Property buyers have already “spoken to three agents” before they call you

The conversation that prompted this

A client told me recently that one of his prospects had already spoken to three agents before he picked up the phone. He did not mean competitors. He meant ChatGPT, Perplexity and Google’s AI Overview.

I have thought about that line a lot since, mostly because of what it says about the sequence. The buyer had not replaced anything. He had added a step, and that step happened somewhere none of us can see, before he was a lead in anyone’s CRM.

The old path still works: browse a portal, shortlist, call, inspect, decide. There is just something in front of it now.

What the Australian data actually shows

The behaviour is real and it is local, not an American import.

Australian National University researchers, surveying more than 3,500 adults, found 48.6% of Australian adults have used generative AI at least once, with adoption much higher among younger Australians, the tertiary-qualified, and people living in capital cities. That last group is your first home buyer market almost exactly.

Google launched AI Mode in Australia on 8 October 2025. Google’s own note on the launch says early testers were asking questions nearly three times the length of a traditional search. That matters more than the launch itself. A three-word search tells you almost nothing about intent. A forty-word question about stamp duty on a Tarneit land purchase for a first home buyer tells you everything, and none of it reaches you.

What buyers use AI for, and what they do not

Buyers are using AI for the “help me understand this” work. Summarising options, comparing suburbs, decoding a contract of sale, working out what body corporate fees actually cover.

They are not using it to choose a developer. The best available consumer data, from Bain’s September 2025 survey of 1,500 US online buyers, found about half trust generative AI for initial research and product comparisons, and considerably less for the decision itself. Semrush’s December 2025 survey of 1,030 US consumers found 86% verify what an AI tells them somewhere else before buying, and 68% of those verify on Google. Both studies are American, so treat the exact figures as indicative rather than local.

This is mostly good news. An AI assistant is doing the education that used to eat the first fifteen minutes of your sales consultant’s first call.

It has one uncomfortable consequence. Buyers now arrive further along than they used to, carrying assumptions you did not give them and cannot see. If your team’s opening script still assumes a blank slate, you will sound behind the person you are talking to. That is a briefing problem, not a marketing problem, and it is cheaper to fix.

Why buyers will not tell you they used AI

No buyer describes ChatGPT as an agent. They say “I asked ChatGPT” or “I’ve been using AI”. Calling software an agent is industry and vendor language, and my client used it that way because he works in this industry. Buyers do not.

The practical problem is not the vocabulary, though. It is that buyers will not mention it at all. Nobody volunteers “I compared you against two other estates in a chatbot last night”. It does not feel like a research source worth naming, and it is a slightly odd thing to admit to a salesperson.

So your consultants have to listen for the tells instead. “I read that the titles were delayed.” “I’ve been looking into it.” “Someone said that estate has better sized blocks.” Those are the moments to ask where, lightly. It is the cheapest AI research you will ever do, and it tells you what assistants are saying about your estate before any tracking tool does.

The language will probably catch up eventually. Deloitte Digital’s Australian research, covering 3,000 consumers, found 40% have already used an AI agent and 50% would trust one to search, compare and transact on their behalf. The capability is arriving faster than the words for it.

How much traffic does AI actually send? Very little

Anyone telling you otherwise is selling something.

Conductor’s benchmark analysis of 13,770 domains across 3.5 million prompts, sampled in May and September 2025, found AI assistants sent about 1.01% of real estate website traffic, of which ChatGPT accounted for 95.2%. AI Overviews appeared on 4.5% of page-one real estate queries. Blog content was cited nearly twice as often as news pages. That benchmark is not Australia-specific, so treat it as the shape of the thing rather than your number.

So the click volume is small. The reason to care is what happens without a click.

Pew Research Center analysed the browsing behaviour of 900 US adults across 68,879 Google searches in March 2025. Where an AI summary appeared, users clicked a traditional search result on 8% of visits, against 15% where no summary appeared. They clicked a link inside the summary on 1% of visits. They ended the browsing session entirely on 26% of pages with a summary, against 16% without.

The research is happening, it is shaping the shortlist, and most of it never shows up in your analytics. You are not losing traffic to AI so much as being discussed without being visited.

The portal followed the buyer. Your website did not.

On 12 February 2026, REA Group launched a realestate.com.au app inside ChatGPT. A logged-in user can search REA listings conversationally and get photographs, property details, pricing and agent details without leaving the chat. REA’s Jonathan Swift put it plainly: “Some property buyers are beginning to use AI tools in their property journey and we’re seeing a higher uptake with younger property seekers.”

REA did not wait for buyers to come back. Faced with them researching somewhere else, it moved the listing into the conversation.

The portals remain central, incidentally. PropTrack’s Buyer Impact Model, which analysed more than 1.3 million Australian sales between August 2023 and November 2025 and was independently validated by Deloitte, found buyer engagement on nine in ten listings that sold. Eventual buyers spend almost seven cumulative hours on listings and view 28 times more property images than people who do not buy.

So the portal did not lose the buyer. It followed them. The question worth sitting with is whether your estate’s own website, brochure copy and suburb pages have followed them anywhere at all, or whether they are still built entirely for someone who arrives via a Google click.

