Zillow’s ChatGPT app tests the long-standing broker control principle central to IDX
For more than 20 years, I have been a student of MLS policy. Every year, there is a process where brokers, agents, lawyers, MLS executives, and technologists bring ideas for change to the National Association of REALTORS® MLS policy. These proposals undergo healthy debate and either result in a policy change or they do not. This deliberative approach has served the industry well, ensuring that changes are considered carefully and collectively.
Lessons from Past IDX Policy Debates
One memorable debate involved Realogy and other franchise organizations requesting a change to the IDX policy to allow an IDX search on their franchise brand websites. On the surface, it seemed reasonable. These franchise brands are connected to brokerages; so why not let them host IDX listings on a national brand site? The industry reviewed the request and ultimately said no. The process of debate allowed participants in the broker reciprocity program to discuss and decide, and not changing the policy was a deliberate choice. The primary reason was that franchise brands are not themselves brokers participating in the MLS, so allowing them an IDX feed would violate the principle that IDX sites must be under the control of a participating broker.
Franchise companies found work-arounds to this limitation. For example, REMAX.com devised a creative solution: when a consumer enters a search on remax.com, the site actually hands off the query to the local RE/MAX brokerage’s own IDX website for that area. The consumer might not realize it, but the search results are displayed on the local broker’s site (with a remax.com wrapper). This preserves compliance because the listing data ultimately appears under the control of a broker’s domain, not purely on the franchise’s central site.
Another debate centered on putting an IDX property search on Facebook. Industry policymakers carefully crafted language to allow this, with conditions. The broker or agent’s Facebook page had to clearly identify the brokerage/agent (in the page name or URL, e.g. facebook.com/victorlund for my page), display their branding and headshot, and the actual listing data had to be embedded in a way that Facebook itself couldn’t crawl or reuse (often via an iframe). In essence, even though Facebook is a third-party platform, the IDX display on a Facebook business page was considered under the broker’s control because the page is managed by the broker/agent and not by Facebook. The guiding principle in these cases was that the display of MLS listing data must remain under the control of a broker who participates in that MLS.
The ChatGPT Integration: A New Challenge to Broker Control
Fast forward to today’s controversy: Zillow’s new integration with OpenAI’s ChatGPT for listing search. I use the word integration because the system works by connecting a server that houses broker listing data (Zillow’s MCP server, in this context) to an application inside ChatGPT via OpenAI’s developer toolkit. Let’s unpack how this works and why it’s raising eyebrows in the MLS community.
When you open ChatGPT, you’re on the chat.openai.com domain – clearly not under the ownership and control of any real estate broker. Zillow did not even attempt to create a subdomain like chat.openai.com/zillow or a co-branded URL. The entire experience runs on ChatGPT’s domain. This fact alone raises a red flag. Historically, as noted, IDX policy has insisted that the website or page displaying IDX listings be under a broker’s control. The ChatGPT domain is obviously not controlled by Zillow (even if Zillow is providing the data to the ChatGPT app). By that simple litmus test, many MLSs could determine that Zillow’s new ChatGPT integration violates IDX rules, just as they would if, say, remax.com directly displayed another broker’s listings without routing through that broker’s site.
We have a recent precedent for comparison. Microsoft’s Bing (the search engine) experimented with a real estate search feature that pulled in property listings via APIs from Zillow, Realtor.com, and Redfin. The listings displayed on a Microsoft site (with links back to the brokers). The industry “flew off the handle” at this, because consumers were able to search and view MLS-sourced listings on a platform completely outside broker or MLS control (the Bing domain). Even though each listing result linked back to the listing broker or portal, the initial display on Bing was enough to trigger backlash. The key issue was the same: the domain where the search and initial results occurred was not under the participating broker’s control. Ultimately, that experiment was shut down or reworked due to industry pressure.
Zillow’s ChatGPT app presents a similar scenario. The search and results are happening on ChatGPT’s domain, not on a broker’s own website. By the same reasoning as the Bing case, MLS organizations could very well choose to revoke Zillow’s IDX feed access for violating display rules. If an MLS would not allow a franchise site or a search engine to host listing displays on their domain, why would it allow ChatGPT?
