Hannah Parker on 'Wicked' Real Estate Portals in the Age of AI Agentic Search

“The gap in technical advancement in our industry is still a big challenge that AI is exposing faster”, says Hannah Parker, founder at Inoki Agency. "When I have conversations about product innovation in other industries, it feels a world away from the technology shipping across the real estate/proptech space.”

Parker's opinion is worth listening to. She has been innovating data-driven products in the real estate portal industry for nearly two decades with tenures at Zoopla, Propertyfinder, Hometrack and more recent consultancy and fractional roles with Inoki.

To set the context for our conversation, Hannah highlights three key AI plays. The first, "AI Theatre", is the shiny wrapper of ‘AI’ that sits above traditional features like filters and keyword search. The next level taps into the entertainment power of AI: virtual staging, playful natural language and speech-based search. These are new and delightful features, but it's unclear what impact they will have on increasing lead volume or quality.

The real value, suggests Hannah, is in the structural gap between the data that portals acquire and build, and how to leverage those datasets for human and AI users. Information such as safety,  environmental risks, crime, building restrictions, noise levels, and investment potential are all namedropped.

To achieve this in the new era of AI-assisted search, she says, portals will need to design products for both the human user and for the agentic search experience; AIx.

"AI agents are like digital assistants working for you. Send this email for me, search for this product, add to my basket and even pay. They are acting on your instructions, but they don't read a website as a human user; they don’t care about the great UX. They can't read it in the same way.

"AI agents are desperately trying to find you an answer, but if your data [as a portal] isn't structured in the right way, it is  either invisible to the AI, or it is forced to work even harder, impacting token use and context windows

“Focusing purely on visibility at the model level (your SEO and GEO strategy) may not be enough, especially for challenger brands. Product teams need to think about a new AI user persona and design for them deliberately. AI agents are good at guessing and are confidently wrong. Data locked inside images with no schema or buried in ambiguous free text cannot be extracted reliably or consistently at scale. Structured schema is the love language for AI agents. It uses fewer tokens, reduces hallucination risk and gives you results you can trust.”

"The days of being only a lead generation machine are over. The portals that will stand the test of time and survive are the ones that enrich the experience beyond the traditional 'search and find'. They will give consumers genuine advice based on reliable data. Portal executives need to get a massive handle on the hard and soft criteria that are key to the decision-making, and how to build an experience around those demands."

At the recent Proptech and Portal Watch conference in Bangkok, Parker (pictured) politely reminded several major portal operators that slapping an AI chat search on their website isn't quite the innovation the press release makes it out to be.

As AI agents take on more daily tasks for users, including property research, having readable data is not a ‘nice to have’; it is a basic requirement that the industry is slow to get on board with and slower to build for. As she said on stage in Bangkok, "If an AI agent can’t read your listings, you risk not existing in the AI search economy."

Parker's presentation in Bangkok shared the results of an experiment she ran in the lead-up to the conference.

In it, a selection of portals were displayed: Homes.com and Zillow in the United States, realestate.com.au and Domain in Australia, the UK's Rightmove, and Germany's Scout24. The portals were all ranked based on the results from an agentic search (AIx) audit and a typical search prompt. 

The higher the score, the more structured the data architecture behind the scenes, and therefore the more readable by the AI agent.

Tested on nine parameters, four out of six portals scored less than 50%, while some of the largest portal leaders passed just two parameters. Only Homes.com, powered by the data specialist CoStar Group, had a deliberate AI-readability strategy on its platform, with structured data for seven out of nine parameters.

Meanwhile, the AI agent struggled to interpret basic search instructions such as the number of bedrooms of a property listed on Zillow, Rightmove, realestate.com.au, Domain, or Scout24.

To be clear, the AI agent didn't fail in its prompt. The data architecture did.

As she explains:

"There is a structural property data problem that was evident in almost every portal I tested. There's a concept in decision science called 'kind' versus 'wicked' learning environments. Kind environments have fixed rules and reliable feedback, like chess, radiology reads in healthcare, and law. Play the same input, get the same output.

"Wicked environments are different. The rules are unclear. The feedback is delayed or ambiguous. The 'right' answer depends on factors you probably don't have, or haven't clearly defined.

"Property search, and therefore search data, is profoundly wicked. 'Close to a good school' sounds like a simple and well-used criterion for users, but it means something different for every family in the room. Criteria can shift mid-process; human factors are simply not reliable.

"The portals that want to secure visibility with Agentic browsers must leverage their rich data into structured, verified data schemas to make those wicked environments more navigable, designing for both the human and the AI agent users at once."

The vast quantity of unstructured, subjective data on major portals is a real problem, embedded in a 'wicked' environment that renders property data unreadable by unfocused, unreliable and subjective criteria. After all, how close does a property need to be to an electric charge point to be deemed objectively 'close'? How does one measure 'a short commute'? What constitutes a 'good school'?

