Real Estate Portals' ChatGPT Apps in 2026: How Zillow, Redfin & Realtor.com Are Redirecting Leads

Consider an agent like Marcus, who works a mid-sized suburban market outside Columbus. For most of his ten-year career, his funnel looked the same as everyone else's: a Zillow Premier Agent subscription, a handful of Google Ads, and a CRM full of leads who'd clicked a "contact agent" button on a listing they'd probably already forgotten. In the spring of 2026, three of his last five buyer inquiries mentioned, unprompted, that they'd "already asked ChatGPT about the neighborhood" before calling him. None of them had touched Zillow's search filters.

That shift — buyers doing their early-stage research inside a chat window instead of a portal's search page — is no longer a hypothetical. It's the direct, structural consequence of a string of corporate decisions made between October 2025 and June 2026, only three of which were actually "an app inside ChatGPT." The rest of the story — Zillow's second, separate AI product; Homes.com's decision to opt out of ChatGPT entirely; Google's re-entry through paid listing ads; and an unresolved legal fight over who's even allowed to hand listing data to an AI model in the first place — is what most coverage of this topic leaves out. This is the fuller picture.

The Full Timeline: Six AI Moves in Nine Months

The sequence matters, because each company built its move with the previous ones' controversies — and successes — in mind:

What Actually Happens When Someone Asks ChatGPT to Find Them a House

All three ChatGPT integrations run on OpenAI's Apps SDK, a developer framework built on top of the Model Context Protocol (MCP) — the same open standard, originally released by Anthropic in late 2024, that lets AI models pull live data from outside systems and render an interactive interface inside the chat instead of just returning text. In practice: a user types "Zillow, show me three-bedroom homes in Denver under $600,000," ChatGPT recognizes the app trigger, relays the query to Zillow's servers, and Zillow sends back a rendered widget — a map, photo cards, pricing — that displays inside the conversation. Zillow has stated that the listing data itself does not get passed to or stored by ChatGPT in that exchange; the widget is served from Zillow's own systems and displayed inside an OpenAI-hosted frame.

That distinction — "displayed through ChatGPT" versus "handed over to ChatGPT" — is the entire crux of the legal argument covered in the next section. It's also worth knowing before you evaluate any of these apps that OpenAI's own broader rollout has been uneven: apps are limited to English-language markets outside the EU, available on Free, Go, Plus, and Pro plans, and — as of the Realtor.com launch in March — OpenAI had only just begun accepting general third-party app submissions two months earlier, expanding what had been a curated group of seven pilot partners into an open directory.

The IDX/MLS Licensing Fight Nobody Fully Resolved

Zillow's launch didn't just introduce a new consumer feature — it reopened a legal argument the industry hadn't had to have in twenty years. WAV Group co-founder Victor Lund published a detailed critique days after the launch, arguing that Zillow's IDX license only permits displaying MLS data on Zillow.com and Zillow's own mobile apps, not on a third-party platform like ChatGPT. NAR's own MLS Handbook requires that "all displays of IDX listings must also be under the actual and apparent control of the Participant" — meaning the broker or portal has to have actually built, or directed the building of, the display, under an agreement giving them authority over what's shown and how. Lund's objection wasn't just about where the data physically sits; it was about process. He pointed out that no MLS committee or broader broker forum appears to have publicly reviewed the integration before launch, and that Zillow's own compliance claim — "we're 100% in control of the experience" — was, in his view, exactly the kind of unilateral, self-graded compliance the IDX framework was designed to prevent.

The National Association of Realtors didn't take a side. In an Oct. 21, 2025 statement, NAR said each MLS is individually responsible for assessing whether a given technology complies with its own IDX policy, and pointed MLSs toward three questions: whether MLS data is being transmitted to an unauthorized party, whether the displaying participant maintains genuine "control," and whether the display meets standard IDX disclosure requirements. NAR did clarify one thing in Zillow's favor — that displaying MLS data on mobile apps is already permitted under existing policy — without settling whether a chat interface inside a separate company's product counts as an extension of that.

MLSs split in response. North Carolina's Canopy MLS said Zillow's app remained compliant with its rules "at this time." Stellar MLS, Georgia MLS, and NTREIS said they were still reviewing the question internally. Several others didn't comment publicly at all. At least one agent, working through the Houston Association of Realtors, filed a formal complaint that HAR confirmed it was reviewing. Separately, when Real Estate News tested the Zillow app directly, reporters found they could prompt it to compile and export a downloadable spreadsheet of roughly 100 listings pulled from the connector — the kind of bulk, structured extraction that a standard web browser showing Zillow listings simply can't do. Lund flagged that capability specifically, warning it could let third parties scrape and repackage listing data for commercial use in ways IDX policy was never written to anticipate.

