Your buyers aren't searching the way you write your listings

A weekly briefing on the gap between what AI can do for your business and what you're actually using.

About a billion people are now searching Google differently than they did a year ago. They're typing full sentences instead of keywords, asking follow-up questions, sometimes talking or uploading a photo, and they're getting answers back rather than a page of links to work through.

None of that will be news to you. The idea that search is changing has been the industry conversation for a solid year now, and you've probably heard a version of it at every conference and in every newsletter since. What's new is the weight behind it. Google put hard numbers on the shift at its I/O event on Tuesday, and they're difficult to wave away. AI Mode has crossed a billion monthly users in its first year. The average AI Mode search runs about three times longer than a traditional Google search. More than one in six searches now involves voice or images, and image-based searches have been climbing roughly 40 per cent month on month since launch.

Your buyer is somewhere in that billion, and the listing you publish next week is almost certainly not ready for the way they're now searching.

What the buyer just stopped doing

For a long stretch now, the reader you've been writing every listing description for hasn't really been the buyer. It's been the search engine. Word by word, you've been feeding the keyword filter. Two-bedroom. Heat pump. Modern kitchen. Mount Eden. Easy commute. The buyer types two or three of those into a portal, your listing comes up, the system does its job.

That buyer is still out there, but a growing number of them don't open a portal first. They open ChatGPT, or Google AI Mode, or Perplexity, and they describe the life they want rather than the property they're shopping for.

A sun-trap weatherboard in Mt Eden under 1.8, with indoor-outdoor flow for two kids and a dog. Single level if possible. Walking distance to a primary school I'd be happy with.

The model reads that, decides which listings and agents and suburbs to put in front of the buyer, and serves them up inside the conversation. When your listing turns up there, it isn't being matched against a keyword list anymore. It's being interpreted, which is a different thing entirely. A machine is looking at your three hero photos and reading your two paragraphs of description, then making a judgement about whether this property answers what the buyer just asked for.

Issue 1 called this the faster-typewriter trap: using AI to do the same job more quickly without asking whether the job itself has changed. Writing the same listing description in half the time doesn't help much when the description was built for a search engine your buyer is using less often every quarter.

Who is handling this for you?

If your agency has someone who looks after the website and the listing feed, this is their challenge. Search visibility has been a constant work-in-progress. What's changed is that the rules underneath it have moved, and the familiar SEO playbook of keywords and meta tags only takes you part of the way now.

The consultants are already circling. The moment AI search becomes a worry that reaches the leadership team, SEO and "AI visibility" consultants start pitching, and some have genuine expertise to bring. But handing the whole thing to a consultant before anyone inside the business understands it is how agencies end up paying for work they don't know how to evaluate. The person who owns your digital presence should be able to explain, in their own words, how an AI engine decides whether to surface one of your listings. That understanding isn't optional anymore, and it isn't the kind of thing to outsource before you've got a grip on it yourself.

The portals already know

You'd be forgiven for thinking none of this has reached your buyer in Mt Maunganui or Halswell yet, but it has, more than you'd guess. Google AI Mode is live in New Zealand. ChatGPT is the default research tool for a generation of buyers under 35, and your last open home almost certainly included people who'd opened an AI conversation about the property before they walked through the door.

The portals are watching the same buyer move and testing ways to stay close to it. Trade Me Property launched inside ChatGPT in February as the first New Zealand company in OpenAI's app directory, and realestate.com.au led its autumn release across the Tasman with a conversational search assistant. Whether any single one of these experiments takes hold is a separate question. What matters more is the pattern they're all responding to, which is the one already sitting in your buyer's pocket. It's worth keeping an eye on across the year, because the portal that ends up owning the AI conversation owns the buyer relationship in a deeper way than any portal has before.

The early evidence

This is where it gets concrete. Cited, a firm that audits how AI engines recommend businesses, ran 324 buyer-style queries through ChatGPT, Perplexity and Gemini across six Australian cities earlier this autumn, asking each engine to recommend a real estate agent. Two findings are worth your attention.

The first is that the three engines barely agreed with each other. Across the six cities, fewer than one per cent of agencies showed up consistently on all three, and in four of those cities not a single one did. Each engine reads different source material and weighs a buyer's intent differently, so being visible on one tells you nothing about the other two.

The second matters more for what you actually publish. The agencies that surfaced for a specific question were the ones whose own content addressed that specific thing. Cited found that agencies with suburb-level pages appeared for suburb-level queries, while agencies without that content were invisible to the same queries, and that held with no exceptions across all 324 tests. Generic "full-service" framing returned almost nothing. The engine surfaces the page that answers the question in front of it.

It's Australian data, and Cited sells AI-visibility services, so treat it as directional rather than gospel. But the mechanism isn't country-specific. The same models read New Zealand listings, and they apply the same logic one level down. The listing that clearly answers the buyer's question is the one that surfaces. The listing that leaves the buyer to infer the answer often won't.

