In November 2024 the Registrar of the Real Estate Authority issued Generative AI guidance for New Zealand licensees. It is general, non-exhaustive — and still the clearest public statement of how the regulator expects Gen AI to sit inside real estate agency work.
The line that matters most is blunt: you remain responsible for the real estate services you provide. If a Gen AI tool invents a fact, invents a comparable sale, or quietly drops a material LIM issue, that is not a defence under the Real Estate Agents Act 2008 or the Code of Conduct. REA’s media release on 12 November 2024 put it the same way: accountability stays with the licensee, irrespective of the tools used.
That message landed at the right moment. Consumer chat tools are already in listing lounges. Marketing teams are already drafting captions with models. Some salespeople are already summarising LIMs because a 40-page PDF feels like friction. REA did not pretend those habits do not exist. It mapped them onto duties you already have.
This piece is a practical reading for agents, branch managers and agency owners. It is not legal advice. Policies, vendor agreements and tool settings change. When in doubt, read the primary sources, talk to your compliance lead, and get legal advice for your agency’s risk profile.
PropertyLM take
REA did not ban Gen AI. It banned the fantasy that “the model said so” is a compliance strategy. Human oversight, privacy hygiene and a written agency policy are the actual product.
What the guidance is — and is not
REA’s guidance was prepared after feedback from licensees, including Industry Advisory Group representatives. It sits under the purposes of the Real Estate Agents Act 2008. It is guidance, not a new set of rules — but that does not make it optional reading. It maps how existing duties apply when you use tools that create text, images, summaries and “valuations” on demand.
REA is explicit that the document is non-exhaustive, that technology will keep moving, and that agencies should put Gen AI policies in place. It also points licensees toward related guidance from the New Zealand Law Society, the Office of the Privacy Commissioner, and DigiGovt public-service material. In other words: treat Gen AI as a professional-risk domain, not a weekend experiment with the office ChatGPT login.
Primary sources worth bookmarking:
REA — Appraisals (rules 10.2 and 10.3)
What Gen AI is doing in real estate already
REA’s own examples are useful because they sound like Tuesday in a busy office: analysing property data and market trends to support pricing conversations; virtual staging; drafting marketing scripts and emails; simple customer engagement such as booking appointments; answering frequently asked questions; and sourcing publicly available information.
None of those use cases are automatically compliant or automatically dangerous. The risk sits in how you run them. A draft open-home script checked by a salesperson is one thing. An unchecked “appraisal” paragraph that invents a comparable sale is another. A chatbot that books a viewing is one thing. A chatbot that invents a flood-zone answer from a half-read LIM is another.
REA’s framing is commercial as well as regulatory: opportunities exist, and so do compliance and customer-harm risks. The guidance is there to help licensees engage with the technology while meeting obligations — not to romanticise either the tools or the fear of them.
REA in plain NZ English
Gen AI creates; it does not understand. Treat every output as a draft until a licensed human has checked accuracy, relevance and completeness — especially before it touches a vendor, buyer or listing file.
The accountability spine
From a regulatory view, your obligations under the Act and other legislation remain with you even when a technology tool is part of the work. Failure can mean customer harm, reputational damage, complaints and disciplinary action. REA’s wording is hard to misread: the fact that any error was made by a Gen AI tool will not be a defence.
Two further points agencies often under-weight:
Supervisors and agencies can be pulled into the same mess. If salespeople are pasting client files into consumer tools without controls, that is a supervision and systems problem as well as an individual one.
Careful human oversight is vital. REA’s framing is not “don’t innovate.” It is “innovate with safeguards that still meet legal and regulatory requirements.”
That maps cleanly onto everyday Code duties: skill, care, competence and diligence (rule 5.1); accuracy and disclosure (including rules 6.4 and 10.7); confidentiality and personal information (including rule 9.17 and the Privacy Act 2020); and written appraisals that realistically reflect current market conditions with comparable sales support (rules 10.2 and 10.3).
