AI & Real Estate: 5 Signals to Watch This Week (Week of September 21, 2026)

For two years, AI has slipped into listings, valuations and client conversations, often without saying so. This week, three very different voices asked for the same thing: that it declare itself, that we measure what it's actually worth, and that we keep it where it's useful. The French government, a data-driven study and two tech executives reach the same conclusion from opposite directions.

Asked by MP Édouard Bénard about AI use in property listings, the French government replied that it doesn't plan to create AI-specific rules. It doesn't need to: the uses mentioned - photo retouching, property enhancement, virtual staging - already fall under the duty of fairness and the ban on misleading commercial practices. Professionals are also bound by their code of conduct. The government added that automated applicant screening remains subject to anti-discrimination rules, under the oversight of France's rights ombudsman. The practical consequence is simple. The question isn't "did I use AI?" but "does this image give a false idea of the property?". A living room virtually furnished to help buyers picture themselves there is fine. An erased crack, a removed overlooking building or an enlarged room is not. That's the line we follow: a reworked visual must stay faithful to the real property, and be identifiable as reworked. Source: Journal de l'Agence, "AI and property listings: the government restates the rules of use", 2026.

Yanport compared its automated valuation model with two general-purpose AI models, Gemini 3.5 Flash and Claude Opus 4.8, across 1,071 properties. The gap is clear. The median error is 10.5% for the dedicated tool, versus 15.0% and 15.3% for the two models. On the hardest cases, one property in ten, the error climbs to 30% for the dedicated tool and 44-48% for the general-purpose models. The study also points to three deeper flaws. General-purpose models systematically overvalue. They give different figures for the same property depending on how the request is phrased. And they ignore qualitative factors, like renovation potential, unless explicitly told about them. Pasting a ChatGPT estimate into a valuation report therefore means signing off on an unstable number that leans high. AI still has a real role in explaining and presenting a valuation. The number itself should come from market data and your knowledge of the area. Source: Immobilier 2.0, "Can AI be trusted to value a property?", 2026.

Speaking to TechCrunch, Mati Staniszewski, CEO of ElevenLabs, whose synthetic voices power many companies' phone lines, said businesses should tell customers when they're talking to an AI. Coming from a company that sells voices indistinguishable from humans, that's a notable stance. In Europe, it's no longer optional anyway: the AI Act already requires informing people that they're interacting with an AI system. The same week, Ron Johnson, who built the Apple Stores, said he doesn't believe in shopping handed off to AI agents. In his view, nobody will let an agent choose and buy a $2,000 laptop for them. Agents will narrow the field, then send better-informed customers into stores. The parallel with real estate is direct. An assistant can handle first-level questions and shortlist properties, as long as it introduces itself as one. The viewing, the negotiation and the decision remain human moments, and that's where trust is built. Sources: TechCrunch, "Twenty minutes with the CEO of ElevenLabs, now reportedly valued at $22 billion" and "The man who built Apple's stores doesn't buy Silicon Valley's bet on AI shopping", 2026.

On September 22, idealista announced an agreement with Orpi: listings from the 1,250+ agencies of France's largest agency network will be published on idealista.fr. The deal comes barely a week after the portal launched in France. idealista claims more than 200 million monthly visits across Europe. We covered idealista's arrival a few weeks ago. With Orpi, the portal instantly gains what every newcomer lacks: inventory. For an independent agency, testing it becomes more worthwhile, since buyers will now find a substantial offer there. The rule stays the same: track where your leads come from from day one. Otherwise, six months from now you won't know whether this channel brought in listings or just more distribution work. Sources: Journal de l'Agence, "Orpi joins idealista: France's No. 1 agency network joins Europe's leading property portal" and Immo Matin, "Orpi joins Idealista a week after its French launch", 2026.

Meta unveiled its Meta VR Glasses, 100-gram virtual reality glasses with micro-OLED displays and a compute puck connected by cable. They'll sell for $1,299 starting spring 2027. The weight, the bulk and the controllers that made headsets tiring after a few minutes are largely gone. As a former VR developer, I've been waiting for this. The main obstacle to immersive viewings was never image quality, it was the headset nobody wants to wear. Let's be clear-eyed though: at that price and on that timeline, these glasses won't be in agencies for a long while. The useful signal is elsewhere. A 3D tour captured today with accurate volumes can be reused tomorrow on these new devices. Better to invest in capture accuracy than in any particular display format. Sources: The Verge, "Meta's next VR device isn't a headset — it's glasses" and Numerama, "I tried Meta's VR glasses and it's exactly what Apple should have done with the Vision Pro", 2026.

The French government reminds everyone that AI doesn't exempt anyone from fair dealing. A study puts numbers on what general-purpose models' valuations are really worth. Two tech executives argue for transparent AI that prepares the decision without making it. The era of discreet AI is ending. What matters now is being able to say exactly what it did, and what it didn't. So: if a client asked you tomorrow which part of your listing came from an AI, could you answer precisely?