Also this week

Rex just bolted AI into the CRM. Elite Agent · 8 May 2026. Rex Software, a CRM with reasonably wide use in NZ, has rolled out a four-component AI suite inside the platform: AI Admin for voice and text data entry (in beta), AI Prospecting for who to call and what to say, AI Nurture for automated high-volume follow-up, and AI Manage for performance and forecasting. If you're on Rex, this lands inside your existing workflow without a tool change. Worth knowing what's in beta right now and what your agency is or isn't switching on this quarter.

A cloud brokerage is recruiting on the strength of its tech. Property Noise NZ · 18 May 2026. eXp Realty, which has run a branchless, cloud-based brokerage in New Zealand since 2022, is reported to be recruiting Kiwi agents harder this year, leading with a model built on remote working, shared technology, revenue share and equity rather than physical offices. Whatever you make of the model itself, the pitch is the part worth noting: it's aimed squarely at agents who feel their current tools should be doing more for them. It's a thread we'll pick up properly in a future issue.

Connected AI assistants just got real. Inman · 11 May 2026. Inman, a long-running US real estate news publication, ran a piece by Audie Chamberlain who wired email, calendar, Slack, ClickUp and Chrome into a single Claude Cowork agent that works across applications rather than answering prompts one at a time. For anyone already using Claude for copy, this is what the next step looks like in practice: an AI running back-office sequences while you're standing at an open home. It's Mac-only for now and needs a paid Claude subscription. We'll show you how an NZ agent can put together a workflow along these lines in a coming issue.

The listing audit

The lead of this issue comes down to one practical question: when a buyer describes their ideal home to an AI, does your listing answer them, or does it leave them guessing? Here's a prompt that runs that check on a real listing.

Paste it into Claude or ChatGPT. The model will ask you for the current listing description, a short text description of each of your three hero photos, and a handful of property facts. From there it produces a three-part audit and a rewrite you can use straight away.

The whole exercise takes around ten minutes. Try it on the strongest listing in your current campaign first, and you'll almost certainly find at least one thing worth fixing before the next open home.

PromptListing AI-Readability Audit

You help New Zealand real estate agents audit a listing for AI-readability. You'll work through this in three passes, gathering what you need through a short conversation. Ask one question at a time. Wait for each answer before moving on. Keep your questions short.

**Question 1. The listing facts.** Address or suburb, asking price or price band, bedrooms, bathrooms, land size, method of sale, and the launch date. One line.

**Question 2. The current description.** Paste the full listing description as it currently appears on the portal.

**Question 3. The three hero photos.** Describe each of the three lead photos in two or three sentences. What's the shot? What's in frame? What's the feeling? Don't write marketing copy, write what a friend would describe over the phone. If a photo is staged or virtually staged, say so.

**Question 4. The buyer profile you actually expect.** One sentence on who you think this property is going to sell to. Be specific. "First home buyer couple, 30s, professional, no kids yet." Not "any buyer".

Once you have those four answers, produce the audit in three sections.

**Section 1. Ten queries this listing should surface in.**

Generate ten plausible buyer queries this property would be relevant to, written the way a real buyer would now talk to an AI. Use the listing facts and the buyer profile. Make them specific to NZ and specific to this property. Mix generic queries (e.g. "three-bedroom family home in [suburb] under [price band] with a north-facing yard"), lifestyle queries (e.g. "single-level home walkable to a primary school in [suburb], dog-friendly section, room for a home office"), and situational queries (e.g. "first-home-buyer-ready property in [suburb] with low maintenance, close to public transport"). Don't pad. Each query should be one that an AI engine would meaningfully attempt to answer.

Output as a numbered list, one query per line, no commentary.

**Section 2. Coverage check.**

For each of the ten queries, rate the listing on whether the current description and photos answer it. Use three ratings.

Green: the current listing clearly supports this query, with specific evidence in the description and visible support in the photos.
Amber: the listing partially supports this query, but the evidence is implied rather than stated, or only one of the description and photos backs it up.
Red: the listing doesn't support this query at all, even though it should.

Output as a table with columns: query, rating, what the listing currently says, what's missing. Be specific. Don't say "the description could be clearer". Say "the description doesn't mention school zone, north-facing aspect, or the home office space the photos show".

**Section 3. The rewrite and the shot list.**

Rewrite the listing description. Pre-emptively answer every amber and red query from Section 2. Keep to roughly the same length as the original. Don't add features the property doesn't have. Don't claim school-zone status, sun aspect, or measurements that weren't in the inputs. If something is implied by the photos but not confirmed in the facts, flag it as "confirm before publishing".

Then list one to three photos the agent should consider adding or replacing before the next open home, with a one-line reason each. Tie each photo recommendation back to a specific red or amber query.

Voice rules for the rewrite. NZ English (analyse, organisation, behaviour, recognise). No em dashes. Use full stops or commas instead. Direct sentences. No estate-agentese. No "stunning", "boasts", "nestled", "must-see", "this won't last". No vendor cheerleading. Write to one buyer, not "buyers". State the facts and let the buyer weigh them.

If you're using Claude, drop the actual photo files into the chat at Question 3 instead of describing them, and the model will run the vision pass directly. The results come back noticeably sharper.

Run it once this week and reply to tell us what surfaced.

Happy selling.

— The Listing Signal