Imagine my husband Mark having a perfectly ordinary conversation with ChatGPT about a tool he’s thinking of buying.
For the purposes of this example, let’s say it’s a cordless drill. There is probably a more precise name involving several letters, electricity usage and a reason the one we already own won’t do.
He starts the conversation with:
“I want to put up some shelves in the garage. What sort of drill do I need?”
ChatGPT asks what the walls are made of, what he already owns and how much he wants to spend. They discuss a couple of options. He asks whether the more expensive one is worth it.
We have now progressed from “put up some shelves” to “invest in the right equipment”, which I understand is an important stage of the process.
Now imagine a clearly labelled ad appears for a drill that looks promising. He clicks it and chooses to chat with the retailer’s AI assistant.
“Will this work with the batteries I already have?”
“Does it come with the bits I need?”
“Could I collect it on Saturday?”
He can ask the questions that stand between that looks interesting and yes, that’s the one.
That is the idea behind Sponsored Agents, which OpenAI announced on 16 September. The company is testing them with select US advertisers: after clicking an ad in ChatGPT, a person can choose to enter a separate, clearly labelled conversation with an AI representing the business.
The distinction matters. The original ChatGPT conversation is helping him explore his options. The sponsored conversation represents a particular seller.
You have gone from discussing what to buy to asking a shop assistant about something on the shelf. Knowing who that assistant works for helps you weigh the answer.
But who actually sees the ad?
OpenAI’s current published policy says ads are being tested with logged-in adults in the United States on the Free and Go tiers. Pro, Business and Enterprise subscriptions are currently ad-free.
An ad may appear when there is a relevant sponsored product or service based on the person’s current conversation.
People can turn off personalisation, and OpenAI says their ChatGPT conversations remain private from advertisers and are not sold to them.
What OpenAI has not explained publicly is whether advertisers will have the familiar targeting controls they use elsewhere – location, age, previous website visits or other audience categories.
For an agent, that distinction could be important. The interesting possibility is not simply targeting everyone in a postcode.
It is appearing while someone is already talking to AI about moving, comparing suburbs or finding a home that suits the next stage of their life.
That could be a much stronger signal of intent than fitting into a broad demographic audience. It is also still only a possibility.
OpenAI has not announced property-specific targeting or an Australian rollout, so there is nothing for you to do today.
But it is worth understanding where the value might eventually lie.
Imagine the purchase under discussion is a home.
Same desire to ask a few questions before contacting someone. Considerably more at stake than another addition to the garage.
Let’s say a buyer is already talking to ChatGPT about moving. They are weighing up suburbs, commuting times and whether downsizing will require finding new homes for half their furniture.
In a possible property version of this advertising format, they might see a relevant listing ad and choose to ask the agency’s AI assistant about it.
“What are the body corporate fees?”
“Is there a floor plan with measurements?”
“Can I inspect after work?”
To be clear, that property example is hypothetical; OpenAI’s announcement does not establish a property advertising product or an Australian rollout of Sponsored Agents. But it may give us something concrete to think about.
The buyer could arrive at the first human conversation having already worked through several practical questions.
And, with their permission and a well-designed handover, the agent could arrive knowing what matters to them.
For example, if someone has explained that they are downsizing, need room for a particular dining table and can only inspect after five, there is an opportunity for you to begin with: “I’ve checked the dimensions, and I can show you through on Thursday.”
You have something to work with straight away (and they are not texting you, “When’s the inspection?”…)
Of course, there is a difference between supplying a floor plan and answering, “Do you think the owner would accept $50,000 less?”
An assistant might have access to the first answer. The second belongs in a human conversation.
The same thing might apply in property management. An assistant could explain the published inspection process or provide a link to an application.
A prospective tenant with unusual circumstances may need someone to listen, ask another question and use their judgement.
This is where conversation design becomes important. How much information helps someone feel comfortable taking the next step? At what point would they prefer a person? And does the technology make that person easier to reach?
I would want the option to say, “Thanks, I’d like to speak to someone now,” and have that lead somewhere useful.
(We have all encountered a helpful little chat bubble that eventually turns out to be a very decorative locked door…)
There is also something worth noticing about the drill example. The retailer gets to hear what is holding up the purchase.
In the drill case, part compatibility matters. Collection time matters. The customer may barely mention the feature that took up most of the ad.
Property enquiries contain those clues too. A question about garage dimensions might be about storage, a hobby or whether someone can open the car door comfortably. An inspection-time request might reveal a shift worker whose schedule makes every Saturday open home impossible.
There is room for AI to help capture and organise those details. There is also room for an agent to recognise their significance and ask a better follow-up question.
That combination interests me.
You can start preparing for it with the enquiries your business already receives. Take ten recent questions, remove identifying details and look at what people needed before they felt ready to proceed.
Which answers were easy to find? Which required a callback? Where did the conversation reveal something useful that the original enquiry form never asked?
Then try this in your approved AI tool:
Here are ten anonymised enquiries our real estate business received.
For each one, identify:
- The question the person explicitly asked.
- Any possible underlying concern, clearly labelled as a hypothesis.
- The verified information we would need to answer accurately.
- One useful follow-up question a team member could ask.
Use only the information provided. Flag missing facts rather than filling them in.
Finally, identify recurring questions we could answer more clearly in our advertising, listing information or enquiry process.The result might help you improve a listing page, prepare for a callback or notice a gap in your information. Those are useful gains while the advertising experiment is still unfolding.
And if conversational ads do become part of property marketing, you will have a much better understanding of what people want to talk about.
Happy Hunting 🚀
P.S. Want to put AI to work in your real estate business?
Transform is my practical 30-day AI challenge for real estate professionals. It isn’t an always-open course. I open it in live cohorts so we can take it one useful step at a time together – building confidence, improving real workflows and keeping human judgement where it belongs.
This column first appeared on Samantha McLean’s Substack, samanthamclean.substack.com.