One portal’s share of agent discovery traffic dropped for the first time ever. Here’s why.

A benchmark study tracking 8.2 million real estate conversations in AI search models finds that buyer behavior has changed faster than any channel in real estate marketing history. And portals are losing ground because of it.

For the first time since AI search tracking began in 2024, Zillow‘s share of agent-discovery traffic declined year over year from 41.2% to 33.8%, according to the 2026 State of AI SEO in Real Estate report.

That’s nearly one-fifth (17.5% relative decline) of its agent-discovery share gone in twelve months. The displaced traffic didn’t migrate to Realtor.com, Redfin, or another portal.

It went to AI search tools.

The report, the largest publicly published study of AI search behavior in U.S. residential real estate, spanning 12,400 AI-generated responses, 8.2 million tracked queries across 192 metros, and a 4,180-respondent buyer survey, 67% of homebuyers now use an AI tool as their primary research method before contacting an agent. 

That figure was 17% just 18 months ago.

Why is this adoption happening so quickly and why are the portals losing out?

The portal model was built for a different buyer

Zillow, Realtor.com and Redfin were built for a buyer who searches in fragments. Type a keyword. Scan a list. Click a profile. Compare headshots and review counts. Pick three, call them all, see who answers.

That buyer still exists. But a growing share of the market now behaves differently. The 2026 buyer doesn’t search portals; they have a conversation with ChatGPT.

Session replay analysis of 12,000 buyer journeys in the study reveals the average buyer asks 8.7 questions before identifying a two-to-three agent shortlist, and 71% of those queries are hyper-local.

The entire sequence from ‘where do I want to live’ to ‘who’s the best agent to work with’ happens in a single chat.

This is incredibly different from the portal journey of the past, which explains the traffic loss.

AI search is the future of how people find agents

Real estate has four characteristics that make it almost perfectly suited to AI-mediated discovery and poorly suited to the portal model that dominated the last 15 years.

The average American completes fewer than four real estate transactions in a lifetime. Buyers have no muscle memory for choosing an agent. They don’t know what to look for, what to ask or how to evaluate. This is exactly the kind of complex, localized, high-stakes question where a conversational AI outperforms a directory listing. 

The buyer needs an explainer, they need value. All things a portal experience can’t offer.

Portals flatten that nuance into standardized profile pages. AI grounding systems do the opposite. They pull agent data from Google Business Profiles, local content and, most notably, third-party consensus. 

Seventy-one percent of buyers won’t contact an agent without third-party validation, and AI models disproportionately weight exactly that: reviews spread across multiple platforms, news mentions, listicle features, and podcast appearances. 

In a conversational-search world, the tool that explains, recommends, and narrows is structurally better matched to the buyer’s actual need. This advantage is reinforced by the fact that real estate has the lowest AI Overview trigger rate of any major consumer vertical at just 4.5%, according to Conductor’s 2026 AEO/GEO Industry Benchmarks Report.

AI tools surface fewer names per search

Across 42,180 tracked leads, AI-sourced leads close at 9.6% within 90 days compared to 2.4% for Zillow Premier Agent leads and 1.8% for Google Ads. Average GCI per lead is $1,180 for AI-sourced versus $240 for Zillow. Time to close is roughly half: 42 days versus 87.

The reason for this huge increase in close rates is that a buyer who has spent 30-plus minutes asking an AI about a market arrives pre-educated. They speak to agents in an almost ‘I’ve been referred to you’ in some way.

Second, AI tools rarely surface more than three to five names per search so competition at the point of discovery is dramatically lower. This is hugely different from hundreds, or thousands, of agents competing for Zillow leads. 

The cost of these leads is significantly less but the value, in GCI, is much higher. Which begs the question: why are agents ignoring this channel?

91% of agents don’t appear in AI search

Despite the aggressive and unprecedented change, only 8.4% of practicing U.S. agents appear in any AI-generated response to high-intent searches in their own market. The top 1% of real estate agents capture 47% of all AI citation share.

The concentration is partly explained by a training-data problem. Zillow, Realtor.com, Redfin, Trulia, and Homes.com collectively account for an estimated 61% of real estate-related URLs in publicly available LLM training datasets. 

The default frame most AI models use to answer real estate questions is portal-shaped and agents are presented as line items inside portals rather than as independent professionals.

Breaking through that default requires building an identity outside the portal context: third-party citations, consistent business information, original local content on an agent’s website and, review distribution across multiple platforms. 

The report finds that agents with citations spread across four or more review platforms are significantly more likely to surface in AI responses than agents with all their reviews concentrated on a single site even when the latter has a higher total count.

AI citations are what drive homebuyers and sellers to reach out to an agent directly. And, they allow people to essentially prequalify themselves before choosing who to work with in their local market.

In 71% of U.S. metros, no single agent currently holds more than 15% citation share. The dominant position is unclaimed in nearly three out of four markets. But the compounding curve means that window narrows with each passing quarter.

Ryan Darani is the co-founder and AI brain of FlyDragon.

This column does not necessarily reflect the opinion of HousingWire’s editorial department and its owners.

To contact the editor responsible for this piece: [email protected]