How ChatGPT Gets the Data: Indexing vs. Scraping
It’s worth understanding how ChatGPT is obtaining listing information even without Zillow’s official app. Under the hood, ChatGPT leverages Bing’s search index to retrieve information from the web. Bing’s search engine continuously crawls websites (like Zillow, Realtor.com, Redfin, and broker sites), indexing their content in a massive database. When ChatGPT is asked a real estate question (outside of the Zillow plugin), it can query Bing’s index and find relevant content from those indexed pages. Essentially, ChatGPT can scrape the needed details from a broker’s listing page via Bing’s cached data, and then answer the consumer’s question using that information. The consumer doesn’t realize it, but ChatGPT might be pulling data directly from an IDX listing on a broker’s site, without the consumer ever visiting that site or the broker having control over the display of that data in ChatGPT’s answer.
Many people have wondered how ChatGPT could possibly know about current real estate listings if it’s not directly connected to Zillow’s database. The answer: it piggybacks on Bing’s web indexing. ChatGPT itself isn’t crawling Zillow or MLS sites, but Bing did, and ChatGPT can ask Bing’s index for the info. In effect, Bing indexed the listings, and ChatGPT is using that index to produce responses. Neither Bing nor ChatGPT has a license to do this.
There’s an important distinction to make between indexing and scraping. When a search engine indexes your site, it’s generally allowed under fair use and with the site’s permission (sites can opt out via robots.txt, etc.). The search engine might show snippets and links, enticing users to click through to the source site. What ChatGPT is doing with Bing’s index goes a step further: it extracts and presents the full listing information directly in the conversation, so the user might never click through to the source. This starts to look like scraping – using the data in a new context outside the source site’s control or permission. This arguably violates the terms of service of the websites from which the data is taken. It could even run afoul of laws like the Computer Fraud and Abuse Act (CFAA), DCA, the Federal Trade Commission Act (FTCA) provisions on unfair/deceptive practices, and the Stored Communications Act, especially if the data usage is unauthorized or misrepresents the source. The legality is a gray area, to say the least, and it underscores how uneasy this type of data use might make the industry.
Using Zillow’s App in ChatGPT: Who Controls the Experience?
Now, setting aside the broader web-scraping issue, let’s focus on what happens when a consumer actually uses the Zillow app within ChatGPT, effectively the officially ChatGPT sanctioned route that Zillow has introduced. To activate the Zillow plugin, the user must first be logged in to ChatGPT (with an OpenAI account). At no point are they required to log into a Zillow account or even visit Zillow’s website directly. The entire interaction starts and stays within ChatGPT’s interface. From the consumer’s perspective, they’re using ChatGPT (just as they might use AOL, Bing, Google, or any other online service) and simply invoking Zillow as a tool inside that environment. No reasonable consumer would think that by logging into ChatGPT, they have somehow logged into Zillow or an MLS site. In my opinion, it’s an entirely distinct application from the consumer point of view.
Once logged into ChatGPT, the user adds or enables the Zillow app and then types a query (for example: “Zillow, show me homes for sale in Seattle under $800k”). What’s happening behind the scenes is that ChatGPT interprets this prompt, recognizes that it needs Zillow’s data, and then calls the Zillow MCP listing server for results. The response is then displayed within the ChatGPT interface as a Zillow-styled output (complete with listing details, photos, and attributions as per IDX rules). Visually, it might appear as if Zillow’s website is embedded in the ChatGPT window. However, crucially, the content is still being delivered through the ChatGPT application. The URL in the browser remains chat.openai.com/…, not zillow.com. In other words, ChatGPT is acting as a third party middleman, fetching data from Zillow’s servers and displaying it on ChatGPT’s domain. Today’s IDX policy specifically rejects third party handling of data.
Zillow might argue that when a user invokes the Zillow app on ChatGPT, it’s analogous to a consumer using a web browser to go to Zillow.com, essentially just another method of accessing Zillow’s services. But I would argue this is a stretch. ChatGPT is not a general web browser; it’s a conversational AI platform. The experience of using Zillow via ChatGPT is fundamentally different from navigating to zillow.com in Chrome or Safari.