Accuracy and completeness: where offices actually get burned
REA calls out the obvious failure mode: Gen AI can produce content that sounds and looks good while being inaccurate or incomplete. That risk is sharpest for marketing material, appraisals, and summaries of LIMs or builder reports.
Rule 5.1 is the general quality duty. On top of that sit disclosure obligations and Fair Trading Act constraints on false or misleading statements. Improper, negligent or incompetent use of Gen AI can become a serious breach — not because Gen AI is “illegal,” but because the licensee still owns the output that went out the door.
Practical pattern that travels well in NZ offices:
Use Gen AI for first drafts of emails, social captions, open-home scripts, or research questions — not as the final voice of the agency.
Never treat an AI summary of a LIM, builder report, title pack or tenancy file as complete without reading the source.
Keep appraisal opinions as licensee opinions under rules 10.2 and 10.3. An automated estimate can help gather comps; it is not the appraisal. REA’s appraisals guidance is explicit that you can’t solely rely on electronic appraisals or market estimates.
Name a human reviewer on anything client-facing. “Someone looked at it” should mean a named licensee, not a Slack shrug.
Keep a recoverable trail for material outputs: what was drafted, what was checked, what was sent.
REA also reminds licensees that consumer-facing obligations — approved guides, recommending legal advice, getting the right information to the right person — still sit with you even if a tool helped send the message. Automation of delivery does not automate responsibility.
Compliance callout
If you would not email the file to a stranger’s Gmail, do not paste it into a consumer Gen AI prompt. Same risk shape — different UI.
Privacy, confidentiality and the prompt problem
Gen AI involves inputting data into a tool. REA warns that the provider may be able to see both inputs and outputs. That engages Privacy Act duties and Code confidentiality — REA specifically flags rule 9.17. Government guidance recommends against putting personal and client information into external AI tools. Cybersecurity risk sits alongside privacy risk; REA points agencies to CERT NZ material.
The Office of the Privacy Commissioner’s AI guidance is aligned: the Privacy Act applies to AI use in New Zealand; do a privacy impact assessment; prefer human review before acting on outputs; and do not put personal or confidential information into a generative tool unless you have clear assurance it is not retained or disclosed by the provider.
For a listing office, that usually means a hard line on vendor and buyer names and contact details; trust account and financials; unsigned agreements and negotiation notes; full LIM / builder / engineer PDFs into consumer chatbots; and “anonymous” prompts that still include street address plus unique property facts that re-identify the client in a small New Zealand suburb.
At the very least, REA says agencies should turn their mind to obligations, safeguards, and what customer disclosure or consent is required. That is not a suggestion to write a 40-page policy nobody reads. It is a suggestion to write a one-pager people will follow — then train it.
What agencies should put in place
REA’s agency recommendations are short and useful: understand how Gen AI works and what its limitations are; understand the legal, regulatory and professional obligations that apply; and ensure clear policies, processes and procedures — including confidentiality and privacy management.
For individual licensees, REA’s list is equally direct: know your obligations; follow agency policy; take personal accountability; protect client interests; seek advice when unsure.
A workable start for any branch this week:
Inventory every Gen AI tool already in use — including “personal” ChatGPT / Claude / Gemini accounts on work files.
Write (or refresh) a one-page Gen AI policy: allowed tools, banned inputs, review rules, escalation path.
Train the team once, then put the policy in induction.
Separate research/drafting aids from appraisal opinions and disclosure work.
Keep an evidence trail for material client-facing outputs you heavily relied on AI to draft.
Re-read REA’s guidance whenever you add a new tool — the technology will move; your duties will not evaporate.
Where PropertyLM fits
PropertyLM is built for the gap REA is describing: NZ property data you can check, workflows that assume human sign-off, and outputs designed to be reviewed — not “the model said so.” Tools should make diligence faster without pretending diligence is optional.
If you only do one thing after reading this: open REA’s guidance with your branch manager, mark the privacy and accuracy sections, and write the one-pager your team will actually follow. The next articles in this series go deeper on appraisals versus AVMs, LIM privacy, industry readiness, black-box valuations, and a copy-ready policy template.
— PropertyLM.