
What the statement says about the bill
The costs section is the part a client can test against an invoice. It separates the firm's own overheads from work done on a matter, and the table sets out the distinction as the statement draws it.
| Item | Treatment in the statement |
|---|---|
| Due diligence on a tool, staff supervision, licence fees per user or per year | Practice overheads: not billed to clients |
| Time actually spent on a client's matter | Recorded and billed in the usual way under time billing |
| A fee charged to the firm per matter | May be a disbursement if the costs agreement defines it |
| Time entries raised to match manual effort | Not permitted: entries must remain accurate |
The last row deserves a plain restatement. If a task that once took an hour now takes ten minutes, a firm billing by time records ten minutes. The statement says time entries may not be adjusted upwards to reflect how long the work would have taken by hand, and that fees must remain fair and reasonable.
It also suggests that a fixed fee may suit work with a significant AI component better than time billing, and advises firms that charge this way to keep records of the indirect production costs of the system so that the fee can be justified. Conveyancing is frequently quoted as a fixed fee, which makes this passage the one most likely to apply to a residential file.
Software subscriptions are the firm's overhead, not the client's cost
Guidance Statement No. 37 treats licence fees charged per user or per year, and the time spent vetting and supervising a tool, as practice overheads that are not billed to clients. Only a fee incurred per matter may be passed on, and only where the costs agreement provides for it.
What a client is told, and what can be asked
The Law Society's statement ties disclosure to significance. It recommends that a practice disclose the AI tools it uses in its retainer agreement or letter of engagement. Where AI undertakes substantial parts of work that would ordinarily be done manually, it says the client should be told of that fact, of the types of data the system will be given access to, and of how the relevant risks will be managed.
That gives a client a short list of questions that the published guidance itself supports. None of them is hostile, and a firm that has done the work the statement describes will have the answers to hand.
- Does the engagement letter or costs agreement mention AI tools, and which ones?
- Will any document supplied for this matter be entered into an AI system, and where is that data stored?
- Who reviews work produced with a tool before it is sent, and is that person a solicitor?
- If a per-matter technology fee is charged, where does the costs agreement provide for it?
The answers will differ from firm to firm, and the statement leaves room for that. A tool built for legal work and run under a confidentiality agreement is a different proposition from a free public chatbot, and the statement's risk assessment is designed to tell the two apart. What the documents do not offer a client is a right to insist that no tool be used at all. They describe how a firm is expected to behave, and a client who wants a particular arrangement can raise it when the engagement is agreed.
The Legal Services Commission's place
The practice directions name the Legal Services Commissioner as the destination for a practitioner referred over inaccurate references, which places the Commission at the disciplinary end of the system. Research for this guide did not locate a guidance document on artificial intelligence issued by the Queensland Legal Services Commission itself. That records what could be found and read in August 2026 and nothing more.
In practice it means that the detailed expectations for Queensland solicitors are set out in the Law Society's statement and in the courts' directions, and that the conduct rules those documents cite are the ones against which a complaint would be measured.
What remains the solicitor's own responsibility
Read together, the documents allocate very little to the machine. Five duties stay with the solicitor in every version of events they describe.
Accuracy is the first. The courts' guidelines put it on whoever files a document, the practice directions put it on a named practitioner, and the Law Society's statement says a system error does not excuse negligent service. In a conveyance that covers the dates in a contract, the reading of a title search and the figures in a settlement statement as much as any citation.
Confidentiality is the second. The decision to put a client's document into a tool is the practitioner's, made after the risk assessment the statement describes, and the consequences for privilege follow that decision.
Supervision is the third. A firm approves its tools and sets conditions, and a person with knowledge of the area checks each piece of work.
Honest charging is the fourth: accurate time entries, fair and reasonable fees, and overheads kept off the client's account.
Judgment is the fifth, and the hardest to delegate. The statement lists among a practitioner's continuing responsibilities the duties to the court, the avoidance of conflicts and the detection of fraud and money laundering. Those are the points in a property transaction where something looks wrong and a person has to decide what to do about it. The guideline for non-lawyers explains why a text generator is poorly placed to do so: it predicts likely words and has no conception of whether they are true.
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.
