In October 2025, Zillow started running inside ChatGPT. A buyer can type a question about homes for sale and browse real listings, with photos and maps, without ever opening a browser tab. Most of the reaction has treated this as news about listings. The more useful question for a working agent is what it changes about introductions, and the honest answer is: it moves them further into Zillow's funnel, and it makes your own website the one AI surface you still control.
What the Zillow app in ChatGPT actually does
At OpenAI's developer conference on October 6, 2025, ChatGPT gained the ability to run third-party apps inside a conversation. TechCrunch's coverage lists the first partners: Booking.com, Canva, Coursera, Figma, Expedia, Spotify, and Zillow. Zillow was the only real estate company in the group.
By Zillow's own announcement, the app works like this: a signed-in ChatGPT user in the U.S. starts a prompt with the word Zillow, asks for homes in an area and budget, and gets real listings inside the chat, with photos, an interactive map, pricing, and the broker and MLS attribution that listing rules require. When the buyer wants to schedule a tour, contact an agent, or look at financing, the chat hands off to Zillow's full site.
Two details in that description matter more than the headline. First, the buyer does not have to name Zillow at all: ChatGPT can suggest the app on its own when a conversation turns to buying a home. Second, every action that creates a relationship, a tour request or an agent contact, happens inside Zillow's normal flow, not in the chat.
The scale question, stated honestly
OpenAI has said that more than 800 million people use ChatGPT every week, a figure Zillow repeats in its announcement. That is OpenAI's own number, not an independently audited one, so treat it as a company claim. But the direction is not in dispute: a very large share of everyday questions now starts in a chat window, and home buying questions are part of that.
What nobody outside these companies knows yet is how many buyers actually search for homes this way, how often, and in which markets. Early coverage does not include usage numbers, and the honest position is that the behaviour is new and the data is not public. What the launch proves is intent: the largest consumer AI product and the largest U.S. real estate portal decided listing search belongs inside the chat.
Whose funnel the app feeds
Here is the mechanism worth sitting with. When a buyer finds a home through the Zillow app in ChatGPT and taps through to schedule a tour, the introduction runs through Zillow's lead routing, under Zillow's rules and Zillow's agent relationships. The listing might be yours. The pipeline that decides which agent gets the buyer is not.
This is not a criticism of Zillow. It is how portals have always worked, now extended into a new surface. But it clarifies the decision an agent is actually facing. Inside the chat, listing discovery is portal territory, and no individual agent can compete there. The question is what an agent can still win.
The queries the Zillow app does not answer
Listing search is one kind of question buyers ask an AI assistant. The other kind sounds like this: "Who is a good buyer's agent for condos in Mount Pleasant?" "Is [name] a trustworthy realtor?" "Should I sell my Vancouver townhouse this year or wait?"
The Zillow app does not answer those. When an assistant gets an agent recommendation or reputation question, it builds the answer from the open web: agent websites, Google Business Profiles, reviews, local directories, and news coverage. We looked at that retrieval mechanism in detail in how AI search changes real estate marketing, and the short version holds here: assistants quote sources that state facts plainly, and most agent websites do not.
That is the open lane. It was open before the Zillow app launched, but the launch makes the split clean: portals now own in-chat listing search, and the recommendation queries are still decided by whichever agent gives the assistant something concrete to quote.
Why this makes your own site matter more
It is tempting to read the app as another reason the agent website is dying. The logic runs: buyers search on portals, portals now live inside ChatGPT, so who needs an agent site?
The logic fails on what a website is for. An agent site was never going to out-search Zillow on inventory. Its job is to answer the questions where you are the subject: who you are, where you actually work, what you specialize in, what your clients say, and what you know about your neighbourhoods that a portal profile cannot hold. Those are exactly the pages AI assistants read when someone asks about you or asks for a recommendation in your area.
The practical standard has changed, though. A site that is a photo, a slogan, and an IDX search box gives an assistant nothing. What gets quoted is specific, stated fact: a bio that names your areas and property types in plain sentences, area pages with real local knowledge, consistent business details everywhere, and structured data (the machine-readable facts block behind a page) that repeats all of it. We keep a running checklist of what that looks like in our AI search visibility guide.
What to check on your own site this week
Three checks, in order of effort.
First, run the recommendation test. In a fresh ChatGPT or Perplexity session with memory off, ask for an agent recommendation in your farm area, phrased the way a mover would ask it. Then ask directly about your name. If you are absent, note which sites the assistant did cite. Those are your actual competitors for this query, and they are usually not the agents you compete with on yard signs.
Second, read your own homepage and bio the way a machine would. Is there one plain sentence anywhere that says your name, brokerage, city, and specialty together? If a human skimming for ten seconds cannot find it, an assistant extracting facts cannot either.
Third, check that your site and your public profiles agree on the basics: name, brokerage, phone, areas served. Assistants cross-reference sources, and contradictions read as uncertainty. Our post on the AI search visibility gap walks through the full test with sourcing on what the surveys actually show.
What not to do about it
Two overreactions are worth naming, because both cost money.
The first is trying to compete with the portal on inventory. Building a bigger listing search, buying more IDX features, adding map filters nobody asked for. Zillow's advantage in listing search is structural, and the app extends it into a surface individual agents cannot enter. Spending there buys a worse version of something a buyer already has open in another tab.
The second is treating the launch as proof that websites no longer matter, then letting the site decay. That reads the news backwards. The app narrows what a website should try to do; it does not remove the questions a website answers. An agent with no credible site is now missing from the only category of AI answers where an individual can still be the source.
The measured response sits between them. Concede listing discovery. Own the questions about you and your neighbourhoods. That is a smaller website than many agents think they need, and a considerably better one.
The takeaway
Zillow inside ChatGPT settles one question: in-chat listing search belongs to the portals, and competing with that is not a good use of an agent's budget. It sharpens the other question: when buyers and sellers ask an AI assistant about people instead of properties, the answers come from the open web, and your website is the part of the open web you control. Treat the app launch as the deadline it quietly is, and get your own site to the point where an assistant has something worth quoting.
