Every Portal Has an AI Assistant. Data Decides Who Wins.

Sometime in the past twelve months, every major real estate portal's product roadmap converged on the same slide: a chat box.

Zillow shipped AI Mode in March 2026, a conversational assistant that answers affordability questions, compares listings, books tours and hands users to an agent. CoStar's Homes.com had gone even earlier and louder: Homes AI launched in February, voice-enabled, powered by Microsoft Azure OpenAI, billed by CEO Andy Florance as a shift as significant as online search itself. Realtor.com followed in June with RealAssist AI, built in collaboration with Google. In Germany, ImmoScout24 launched HeyImmo, a GPT-5 powered assistant its parent Scout24 developed with OpenAI, which asks clarifying questions and guides seekers through the whole platform. Spain's Fotocasa announced its own AI-assisted search the very same week. Rightmove is pouring money into AI — enough, in fact, that the announcement knocked roughly 12% off its share price.

Different continents, different business models, different regulatory environments. Identical product.

This is not a coincidence, and nobody is winning this race. It's a re-run of a pattern the portal industry has seen twice before, and the portals that misread it this time will spend a fortune building something that gives them no edge at all.

We've seen this movie before

When map-based search arrived in the late 2000s, it was a genuine differentiator for about eighteen months. Then every portal had maps, and maps stopped mattering as a competitive weapon. The same thing happened with mobile apps in the early 2010s. First movers enjoyed a brief window of advantage, then apps became the price of admission.

Conversational AI search is following the exact same curve, just faster. The window between 'innovative feature' and 'table stakes' used to be measured in years. With foundation models it's measured in months, because unlike maps or apps, the hard part isn't built in-house. Zillow, Scout24, Realtor.com and Homes.com are all building on the same handful of foundation models from the same handful of labs. Scout24 publicly credits GPT-5. Realtor.com built with Google. Homes.com runs on Microsoft Azure OpenAI. The underlying intelligence is rented, not owned.

When your competitors can rent the same brain you rent, the brain is not your moat.

CoStar, to its credit, seems to understand this. Read the Homes AI launch announcement closely and the pitch isn't the chatbot at all. It's that the assistant draws on Homes.com's property database, Matterport 3D digital twins, proprietary school data and neighbourhood insights. The interface gets the headline; the data carries the argument. Even the loudest assistant launch in the industry is, underneath, a data story.

There's a second force accelerating the convergence, one analysts at Online Marketplaces have called homepage homogenisation: portals copy each other's interfaces relentlessly, because in a winner-takes-most market nobody can afford to let a rival's UX innovation go unanswered. AI search is homogenising before it has even finished launching. By 2027, a chat box that answers 'can I afford this if I move in June?' will be as unremarkable as a map tab.

What Zillow actually said when nobody was listening

Here's the part most coverage missed. In its open letter on AI published in March 2026, Zillow didn't argue that its assistant was better than anyone else's. It argued something much more interesting: that horizontal LLMs, brokers, MLSs and rival listing sites lack Zillow's proprietary collection of integrated assets, and that every consumer interaction generates context that improves personalisation, which deepens engagement, which generates richer data. A compounding flywheel.

Read that carefully. Zillow's stated bet for the AI era isn't the chat interface. It's the data underneath the chat interface.

That is the correct read of the competitive field, and it's worth being precise about what 'data' means here, because it's not one thing. It's at least four.

Listing depth and exclusivity

Direct MLS feeds versus scraped approximations. Coming-soon and private inventory access — which in the US just became a live battleground as Zillow and Realtor.com move to advertise private listings while the Clear Cooperation Policy collapses in everything but name. The portal whose assistant can surface inventory the others literally don't have wins conversations before the AI even matters.

Behavioural context

Hundreds of millions of sessions teach a portal what 'family-friendly' actually means to a user in Hamburg versus Houston: which trade-offs people accept, which dealbreakers are real. Scout24's team discovered this building HeyImmo. Users didn't want better search results, they wanted a sparring partner that understands intent. You cannot rent that understanding from a foundation model. It comes from your own traffic.

Transaction and outcome data

Knowing not just what people searched, but what they toured, offered on, financed and closed. Zillow's vertical integration across search, touring, financing and closing is precisely about owning this loop. An assistant that knows how transactions actually end gives advice an assistant trained on listings alone cannot.

Proprietary enrichment

Verified floor plans, energy data, noise scores, valuation histories. The structured features that let an assistant reason instead of retrieve.

The uncomfortable question for mid-size portals

All of which leads to the question that should be keeping mid-market portal executives up at night. If you're building an AI assistant on the same rented foundation models as the market leader, with a fraction of the behavioural data and thinner inventory, what exactly are you shipping?

A worse version of the same thing, at a meaningful fraction of your product budget.

This doesn't mean smaller portals shouldn't build conversational search. They probably have to. Table stakes are still stakes, and a portal without an assistant in 2027 will look like a portal without a mobile app in 2015. But it changes what the assistant is for. For the data-rich market leader, the assistant is a moat-deepening machine where every conversation feeds the flywheel. For everyone else, it's defensive parity: necessary, expensive, and strategically neutral.

The genuine opportunities for non-leaders lie elsewhere. Local data depth that national players can't match. Vertical specialisation in luxury, rentals or new construction, where a focused dataset beats a broad one. Or becoming the best-distributed inventory source for the AI layer everyone else is building. Idealista already exposes an MCP server that lets third-party AI assistants search its listings directly — a portal positioning itself for a world where the conversation happens somewhere else entirely. Which brings us to the elephant in the room.

The threat nobody's assistant solves

There's a scenario in which this entire arms race is beside the point: the one where consumers stop visiting portals and send their own AI agents instead. ChatGPT, Gemini and their successors increasingly answer property questions directly, and a growing chorus of commentary argues that agentic AI will disintermediate the portals entirely, with buyers' personal agents querying primary data sources while portals lose their traffic monopoly.

Rightmove's CEO Johan Svanstrom has called the idea of AI replacing the portal 'unthinkable.' Maybe. The honest answer is that nobody knows the timeline, and portal traffic dominance today remains overwhelming. But notice what would actually protect a portal in that scenario, and what wouldn't. A beautiful chat interface on your own website does nothing when the user never visits your website. Exclusive inventory, proprietary data and being the source AI agents must query? That travels.

In other words, the agentic threat and the convergence problem have the same answer. The interface is perishable. The data is portable.

What this means in practice

For portal operators, three implications follow.

First, stop measuring the AI assistant as a differentiator and start measuring it as a data acquisition engine. The conversations are worth more than the conversions, because they capture intent in users' own words, and that is training data money can't buy. The early evidence supports this: CoStar reports that users who engage with Homes AI spend significantly more time on site and submit more leads. The assistant's real output isn't answers. It's engagement and intent data.

Second, audit what data you hold that the market leader genuinely cannot replicate, and concentrate AI investment there rather than on interface polish that will be copied within a quarter.

Third, decide deliberately whether you're building a destination or a source. If there's any chance the AI layer ends up owned by someone else, make sure your inventory and data are positioned to be indispensable to it.

For MLSs and data providers, the convergence is quietly excellent news. If assistants are commodities and data is the moat, the organisations that govern, license and enrich the data just became more strategically important, not less — provided, of course, they price and protect access accordingly.

The portals that win the AI era won't be the ones with the best chatbot. They'll be the ones whose data makes any chatbot, theirs or anyone else's, smarter.

How to cite this analysis