When Omny, a property portal developed by Franck Le Tendre, a former Aviv Group executive, debuted in France in February, it could have been written off as another startup piggy-backing the AI wave, dressing up keyword filters in a conversational interface and calling it intelligence.
But Omny combines natural language search, AI and computer vision to interpret buyer intent and capture all the lifestyle preferences that traditional portals typically ignore. (View Highlight)
Asked to find two-bedroom apartments with natural light and parking for one car near the Spanish border, the portal paused. It needed to know the budget first and had one more question: “Are you moving here for work, retirement, or a fresh start?” “I have never lived in France, so I’m not sure,” the reporter said in English. After a few seconds, the portal returned 45 results for apartments in Perpignan, flagging three as best matches. “All three have excellent energy performance ratings (A or B), meaning plenty of natural light and low utility bills,” it said. “The prices are right in line with the local average.” (View Highlight)
Powering Omny’s algorithms is access to some 50 external databases that enrich its listings with data about transport links and schools to build a granular picture of the neighborhood around every listing. More astonishingly, by pulling listings directly from agent CRM systems, the site launched with around 750,000 property listings — just shy of the roughly 800,000 on SeLoger, the market-leading portal operated by Le Tendre’s former employer.
Even so, Omny is small. It generates around 5,000 total visits each month, according to Similarweb — a tiny sliver of the roughly 20 million who visit SeLoger each month. But the startup’s arrival challenges the assumption that dominant property sites are insulated from AI disruption by the completeness of their inventory and the proprietary data they spent two decades collecting. (View Highlight)
For any property site attempting to roll out natural language search, the real challenge isn’t the volume of data they hold, but what the data contains. The data in a standard property listing is typically very shallow, amounting to little more than a description, bedroom count, land size, and a price. (View Highlight)
“You can’t make a natural language query with shallow data,” said Mathew Heywood, founder of Neural Index, a company that enriches property listings using computer vision and geospatial data. “There have been a bunch [of portals] where you can make a natural language search, and it will just either ignore your query entirely or give you half a result.” That’s because for all the LLM integrations and conversational search boxes marketplaces have announced, few are anything more than a natural language wrapper over existing checkbox filters. When a user asks for a “home with a big backyard,” the system still scans for those words in the listing copy. If the agent hasn’t manually entered the attribute, the property won’t appear.
“It’s completely worthless to the AI and that means you’re invisible in that ecosystem,” said Hannah Parker, a product consultant who has worked across the property data industry including stints at Zoopla and HomeTrack. (View Highlight)
Property data poses a stubborn problem for AI because it exists in what’s known in decision science as a “wicked environment” — messy, unstructured information layered with human bias and subjective interpretation. Unlike medical imaging — a “kind environment” where AI thrives because inputs are standardized — in real estate there’s no single objective answer to what constitutes a “good neighbourhood.” “For AI that is a nightmare,” Parker said. The data needs to be structured explicitly — bedroom count: 1; garden: true/false — using schema.org, the shared vocabulary for structuring data on the web. “The problem for all these portals is they’ve got all this data, but they’ve been very bad at managing it.” (View Highlight)
Parker conducted an audit of around 20 portals globally and found CoStar-owned Homes.com had done more than any of the major portals to structure its data in ways that AI systems can use. She also pointed to ImmoScout24 in Germany and Jitty in the U.K. as two other portals with superior AI search experiences, underpinned by superior data structures. (View Highlight)
Consumer Search Behaviours Pre-date LLMs The speed at which agentic AI and LLM search advanced may have taken the marketplaces by surprise, but the underlying consumer behaviour had been building for some time. “Even before AI became a real thing, people had been really prioritising things beyond classic filters,” said Taha Dah, CEO of Search Smartly.
“What we have been observing since about 2020 was the shift toward neighbourhood attributes becoming more important,” Dah said, adding that lifestyle preferences like commute times had become as important to consumers as price or bedrooms. “Covid accelerated that, and we never really saw that reverting back to the old way.” Until recently, dominant marketplaces had enjoyed the luxury of making incremental improvements to the user experience over more dramatic product overhauls. But external forces — native AI challengers, nervous boards, rapidly shrinking share prices — have left the portals scrambling to catch up. (View Highlight)
Develop Tools That Build Confidence, Not Uncertainty Since the marketplaces haven’t addressed their unstructured data problem, most new features amount to what Parker calls the “entertainment” layer: neighbourhood insights, walkability scores, sustainability indexes, crime rates, and black-box recommendation engines that provide consumers with the least value.
Parker thinks the marketplaces should be investing in harder-to-get data, like planning applications and approvals. “Go after the harder stuff first. Everyone is taking the low-hanging fruit, but the valuable data is normally the stuff that’s unsexy — getting this harder data, that’s your moat, because no one else has got it.” (View Highlight)
With more complex data sets, marketplaces can develop tools that build consumer confidence and help them make more informed decisions during the buying process — features like AI-powered affordability assistants or property comparison tools.
“So don’t just give me seven properties,” explained Tilen Pigac, a senior lecturer on AI marketing at Hong Kong Metropolitan University. “Give me seven properties and then distil for me how those properties match to each other and which are most suitable for me.” (View Highlight)
d a property. “A match score is a really powerful way to show value, because there’s a (View Highlight)
Heywood and Dah suggest providing match scores in listings that explain why the AI engine has included a property. “A match score is a really powerful way to show value, because there’s a lot of stuff that happened under the hood that the consumer may not know about,” Dah said. (View Highlight)
Enriched Listing Models There is some lingering concern that moving to natural language search and rationalised listing recommendations will supplant the marketplaces’ existing business models, but Heywood doesn’t think that’s the case. “We can have portals feed through a number that comes with the listing and then edit where it appears in the ranking.” Parker suggests introducing an enriched listing model. “If you have all this great data that you’re able to enrich so that AI agents can answer great questions, maybe there’s a premium attached to that — premium data listings rather than premium position listings — and that could be a good way of gently shifting the monetisation model without upsetting anything too much.” (View Highlight)
Consumers will no longer treat marketplaces primarily as the place to search for property; they will evolve into being the place to conduct research.
“This is what’s most misunderstood. Buying a house is more similar to investing than shopping on Amazon — and that is what a lot of portals are misunderstanding.” — Tilen Pigac, Hong Kong Metropolitan University “I don’t think it’s going to flip the switch and replace it overnight,” Heywood concluded. “It’s going to run in parallel, and people will slowly wean off filters. Conversational search becomes the new normal in five years.” (View Highlight)
