5 Reasons ChatGPT Isn’t Going to Take Over Real Estate Search, Unless...

Ever since generative AI went mainstream, a familiar anxiety has crept through sections of the real estate industry. If ChatGPT can write essays, code software, and scrape and present real estate agent listings, why wouldn’t it also replace portals?

It is a fair question. Portals have extracted all that money from agents over the years just by owning the top of the real estate funnel. To some of their customers operating at the transaction end of that funnel, LLMs with their radically new approach and sensational headlines look like knights in shining armour come to free agents of their subscription shackles.

But on closer inspection, the idea that ChatGPT will become a full-blown real estate listings destination quickly unravels. AI will change how people interact with property data, but taking over the search layer entirely is a very different proposition. Here are five reasons why...

1. Listings Data Is Compute-Heavy and Operationally Painful

Property listings look deceptively simple until you try to scrape, interpret and serve them at scale.

Keeping a national or international catalogue fresh requires constant crawling, deduplication, image processing, attribute normalisation and compliance checks. Listings expire, prices change, photos are swapped and agents make mistakes. All of that has to be handled in near real time.

For LLMs, scraping the web to replicate what portals already do would be brutally computationally intensive. As EIV founder, Malcolm Myers said on a recent episode of the PPW Podcast,

"It's not very efficient either in time or in money for them to basically recreate a Rightmove every time someone has a query."

For portals, paying for the people and infrastructure to make and maintain the relationships needed to get listings feeds makes sense because listings are the core product. For an AI company, it would be an enormous ongoing expense that sits awkwardly alongside a business model built on usage-based APIs and subscriptions.

2. Large Language Models Are Not Systems of Record

This is the technical heart of the issue, and it is often misunderstood. To quote the distinguished mathematician and computer scientist Stephen Wolfram, who has written perhaps the best and most accessible material on how ChatGPT works, “ChatGPT is fundamentally doing next-token prediction.”

LLMs are built to predict the next word. They are probabilistic systems, not databases. By themselves, they cannot be a reliable property catalogue.

Ask the same question twice, and you may get different answers, which is fine for brainstorming or summarisation, but deeply problematic for something as transactional as property search. There are several knock-on effects:

• LLMs are non-deterministic, which undermines trust.
• They have no native concept of which is the 'true' canonical listing in a set of duplicates
• Because they use neural networks, they cannot reliably explain why a particular property was shown.
• As recent studies have shown, they are far more vulnerable to spam and manipulation.

Portals, by contrast, are built as systems of record. They know which listing is live, who paid for it, when it expires and who is legally responsible for its accuracy.

3. AI Search Is Not Actually Better Than Portal Search

There is a tendency to conflate conversational fluency with search superiority.

ChatGPT is excellent at reasoning about a single listing you describe to it. It can discuss trade-offs, suggest questions to ask, and even help interpret floorplans, which makes it a great house-hunting assistant.

What it cannot do, at least by default, is ingest and maintain millions of listings across thousands of sources with a consistent structure.

Below: Popular LLM Manus taking over six minutes to do a home search. When it finished, it gave me a text file with no images

It also cannot freely browse every portal, crawl agent sites at scale and parse listing photos to infer features across an entire market. If you want to know whether a specific listing probably has a dishwasher, an LLM can help. If you want to filter listings from an entire postcode by that attribute, portals still do it better.

4. Portals Control Supply, Relationships and the Legal Perimeter

Even if the technical challenges were solved, the commercial ones remain.

Portals control access to the listings supply. They can and do block unauthorised scraping, and the largest players would have little incentive to allow a third party to repackage their inventory without compensation.

The legal position is also murky, as Zillow found out recently when it syndicated its MLS-sourced listings onto ChatGPT. Is an AI model that presents listings an aggregator or a search engine? There is no clear precedent for how large language models should be treated when it comes to property data.  Until that's resolved, any attempt to operate at scale would carry significant risk.

Running a listings marketplace also requires people. Even faceless aggregators historically needed large teams, local expertise and ongoing relationships with listings sites. If ChatGPT went down this path, it would quickly start to look like a portal itself, complete with sales staff, account management and market-by-market nuance.

5. OpenAI Has Already Chosen Partnership Over Disruption

This clue sits in plain sight. OpenAI has positioned itself as an enabling platform, not a vertical operator. To date, it has partnered with Zillow (full listings syndication), REA Group (conversational assistant for valuations) and Scout24 (helped build its search assistant).

Its commercial logic points towards powering other products rather than competing with them. In real estate, that means helping portals, brokerages and software vendors layer AI on top of existing inventory, not trying to own the inventory itself.

Longer term, OpenAI’s ambition looks closer to an app-store model. The AI becomes the operating system, while vertical specialists build applications on top of it.

But Hang On...

None of this is to say that OpenAI and others will be completely benign to portals in the long term. A partner can turn into a competitor, or it can subsume the lesser party over time.

LLMs can't really be reliable catalogues, but they can be very good assistants sitting over a reliable catalogue.

The app-store analogy is apt here. Zillow won plaudits for being the first real estate marketplace in the world to enter into one of these partnerships with OpenAI and have the listings on its site discoverable on ChatGPT. It does stand to learn a lot and benefit from increased traffic by being an early syndicator. But it has taken a long-term risk in doing so.

Like all portals, Zillow is a shop window business. As shopkeepers know, putting all of your wares into someone else's window can be dangerous. A shop might move locations to the big shiny mall with lots of foot traffic, but if the mall gets sued, sees traffic decline or decides to increase the rent, the shop suffers.

As Airbnb CEO, Brian Chesky said recently, big marketplace companies will soon "have to decide how to participate with platforms like ChatGPT". Do they acquiesce to OpenAI, becoming mere app creators and potentially giving up the customer relationships, or do they back their own AI, which is native to their niches and verticals?

The decision on which path portals individually or collectively take may be looked back on as pivotal in much the same way agents around the world now lament having given up their listings to the shiny, new portals twenty years ago.

One CEO who seems to have made a decision is CoStar founder Andy Florance, whose comments on his company's Q3 investor call addressed the debate between general and domain-specific technology in real estate.

"There was a time when AOL Yahoo, or eBay were ascendant and uniquely dominant and Microsoft and Google are still dominant. Though perhaps past their zenith of dominance. All of these impressive general purpose transformative technology innovations enthusiastically built real estate portals, and tried to dominate, digital real estate. All failed...

Specialized solutions often leveraging these companies' capabilities repeatedly ultimately dominated the real estate vertical. I believe the past is prologue here. There are a number of incredible generative AI companies that are building invaluable tools. Those tools will be leveraged by specialized digital real estate companies to create specialized value. The specialized digital real estate company that does it best among them will unlock huge value for its investors."

The real risk for portals is not that ChatGPT wakes up one morning and decides to become Rightmove, but that it quietly redefines the interface layer and relegates portals to being very good app makers underneath it. If consumer attention, discovery and loyalty migrate upstream to a general AI layer, portals could find themselves supplying inventory, tools and expertise while someone else owns the customer relationship.

This kind of backdoor disintermediation is far more plausible than a head-on assault, and it is exactly why decisions being made now about syndication, partnerships and proprietary AI will matter more than whether ChatGPT can run a decent property search today.