rw-book-cover

Metadata

Highlights

  • Most of the AI property search launched this year is a friendlier way to fill in the same filters. That is a useful improvement, but it is not the change that matters. The change that matters is search that behaves like a good agent, and when it arrives it will put pressure on the way most portals make money. (View Highlight)
  • The launches are real. Rightmove opened a conversational search, built with Google’s Gemini models, to part of its audience in February. Zillow launched AI Mode in March. Bayut put an app inside ChatGPT in June and added AI Search to its own platform in September. In each case a buyer can describe a home in plain language rather than clicking through a filter panel. (View Highlight)
  • Look at what comes back, though. It is still a list of listings that match a set of criteria, in an order the portal decides. The buyer’s sentence has been translated into the same price, location, bedroom and property type fields the filter panel already had. The input is nicer. The output has barely changed. (View Highlight)
  • In my experience of portal search data, community and property type are the criteria seekers almost never relax online. Someone searching for an apartment in one area rarely clicks on a townhouse in another. (View Highlight)
  • What I found more interesting came from talking to agents after they had received those leads. The criteria that looked fixed online were not fixed once the seeker was talking to a person. An agent who properly understood what the seeker needed could show them options they had ruled out without realising: a different property type, a different location, and sometimes a move from renting to buying, because the seeker had never worked out that they could afford to buy. (View Highlight)
  • That is the part of the job that search has never done. A filter cannot tell you that what you asked for is not what you need. Until AI search can, it is another way of filtering the results. (View Highlight)
  • Most portals sort results by default under a label such as “relevance” or “featured”. A seeker reads that as relevance to their criteria. Anyone who has worked inside a portal knows it means something more complicated: the order produced by that portal’s own algorithm, which balances what the seeker wants against the commercial need to give visibility to the agents and listings paying for it. (View Highlight)
  • That is not a criticism. Much of portal revenue is ultimately tied to position in search, through featured and premium listings, boosts, and credit models where agents spend to move up the page. The model works because the seeker has to scroll a long list, so a place near the top of it has real value. (View Highlight)
  • Now picture the same seeker describing their ideal home in the detail that usually only comes out with a good agent. Walking distance to a particular school, morning light in the living room, a building that allows dogs, a budget that could stretch if the service charge is low, and an open mind about buying if the numbers work. (View Highlight)
  • An AI assistant with the right access could check listings across portals and brokers’ own websites, pull nearby amenities from maps that were never in the listing text, read the photos to judge style, space and light, and work out whether buying is within reach. It would then return the one or two homes that actually fit, with its reasons, instead of several pages sorted by relevance. (View Highlight)
  • Parts of this are possible now. Current models describe photographs well, and map and amenity data is widely available. The harder parts are reliable inventory and getting the hard constraints right, because a buyer who said AED 1.6m and is shown a home at 2.1m stops trusting the next answer as well. (View Highlight)
  • It is also early. An industry analysis in September put it plainly: any business getting more than 1% of its traffic directly from ChatGPT or Gemini is an outlier. Yet Rightmove told investors last November that its AI investment would hold operating profit growth to between 3% and 5% in 2026. The larger portals are treating this as a business model question, not a feature, and they are right to. (View Highlight)
  • The question is not how to add a chat box. It is how to serve property seekers when search no longer produces a long list to rank. (View Highlight)
  • That points towards revenue tied to outcomes rather than position: a qualified enquiry or a booked viewing, from a seeker who was matched rather than shown. Verified data becomes a product in its own right, including for AI assistants on the portal’s own terms; Bayut’s and Zillow’s apps inside ChatGPT are early versions of that. The hardest step is building an agent’s judgement into search itself, so the product asks questions, suggests alternatives and runs the affordability numbers rather than only parsing the request. (View Highlight)
  • The portals that come through this well will be the ones that accept, before their users do, that a ranked list of 400 homes was never what the seeker wanted. (View Highlight)