Realtor.com's March 2026 launch reads like a direct answer to all of this. The company built in a "limited preview" ceiling on listing data, an explicit contractual prohibition on using MLS data to train ChatGPT's underlying model, and public statements from its Realtor.com Next team framing the integration as designed to keep MLSs and listing agents, not the portal, at the center of the transaction. Homes.com went a step further and sidestepped the question entirely by keeping its AI inside its own walled ecosystem, where none of the third-party-transmission concerns apply in the first place.

Google Enters the Fight, From a Completely Different Angle

While the ChatGPT apps grabbed headlines, Google was building a parallel — and arguably more consequential — threat to portal traffic: putting home listings directly inside Google search results and ads, without requiring a click to any portal at all. A limited pilot quietly launched in late 2025 through a partnership with HouseCanary, an AI-enabled property data company, and its ComeHome.com listing site. When real estate analyst Mike DelPrete first reported the pilot in December 2025, Zillow's stock dropped nearly 9% in a single session, and WAV Group's Victor Lund raised the same IDX-control objection he'd made about ChatGPT: that turning listings into ad inventory on a global ad network was never what the IDX cooperation agreement was meant to allow.

That first version drew pushback specifically because HouseCanary was accessing listing data through its own brokerage status rather than explicit MLS and brokerage consent. Google paused it. It came back in spring 2026 in select markets — Miami, New York, Austin, Los Angeles — this time built exclusively on opt-in participation from MLSs and brokerages. On June 11, 2026, Google took the rebuilt version, branded as enhanced Local Services Ads for home listings, nationwide across all 50 states. Buyers now see pricing, photos, and core property details directly inside Google's search interface, with a button to call, message, or book a tour with a local agent — no portal visit required. The announcement triggered another round of selling: Zillow's stock fell 4.9% and CoStar's fell 3.6% on the day of the news, per Barron's. Zillow's public response was that roughly 80% of its traffic already arrives directly rather than through search engines; a Redfin spokesperson argued that serious buyers need more than a single search result, since exclusive inventory is what actually differentiates a portal.

Whether that confidence holds is genuinely an open question analysts are split on — Goldman Sachs has called the long-term risk to portals real even while seeing limited near-term impact, while Piper Sandler and others have called the immediate concern overblown. What's not in question is that Google, Homes.com, and Zillow are all now running live, MLS-sourced AI or AI-adjacent search experiences that sit entirely outside the ChatGPT conversation this topic usually gets reduced to.

Where the Leads Are Actually Going: The Data

FlyDragon's 2026 State of AI SEO in Real Estate report is the largest published study of this shift: 8.2 million tracked search queries across 192 metros, 12,400 AI-generated responses collected from ChatGPT, Perplexity, Gemini, Claude, and Google's AI Overviews between January and March 2026, a 4,180-respondent buyer survey, and 42,180 tracked leads run through to a close/no-close outcome. Worth saying plainly: FlyDragon sells AI-visibility services to agents, so a report titled "91% of agents are invisible to AI" is also, transparently, a marketing document. That doesn't make the underlying traffic and close-rate numbers wrong — several are corroborated by Conductor's independently run report below — but it's a reason to treat the framing with the same skepticism you'd apply to a portal telling you their own leads convert better than everyone else's.

With that caveat stated, the numbers are hard to wave away. The report found that 67% of homebuyers now use an AI tool as their primary research method before contacting an agent, up from 17% eighteen months earlier — a faster swing than mobile adoption or MLS digitization saw in their own early years. It also found that 61.3% of buyer-side real estate searches now begin inside an AI engine rather than a traditional search engine. Session-replay analysis of 12,000 buyer journeys showed the average buyer asks 8.7 questions in a single AI conversation before narrowing to a two-to-three agent shortlist, and 71% of those questions are hyper-local — "which agent knows the Lincoln Park condo market," not "who's a good real estate agent."