What to do about it

1. Make your basics consistent everywhere

Estate name, location, builder panel, lot sizes, price ranges, school catchments, stage releases, display village hours. If those are thin, contradictory or buried in a PDF, that is exactly what gets dropped or garbled when an assistant summarises you. This is unglamorous and it is the highest-return work available.

2. Write pages that answer one question each

An assistant retrieves passages, not pages. A page that plainly answers “what does buying off the plan mean in Victoria” will be used. A page of brand copy about visionary community living will not. Same for a buyer skimming on a phone at a display village, which is the point. Content that reads well to a person and extracts cleanly for a machine is largely the same content.

3. Be honest about what AI optimisation actually involves

Google’s developer documentation is unusually direct on this. It states there are “no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary”, that there is “no special schema.org structured data that you need to add”, and that “you don’t need to create new machine readable files, AI text files, or markup to appear in these features”.

FAQ rich results also stopped appearing in Google Search from 7 May 2026, and Google removed the documentation for them in June. Plenty of agencies are still selling FAQ schema on the promise of a search feature that no longer exists.

Which means most of what is marketed as AI optimisation is SEO fundamentals with a new label: crawlable pages, clear information architecture, accurate structured data, content that answers real questions, and consistent facts about your business across every place they appear. That should be reassuring. It is work you can specify and check.

4. Brief your sales team for a buyer who has already done homework

The first human conversation is now the second stage of research, not the first. Your consultants’ value is local knowledge, judgement and negotiation, which is the part an assistant does badly. Give them the five things AI commonly gets wrong about your estate and let them correct the record. It is a good opening, and it is genuinely useful to the buyer.

Why you should not cut Google Ads and Meta

This is the most common misread of everything above, and it is an expensive one.

Google Ads and Meta do something AI-assisted research does not. They put you in front of a buyer who is already in market, at the moment they search “house and land packages Melbourne”. That is measurable demand at the point of intent. Across the same ten estates, paid search, paid social and Performance Max together drove 55.5% of all key events in the last 90 days, comfortably the largest source of qualified enquiry. AI-assisted discovery sits in front of that. It does not replace it.

What has changed is the job of the landing page. A page built only to convert a click now has a second job: being clear and specific enough to be summarised accurately if it is ever pulled into someone’s research. Direct answers rather than vague brand copy. Consistency between what the ad promises and what the page says.

If your Google Ads are not producing qualified enquiry today, AI is almost certainly not the reason. We wrote separately about the most common reasons property developer campaigns waste budget, and it is usually keyword controls rather than anything to do with AI.

How to measure this, and what we found

Start with GA4, which now does more than most people realise. Google has added an AI Assistant channel group. Open Reports, then Acquisition, then Traffic acquisition, set the dimension to Session primary channel group, and it is sitting there next to Organic Search and Paid Search. Most estate reporting templates were built before it existed, so it is probably missing from yours.

Across ten estate websites we manage, AI assistants accounted for 0.42% of sessions over the 90 days to 22 September 2026. That is 410 sessions out of 97,426. Over the same period paid search delivered 11.1% and organic search 20.2%. Individual estates ranged from 0.10% to 1.12%, so the portfolio average hides a lot.

The enquiries follow a different shape to the sessions. Those 410 AI sessions produced 12 key events, which is 1.3% of the 893 recorded across the portfolio. Put another way, AI-sourced sessions converted at roughly three times the portfolio average. Twelve is a small enough base that I would not build a strategy on it, but it points the same way as everything else here: not many visits, unusually committed ones.

Treat that 0.42% as a floor, not a ceiling

Three things it does not count, and the first is the big one.

Google’s own AI does not appear in it. When someone clicks a link inside an AI Overview or inside AI Mode, it arrives in GA4 as google / organic, indistinguishable from an ordinary blue link. The AI Assistant channel group catches ChatGPT and Perplexity. It does not catch Google’s AI surfaces, which are the ones most Australians are actually using.

The answers nobody clicks are invisible. A buyer asks which estates are worth looking at in your corridor, gets three named with a paragraph on each, and closes the tab satisfied. There is no session, so there is no data, and on the Pew numbers above that is the common case rather than the edge case.

You cannot see whether you were named, or what was said. Analytics counts arrivals. It cannot tell you that an assistant described your titles as delayed, or recommended the estate down the road instead, or left you out of the answer altogether.

Search Console helps a little, and less than you would hope. Google’s Generative AI performance report gives you impressions in AI Overviews and AI Mode, combined into one number with no split between them, and no clicks, no click-through rate, no average position and no query data. Those impressions are also already inside your Web search totals, so you cannot add them on top of anything.

So we track the part nobody reports on directly. A fixed set of the questions buyers actually ask, put to ChatGPT, Gemini, Perplexity and Google AI Overviews each month, recording which estates and developers get named and what gets said about them. Same questions every time, so the answers are comparable month to month. It is the only way to see whether you are gaining or losing ground in the research that happens before the click.

AI is not yet a traffic channel for property. It is a reputation channel. It decides whether you are one of the two names a buyer brings to their first phone call.

What happens after the enquiry arrives is a separate problem, and we wrote about that in the changing role of chatbots in property lead generation.