ChatGPT Prompts for Writing Property Listings (NZ Agents)
Most agents using ChatGPT for listing copy get the same result: something fluent, confident, and completely interchangeable with every other listing on the portal. "Stunning family home nestled in a sought-after location, offering the perfect blend of comfort and style."
The model isn't writing badly. It has nothing to work with.
A listing description is a factual document with a persuasive job. The model can do the persuasive part. It cannot invent the facts, and when you don't supply them it fills the gap with adjectives. Every generic listing you've read is that gap.
Here's how to close it, and how to stay on the right side of the rules while you do.
What the model needs from you
Before any prompt, gather these. It takes two minutes and it's the difference between usable copy and filler.
Bedrooms, bathrooms, car spaces
Floor area and land area, and whether the floor area includes the garage
Year built, construction material, roof type
Heating, insulation, and any recent compliance work
School zones by name
The three things a buyer would actually notice walking in
The one compromise the property has
Who it suits: first home buyer, downsizer, family, investor
Method of sale and deadline
That second-to-last item is the one agents skip, and it's the one that makes copy specific. A house written for a downsizer reads nothing like the same house written for a family.
What not to paste in
Before you paste anything into ChatGPT, know that government guidance advises against inputting personal and client information into external AI tools, and the Real Estate Authority points licensees at exactly that in its own generative AI guidance.
The reason is simple enough: as the REA puts it, Gen AI providers "may well be able to see both your input data and the output". Anything you paste in has left your control.
In practice that means:
Property facts, yes. Bedrooms, floor area, school zones, chattels, method of sale.
Vendor names, contact details, financial position, reason for selling: no. None of that belongs in a prompt, and none of it improves the copy anyway.
Anything from a client file, no. If it would be a privacy problem in an email to the wrong person, it's a privacy problem in a chat window.
Your Privacy Act obligations don't pause because you're using a tool. Write your prompts with the property, not the people.
The prompts
1. The main listing description
The banned-words list matters more than anything else in that prompt. Those specific words are what make a listing read as generic, and the model reaches for them by default.
2. The headline
3. The social caption
4. The buyer database email
5. Rewriting copy that's already too generic
That last prompt is the most useful one on this page. "Would this sentence be equally true of a different house?" is the entire test for listing copy, and the model applies it more ruthlessly than most writers will to their own work.
What the REA actually requires
The Real Estate Authority published guidance on generative AI in November 2024, downloadable as a PDF, and it's worth reading in full if you're using these tools on client work. The short version:
You are responsible, not the tool. The REA is explicit that licensees are ultimately responsible for the real estate services they provide. An error introduced by an AI tool is not a defence.
Skill, care, competence and diligence apply. That's rule 5.1, and the guidance is specific that it covers "the production of customer facing documents such as appraisals and marketing materials". It warns that Gen AI "can create content that may sound and look good but is actually inaccurate or incomplete".
Fact-check everything. The guidance asks that "at the very least" steps are taken to check the "accuracy, relevance, and completeness" of anything a Gen AI tool produced. Separately, the Fair Trading Act prohibits false and misleading statements, which applies to how a property is advertised.
Disclosure isn't explicitly required. The guidance doesn't require you to state that AI was used. Your existing transparency obligations still apply.
Supervisors are on the hook too. Agencies and supervisors "may also be held to account for breaches by salespeople and other staff members under your supervision", which is why more agencies are writing AI policies rather than leaving it to individuals.
Practically, three rules cover most of it:
Every factual claim gets verified against the record of title, the LIM, or your own inspection. School zones especially, because they change.
Never let the model estimate. If you didn't give it the floor area, it will sometimes produce one.
Read the output as though a buyer will quote it back to you at the tribunal. That's the standard.
Used properly, AI is a drafting tool that saves you twenty minutes. Used carelessly, it's a written record of a misleading representation with your name on it.
How to not sound like AI
Even accurate AI copy has a texture that people are increasingly good at spotting. The common tells:
Sentences built as "not just X, but Y"
Threes everywhere: three adjectives, three clauses, three benefits
Em dashes used for emphasis in every second sentence
Openers that restate the question before answering it
A closing paragraph that summarises what you just read
I've written a longer piece on this: 17 AI writing tells that give real estate agents away. Worth reading before you publish anything a model drafted.
The quickest fix is to read the copy aloud. Anything you wouldn't say to a vendor in their kitchen comes out.
What AI still can't do
It can't walk the property. It doesn't know that the afternoon sun hits the deck until four, that the neighbour's tree drops onto the drive, or that the third bedroom is the one that sells the house.
That's your job, and it's the part that makes a listing worth reading. The model handles structure and speed. The specifics, the ones that make a buyer book a viewing, still come from you standing in the house.
Photography works the same way. The camera doesn't know which room matters either.
Want your listings to look as good as they read? See listing media packages · 17 AI writing tells that give agents away
Frequently asked questions
Do I have to tell anyone I used AI to write a listing?
The REA's generative AI guidance does not explicitly require you to disclose that AI was used. Your existing obligations still apply to the content itself. What the listing says has to be accurate and not misleading, regardless of who or what drafted it.
Will Google penalise AI-written listing descriptions?
For listing copy this is mostly the wrong thing to worry about. Your description lives on realestate.co.nz or Trade Me, not on your own website, so your own search visibility isn't really what's at stake. Google's stated position is that it rewards helpful content however it was produced, and acts against spam generated at scale.
The real exposure with AI listing copy isn't ranking. It's the Fair Trading Act.
Can I paste vendor details into ChatGPT?
No. Government guidance advises against entering personal and client information into external AI tools, and the REA points licensees at it. Property facts are fine. Names, contact details, financial position and reason for selling are not. The What not to paste in section above covers this in full.
My agency has its own AI policy. Does that override this?
Yes, follow it. The REA notes that agencies and supervisors may be held to account for breaches by salespeople under their supervision, which is why more agencies are writing formal policies rather than leaving it to individuals.