Conductor's independently run 2026 AEO/GEO Benchmarks Report — analyzing 21.9 million Google searches, 13,770 domains, and 100 million AI citations across ten industries, first published in November 2025 and updated through April 2026 — backs up the directional story from a different angle. Real estate triggers a Google AI Overview in just 4.48% of searches, the lowest of any industry Conductor tracked (healthcare tops the list at 48.75%), because real estate queries tend to be transactional and hyper-local in a way that resists a tidy AI-generated summary. That's precisely why buyers are bypassing AI Overviews altogether and going straight into an open-ended AI conversation or a dedicated app instead. Conductor's data also offers one useful reality check on brand power: despite Zillow's scale, it doesn't crack the top five most-cited domains in real estate AI answers, though one breakdown of the underlying data found it still captures roughly a 7.36% brand-mention share — AI models recognize "Zillow" as a name even when they're not linking to zillow.com directly.

Conductor's report also contains the single most useful data point for anyone deciding what content to actually write: across the industries studied, data-rich guides built around tables earned a 67% AI citation rate, comparison content 61%, and FAQ-formatted content 58% — while thought-leadership and opinion pieces, however well-written, earned only 18%. If you're going to spend an afternoon writing anything meant to get cited by an AI model, a specific, numbers-driven neighborhood comparison beats a think piece by more than 3-to-1.

Before You Panic: What the Data Doesn't Say

It's worth sitting with an inconvenient fact that gets left out of most coverage of this topic: OpenAI's own third-party app ecosystem — the very apps this article is about — has had a rough six months. Bloomberg's late-March 2026 reporting, based on interviews with app makers including several launch partners, found that despite more than 300 available integrations, the apps are hard to find inside ChatGPT, deliver little measurable traffic back to partner companies, and have been deliberately limited in functionality by partners unwilling to hand OpenAI control of customer relationships and payments. An OpenAI spokesperson told Bloomberg the company recognizes the developer experience "needs to improve." None of that erases the underlying behavioral shift — 61.3% of real estate searches starting in an AI conversation is a real, independently corroborated number — but it does mean "get listed in the Zillow ChatGPT app" is a much smaller lever than "be the source AI models already trust when they're forming an answer," which happens well upstream of any single portal's app.

It's also worth noting that AI referral traffic remains genuinely small in absolute terms — Conductor puts it at roughly 1.08% of all website sessions across the industries it studied, with ChatGPT responsible for about 87.4% of that referred traffic. The volume is tiny; what FlyDragon's close-rate data suggests is that the intent behind it is unusually high. Both things are true at once, and neither on its own justifies either ignoring AI search or treating it as the only channel that matters this year.

What Agents Should Actually Do About It

None of the levers below depend on which portal a buyer happens to open first, which is exactly why they're worth the time. Four things carry real evidence behind them:

1. Make your portal profiles complete, not just present. Because roughly 61% of real estate URLs in publicly available AI training data trace back to the five largest portals, an AI model's baseline picture of "who the agents are" in your market comes overwhelmingly from those profiles. A profile with three reviews from 2022 and an unclaimed headshot is close to invisible next to a fully claimed profile with recent reviews, verified past sales, and a clear specialty description.

2. Add structured data an AI model can parse without guessing. RealEstateAgent and LocalBusiness schema on your homepage and about page, Person schema linking your credentials and certifications, and FAQPage schema on pages that actually answer buyer questions all give a language model a machine-readable summary card instead of a wall of unstructured text. A simple starting block looks like this:

{
  "@context": "https://schema.org",
  "@type": "RealEstateAgent",
  "name": "Your Name",
  "worksFor": {
    "@type": "RealEstateAgency",
    "name": "Your Brokerage"
  },
  "areaServed": "Columbus, OH",
  "knowsAbout": ["First-time buyers", "New construction", "Relocation"],
  "aggregateRating": {
    "@type": "AggregateRating",
    "ratingValue": "4.9",
    "reviewCount": "62"
  }
}

Keep listing pages on RealEstateListing schema and reserve Article schema for genuine blog and market-report content — mixing the two confuses the parser and can knock a page out of the query type it should be matching.

3. Make your name say the same thing everywhere. Your name, brokerage, title, and service area should match, word for word, across your website, Google Business Profile, portal profiles, and any directory listing. Small mismatches — "team lead" on one platform, "agent" on another — read as inconsistency to a model cross-checking sources, even when a human would read them as interchangeable.

4. Write the specific thing, not the general thing. Per Conductor's citation-rate data above, a short, current post with an actual table — "Median price and days-on-market by neighborhood, updated June 2026" — is measurably more likely to get pulled into an AI answer than a well-written opinion piece on the market. Run a monthly check by asking ChatGPT, Perplexity, and Gemini a buyer-style question about your market and noting who gets named; the same three or four names tend to recur, and that list tells you exactly which sources a model in your market already trusts.

Frequently Asked Questions