IDX policy was written long before such AI platforms existed, so we are in uncharted territory, but we can draw analogies. To my mind, using Zillow through ChatGPT is not the same as a consumer going directly to Zillow’s website under the broker’s control. As an analogy, going to a franchise’s national site (e.g. ERA.com) is not the same as going to an individual broker’s site (e.g. HuntRealEstate.com). The context and control are different.
An Exclusive Deal: Is It Fair?
It’s also important to note the exclusivity of this arrangement. As of now, you can only invoke Zillow in ChatGPT for real estate search. There is no equivalent plugin for other brokerages like Compass, eXp Realty, Redfin, or any independent broker. Zillow and OpenAI (ChatGPT’s creator) have essentially struck a deal to make Zillow the first (and for now, the only) real estate search tool on the platform. It brings to mind other deals OpenAI has made. For example, ChatGPT reportedly pays Reddit around $70 million per year in a multi-year licensing arrangement for Reddit’s data. We don’t know the terms of the Zillow-OpenAI arrangement. Does Zillow pay for the privilege to be the exclusive real estate app, or does OpenAI pay Zillow for access to the data? The details haven’t been disclosed publicly.
What we do know is that this exclusivity won’t last forever. OpenAI has introduced an Apps SDK for ChatGPT, and presumably more apps and connectors will be coming. In fact, I spoke with a technology vendor in the real estate space (one that powers around 600,000 IDX websites and an equal number of mobile apps for brokers and agents). They could build a ChatGPT integration just like Zillow’s with relative ease using OpenAI’s toolkit. The only thing stopping them right now is that OpenAI’s ChatGPT Apps platform is in a limited preview. Zillow got in early, and others have to wait. (Perhaps this exclusivity or head-start lasts until December, if rumors are to be believed.) From an industry fairness perspective, it’s problematic that one giant broker/portal gets a first-mover advantage in this new AI-driven channel, while others are effectively locked out in the near term.
For those interested in the technical side of how the ChatGPT Apps work, OpenAI’s announcement provides some insight. Read OpenAI’s introduction to ChatGPT Apps for an overview of how developers can connect external services like Zillow to ChatGPT.
“Reasoning” with the MCP Server vs. Traditional API Calls
There’s another technical nuance here: the nature of ChatGPT’s interaction with Zillow’s data is through “reasoning” rather than a straightforward API call. Traditional APIs (such as the RESO Web API commonly used in real estate) are rigid and explicit. You ask for specific data with specific parameters, and you get a response. The ChatGPT-Zillow integration is built differently. When a user asks a complex question, ChatGPT doesn’t just fill in fields on an API call; instead, it engages in a kind of dialogue with Zillow’s MCP server. The MCP (Model Context Protocol) server is designed to allow an AI agent to interact with the data in a more flexible, goal-oriented way.
In simple terms, the AI “agent” (ChatGPT) is interpreting the user’s intent and then dynamically deciding how to fulfill that request using the tools and data Zillow has exposed. It’s not blindly calling a single endpoint; it might reason, apply filters, refine queries, and use multiple steps to get to the answer the user wants. This is a new frontier: two machines collaborating, with the AI agent orchestrating a sequence of actions. The result is a more natural, conversational search experience for the user, but it’s also a more complex and less transparent interaction than a typical API call.
Zillow, presumably, has put guardrails on what the ChatGPT agent can do through its MCP server. They might restrict certain data fields from ever being accessible (to comply with MLS rules), and they likely audit the interactions. We have to trust Zillow’s implementation, because no MLS or third-party is overseeing those technical details in real time. Zillow claims that the app is IDX-compliant and that they have full control over what data is served (more on their specific claims in a moment). We can hope that Zillow’s guardrails are robust and aligned with industry rules and norms. But it’s worth noting that this entire AI-driven interaction model is outside the direct supervision of any MLS. MLSs provide IDX feeds under certain rules, but they haven’t historically had to think about an AI reasoning agent pulling data in a conversational context.
