Published on the 04/06/2026 | Written by Heather Wright

Search is shifting, but intent isn’t…

How people expect to search is changing – but their search behaviour itself remains the same. That’s one – very early – takeaway from realestate.co.nz’s rollout of AI-powered image search, which allows users to search for features directly within property photos, rather than relying solely on keywords and listing descriptions.

Within two weeks of launching the feature in the realestate.co.nz app, more than 60,000 searches had been clocked up. It’s still a tiny percentage of the company’s searches, but  Simon Hargraves, realestate.co.nz chief information officer, is optimistic.

“We had just finished a project to get our data into a good state. Without that, it would have been extremely difficult to do this work.”

He told iStart users are increasingly comfortable working with AI and expectations for search are changing. “They want to be able to express their search intent more naturally, rather than just using predefined filters or having to use exact keywords.”

The AI-image search gives users the ability to search ‘at a completely different level’ – searching what’s actually in photos.

What users are searching for, however, hasn’t changed in early searches. “We have a list of traditional keywords people were searching for and generally it’s following the same trend,” Hargraves says.

What has changed is how effectively results are being surfaced. Traditional keyword search relies on exact inputs from both sides, with agents required to include the right terms in listings and users having to guess and input the same wording. AI image search removes that dependency, Hargraves notes, with the AI system examining the photos and pulling out key features.

The system also changes how results are presented, with the platform able to surface the image where the feature has been detected, rather than returning a listing with no obvious visual confirmation.

The accuracy challenge

Hargraves says the feature itself was built and released quickly – one month to build and another month spent testing, tuning and getting feedback from the business. “The really challenging part was more the focus on accuracy and making sure the results aren’t frustrating for users because the way the semantic search works is it expands what the user is searching for to understand intent and then expands the search out from there.”

As an example, early on a search for wine cellars associated them with luxury properties – returning a broad range of luxury properties rather than those with wine cellars. Worse, searching for a property with an accessible ramp expanded out to include staircases.

“Those kind of cases are obviously going to be super frustrating to users when you’re getting the opposite of what you’ve actually search for. So we spent probably half the project time actually testing it internally, across the whole company and getting feedback from users of examples where it was wrong. Then we spent quite a bit of time tuning the system.

“Making sure the experience wasn’t frustrating was the most important part.”

Data does the heavy lifting

Behind the feature sits a significant data effort.

Realestate.co.nz had just completed an 18-month re-platforming and centralisation of all of its data into a single Snowflake platform, standardising definitions and consolidating multiple data sources across the business.

“That was a huge foundational piece that allowed us to build it very quickly.”

The work enabled the rapid development of AI features, but also highlighted a clear dependency.

“You have to have that data foundation in place… if you have bad data, then you’ll get bad results out from the AI.”

In image-based search, that extends beyond structured data to the quality of photos themselves – something the company doesn’t control.

“The AI performs much better when the features [are] clearly visible,” Hargraves says.

But he notes the higher the quality the images and the larger they are, the more cost to process via AI. “There are quite a few variables that we had to take into account and tune to try and get the right kind of cost versus quality.”

He deliberately frames the AI image search as an experiment. “The technology has evolved really rapidly and these kind of features are becoming easier and easier to implement so it makes sense to build the feature and release it to users and see if it’s used and useful.”

The company is less focused on initial uptake than on whether users will return. As Hargraves notes, if users don’t find it useful, they won’t come back to it again – a key test for the future of the system.

To manage expectations, it has been clearly labelled as a beta. “Because… with AI, it’s never going to be perfect.”

Beyond search

The company’s broader approach reflects a pragmatic view of AI. Internally it is being used across the engineering process, and teams are encouraged to experiment through hackathons and rapid prototyping.

“These tools are making it super easy and super quick to build out things and test them internally and build proof of concepts,” Hargraves says.

At the same time, decisions to productise those ideas need to be made carefully, he says, noting the challenge of ensuring projects show ROI and aren’t constantly ‘experiments’.

“You have to make sure projects make sense… and as with anything its prioritising what you have in your roadmap case by case with cost versus value.”

In that context, AI image search is seen as a logical extension of existing capabilities, rather than a standalone transformation.

Hargraves says the company is already working to merge AI image search with traditional keyword search into a single, unified experience.

Longer term, there is potential to expand search into other areas, including accessibility requirements and proximity to amenities, though that will depend on the availability and quality of additional data sources.

“Finding a good data source and being able to ingest it and surface it via AI search is going to be the challenge. But I think you’ll see a lot of those features coming over time. It’s just about getting data into a good place first because without that solid foundation you can’t build high quality AI experiences on top of it.”

Beyond real estate

So what’s Hargraves advice for any other companies considering similar tools?

“Getting the data into a good place is probably the most complex thing. We’re lucky because we had just finished a project to get our data into a good state. Without that, it would probably have been extremely difficult to do this work.”

The data needs to be high quality, well-structured and accessible and good governance around it is also required as well.

“Without that, it’s going to take a lot longer and there’s going to be much higher levels of complexity.”

And be prepared to spend time tuning. “The tuning and accuracy piece is going to be the most difficult part to get right from a product perspective,” Hargraves warns.

“With AI image search you don’t have so many of the concerns you’d have with AI in other areas, especially where you have chatbots and people interacting directly with the AI. AI search is an easier place to start but as we get more into potential AI features in future – chatbots and that kind of thing – it becomes more and more concerning from a risk perspective.”

On the cost front, he says there wasn’t ‘massive’ cost. The time spent tuning was the bigger cost, with the company now spending around US$1000 a month for processing all of its listing images.

“Costs are increasing in AI across the board though, so it’ll be interesting to see how it changes over time and it’s something we’re going to have to keep an eye on.”

Hargraves also acknowledges that for some companies, AI image search could surface unexpected content.

“With our platform, the photos have all been taken and verified by vendors and agents, so personally identifying information or anything sensitive is usually scrubbed from the photos anyway. On different platforms, you would definitely have concerns though,” he says.

Despite that, Hargraves says it’s important for all local companies to be experimenting with AI.

There are huge changes coming across every industry. There’s a massive shift in the way people are working in businesses and how they are interacting with AI and that’s changing their expectations of how they use other products too.”