And how negotiable are these criteria when you're engaging in a highly emotional search? If you're buying your first home and planning the next phase of your life, perhaps that extra five-minute commute doesn't matter as much as enrolling your young child in a school with an 'Excellent' rating.

So to recap: unstructured data creates a 'wicked' environment that is difficult for AI agents, who thrive in 'kind' environments, to parse. And surprise, surprise, home search is as wicked as it comes.

The argument is compelling, transformative, and backed by evidence (a strong correlation from an admittedly small sample size). But if we assume that every portal knows this, why are they choosing to stay wicked?

"It's a really good question. You look at the situation, and you think, why would everyone not just have well-structured data on all the listings? It's crazy.

"Take Zillow and Rightmove. [They both] scored low on the AIx audit, despite having some of the best SEO in the industry.  The trend from the #1 portals seems to be one that’s more protective of their data. Why give away your gold if your brand alone is strong enough to pull in the users?

"There is a question of commercial conflict here too; For brokers/real estate agents, less data (on the listing) is great because it gets more questions coming through, increases lead volumes and at a basic level, demonstrates to vendors their marketing strategy is working. Meanwhile, consumers would love to see more information that helps them make more informed decisions, often trusting the reputation and advice of the portal over the broker.

"Before making radical changes to our search behaviours, it is worth taking stock that we are talking about a small amount of traffic here. Many of the discussions I am having with leadership teams of portal businesses right now are about weighing up data exposure risk. Do we give away too much and then reduce the value of what we have? Or do we risk not giving away enough and being invisible?

"There is a level of brand arrogance in some of these bigger businesses, and I think that will impact their speed and innovation urgency. These portals can make a lot of money from their data. I think their greater risk will come from [below]. [Challenger businesses] are more interesting because they can't rely on brand power alone to drive traffic. They want the highest amount of visibility and pull new users in at the top of the funnel. I don't think a challenger has ever had a better opportunity to bypass big brands, and democratise one of the biggest assets."

Parker predicts that all portals that scored lower in the February AIx audit will improve their score, or have a more evident data and AI strategy, by the time she shares updated findings at the Proptech and Portal Watch conference in Madrid later this year.

And while AI provides cost-cutting potential for portals, there is a revenue-generating angle worth discussing. Excellently-built products fuelled by valuable data and specialised, centralised AIs—are easily monetisable for buyers, sellers and agents.

Innovation is accelerating rapidly, and alongside this, sophisticated products are being developed that boil down to a supplementary SaaS business. Lightweight software that does one thing really well can be rolled out to attract high-value leads that are willing to pay for the pleasure.

Say it quietly, it's a revenue stream that smells a little of consumer-paid. High intent and motivated buyers investing their cash to access products that facilitate a faster, more informed sale or purchase have the potential to rock the subscription model boat, but will it come from the portals, the brokers or the CRM’s?

This is already happening around the world. Scout24 and Swiss Marketplace Group have already successfully launched paid-for consumer products that deliver value and insight to buyers and sellers. In the UK, Zoopla's MyHome product puts great data in the palm of its users' hands (though its monetisation model lags between Scout24 and SMG). These products, designed at their core to enhance user engagement and increase lead quality to agents, will only become more streamlined, searchable and powerful as AI evolves.

Final question: Will the makeup of the workforce change with the evolution of AI? Parker’s view is that AI is not yet sophisticated enough to be trusted with genuine innovation. The human creative and strategic layer remains essential.

"The opportunity is to use AI to accelerate specialist skills, not replace them. That said, the makeup of product and engineering teams is already shifting. As AI coding tools compress delivery timelines, the bottleneck moves upstream to design and product definition. Traditional execution roles are evolving into more architectural ones, directing and shaping AI output rather than writing every line from scratch.

"Think of it as a simple hierarchy of risk and opportunity: specialists who use AI to sharpen their specialism sit at the top, more valuable than ever. Specialists who do not engage with AI are next, still protected by depth. Generalists who adopt AI can hold ground, but are more exposed than domain experts who do not. Generalists who ignore AI entirely are the most vulnerable. Depth compounds. Breadth, without AI, gets commoditised. As Parker puts it, data engineers will not be writing code. They will be creating the architecture for the AI to write it."

What do we do about closing the skill gap in tech in the age of AI? At the heart of it, this is still a people industry. Ben Horowitz famously quotes, 'Take care of the people, the products, and the profits, in that order.’ That sequence has never mattered more than right now. As Hannah says:

“Today, your engineers and product teams are losing sleep. Every morning brings the doom scrolling about a new model, a new tool, a new competitor announcement; the noise is overwhelming. The teams that will build the best products are probably not the ones who move fastest in ‘panic-mode’, but the ones who have space to learn and experiment, and are supported by leaders who stay honest about the problems they are actually solving vs FOMO in shipping the next shiny thing. The next great idea is just as likely to come from a junior engineer experimenting as from your leadership team.”