IDX Reciprocity and the Spirit of Fair Play
Stepping back, let’s recall why IDX exists at all. Broker reciprocity (IDX) is fundamentally about brokers cooperating and sharing listing exposure. As my friend Lennox Scott (an early proponent of IDX policy) has explained, the exchange of value is simple: “I’ll let my listings appear on your website, and in exchange, you let your listings appear on mine.”It’s a mutual, fair deal that increases exposure for all listings and provides consumers a more comprehensive search experience, regardless of which broker’s site they use. Crucially, though, this deal assumes a level playing field with each participating broker that abides by the same rules and has the same opportunity to display each other’s listings on an equal footing.
When I look at the Zillow-ChatGPT integration, I have serious concerns through this lens of fairness and reciprocity. In my view, this integration is one broker (Zillow) taking IDX data into a new channel that no other broker currently has access to, and doing so in a way that skirts the established rules (like the domain control issue). The brokers who contribute their listings to the MLS never specifically consented to having their listings show up via an AI chatbot on OpenAI’s domain. They consented to IDX displays by other brokers under the IDX policy framework, which did not contemplate something like ChatGPT.
To me, this feels like Zillow is pushing the boundaries of IDX reciprocity to the breaking point. Yes, Zillow is a broker and an MLS participant, so it has the right to display other brokers’ listings on its own website or apps under IDX. But is a ChatGPT app an extension of Zillow’s website/app, or is it an entirely new, third-party platform display? Reasonable people can disagree, which is why this is such an important debate to be having. I have chosen to side with the brokers and MLS leaders who view this as a step too far. In fact, some have even labeled it a “bridge too far.” It just doesn’t feel like the fair, mutual exchange of the original IDX spirit.
I suspect we will see a rapid proliferation of similar integrations soon. In the coming weeks and months, we could see real estate apps on Anthropic’s Claude, on Perplexity.ai, on Microsoft’s Copilot, on Google’s upcoming Gemini, on IBM’s Watson, and on countless other AI platforms. The idea of conversational search for homes is out of the bag, and many tech providers and brokers will want to deploy their own solutions. Today, our industry lacks a clear framework or guidelines for MLSs to approve or deny such AI-based IDX displays. We are, in effect, operating in a policy vacuum. In an article I wrote on October 7, 2025, titled “Zillow Seeks Forgiveness, Not Permission: MLSs Must Enforce Cooperation in the AI Era”, I argued that Zillow went ahead with this ChatGPT app without broad industry permission, betting that it’s easier to ask forgiveness afterward than to get permission beforehand. I called on MLSs to step up and enforce the traditional cooperation principles in this new AI context.
I’m not persuaded by the argument some have made that “ChatGPT (or a similar AI) is just like a search engine or an operating system, so it shouldn’t be regulated by IDX rules.” In my opinion, that’s a flawed analogy. ChatGPT is neither just a search engine nor just an OS; it’s a new type of platform. We shouldn’t let unfamiliarity or technicality prevent us from applying the same core principles of fairness and broker control. The real estate industry, through its collective process, should deliberate and update policies to address this new breed of technology. Until that happens, Zillow could easily pause the ChatGPT integration (i.e. turn off OpenAI’s access to its MCP server) while the industry sorts out the rules collectively. That would be the cooperative thing to do.
Zillow’s Response and My Rebuttal
It didn’t take long for Zillow to respond to the concerns raised by myself and others. In the wake of my initial article and others’ commentary, Zillow’s industry relations team went into action, reaching out to stakeholders and publishing a public defense of their ChatGPT app. They released an article on Zillow Group’s website titled “How Zillow’s App in ChatGPT expands listing reach and protects industry rules”. In it, Zillow makes a few key assertions:
“Zillow is not sending OpenAI a feed of MLS data.”
Zillow emphasizes that they are not directly handing over MLS data to OpenAI. Instead, as they explain, OpenAI’s system sends a request to Zillow for information when a user prompts Zillow in ChatGPT, and Zillow sends back the results which are then displayed within ChatGPT. From a purely technical standpoint, this is accurate. Zillow isn’t giving OpenAI an ongoing data feed. However, to many of us, this is a distinction without a difference. Whether the data is pushed or pulled, the end result is that MLS listing information is being displayed on a non-broker controlled domain. The consumer doesn’t know or care about the technicalities; they’re seeing listings via ChatGPT. In other words, it doesn’t matter who “picks up” the data: the outcome (listings showing on ChatGPT) is the same.
“Zillow worked directly with OpenAI to ensure this experience was built with the industry’s rules and safeguards at its core.”
Zillow claims it took care to build the ChatGPT app in compliance with industry rules (IDX policies, etc.), and that it has full control over the app’s experience. The question I have is: under what authority is Zillow interpreting and enforcing “the industry’s rules”? Typically, MLS rules are enforced by the MLS themselves, not unilaterally by one broker’s interpretation. Zillow having “100% control” over the experience (as they tout) is precisely the concern: it means no outside oversight. We are essentially being asked to trust Zillow’s word that everything is compliant. This is not how industry safeguards usually work. Compliance is usually verified by the MLSs and brokers collectively, not just asserted by one participant.
“We worked closely with industry stakeholders to ensure that this experience complies with MLS data standards…”
This statement raises more questions than it answers. Who are these “industry stakeholders” that Zillow worked with? Were MLS executives involved, and if so, which ones? Were brokers consulted beyond Zillow’s own team? From my vantage point, no MLS committees or broader broker forums publicly discussed this integration beforehand. If some stakeholders gave input, it feels like it happened in a back room. The lack of transparency around who blessed this integration (and by what authority) is exactly why many in the industry are uncomfortable. It gives the appearance that Zillow made a deal or got an approval from a select few, and now presents it as “complying with MLS standards.” But compliance is usually determined by collective agreement and explicit policy, not by private assurances.
In short, Zillow’s carefully worded response hasn’t convinced me (or many others) that this ChatGPT integration truly respects the letter and spirit of IDX rules. It comes down to whether we believe this is just an “extension of Zillow’s website” (as Zillow portrays it) or a fundamentally new type of third-party display that needs its own policy vetting.
A Better Path Forward: MLS-Controlled MCP Servers
I believe there is a better way to embrace this new AI technology without sacrificing the principles of broker reciprocity and MLS oversight. The solution I advocate is for MLS organizations themselves to deploy and control the MCP servers that interface with generative AI platforms. In other words, instead of Zillow (or any single broker) being the gatekeeper of MLS data in the AI world, the MLS would operate the gateway for AI access to listings leveraging the fair display guidelines. This could ensure that all brokers have equal opportunity to leverage AI channels, under rules that are set collectively by the MLS membership.
I wrote a paper on this topic back in July, titled “Why Every MLS Needs to Understand MCP Servers Before Someone Else Builds One Without You”. The core idea is that if the industry doesn’t proactively build the infrastructure for AI (like MCP servers managed by MLSs or broker cooperatives), then a large player (like Zillow, or others in the future) will do it unilaterally and potentially consolidate even more power in the process. If MLSs provide a standardized, policy-compliant way for AI systems to query listing data (with proper guardrails), then any broker or authorized app can use it and we maintain a level playing field and oversight. It would uphold the spirit of IDX reciprocity by keeping the MLS in the driver’s seat of how listings are used in emerging technologies.
Conclusion: Choose a Side or Wait and See?
Zillow’s move to integrate with ChatGPT has, without question, gotten a jump on the rest of the real estate industry. It’s a bold play that tests the boundaries of current policy. Now the industry has a choice in how to respond. Will MLSs and brokers simply wait and see, effectively allowing this precedent to stand? Will they push back and say “No, this isn’t allowed under our rules” and force changes or the shutdown of such integrations? Or will they say “OK” and perhaps hurriedly adapt the rules to formally permit this kind of app (maybe with additional conditions)?
As for me, I’ve made my stance clear. I stand beside the brokers, MLS executives, and others who view Zillow’s ChatGPT integration as a bridge too far under the existing IDX framework. I believe in the IDX principles of fairness, mutual benefit, and broker control over their data’s presentation. Those principles have served our industry well for decades of internet disruption, and they should not be abandoned just because the technology has taken another leap forward. We can embrace innovation like AI in home search and still uphold cooperation and fairness. But it requires deliberate action from industry leaders, not just one big player forging ahead alone.
In this debate, I’ve picked my side. Now the question is, how will the rest of the industry respond?
