Real estate triggers an AI Overview on 0.14% of relevant searches. That is the lowest rate of any major U.S. industry, according to joint research from 5W AI Communications and Haute Living published in April. The same study found 82% of agents now use AI tools every day. So agents pour effort into AI while the buyer-facing answer layer stays almost entirely closed to them.
On June 2, Realtor.com widened that gap. It opened a limited beta of RealAssist, a conversational home-search experience built on Google Cloud and Gemini and grounded in 30 years of its own buyer data. RISMedia confirmed the rollout on June 8. A buyer can now ask plain questions and get affordability math, commute estimates, school ratings, and side-by-side neighborhood comparisons without ever loading a listing page or contacting an agent.
The portals are not trying to rank in AI answers. They are becoming the answer. If you are a solo agent or a small team, your counter-move is narrow and specific, and you can start it this week.
RealAssist moves the first conversation away from you
For years the agent owned the early buyer conversation. A lead filled out a form, you replied, and the relationship started there. RealAssist takes that exchange and resolves it inside Realtor.com. The buyer asks “can I afford this in Tempe on 140k with two kids,” gets a grounded answer, and keeps going. By the time a human enters, the buyer has already formed opinions the portal shaped.
This is different from the ChatGPT visibility problem agents have chased since last year. That fight was about getting cited in a general model. RealAssist is a closed system grounded in proprietary listing and behavior data you cannot feed or edit. You will not optimize your way into it. So stop trying. The winnable ground is the answer the portal cannot give: the ground truth of your specific listings and your specific streets.
Build the one answer asset the portals cannot fake
Realtor.com knows the median price on a block. It does not know that the HOA just approved a special assessment, that the seller will include the casita furniture, or that the elementary school redrew its boundary in March. That field-level detail lives in your head and your CRM. Turn it into a machine-readable answer feed, and you own a layer the portal structurally cannot reach.
Start with your active listings. For each one, write eight to twelve buyer questions and short factual answers. Not marketing copy. Real questions: What are the actual monthly costs including the HOA. What did the last comparable sale close at and why. What is the commute to the nearest hospital at 7am. Then mark up each listing page with FAQPage schema and RealEstateListing schema so machines can read the pairs cleanly. Here is the minimum shape:
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [{
"@type": "Question",
"name": "What are the monthly costs on 412 Mesa Verde?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Principal and interest at current rates run about 2,180. HOA is 165 a month and covers front-yard landscaping and the community pool."
}
}]
}Validate it in Google’s Rich Results Test before you ship. The point is not a rich snippet for its own sake. It is giving every model that crawls you a clean, attributable fact it can lift. When a buyer asks an AI assistant about that street, your page is the only source that knows the assessment detail.
Do not stop at listings. Build one standing page per farm area with the same treatment: ten dated, factual question-answer pairs about taxes, build-out timelines, school boundaries, and water rights if they matter where you sell. Date every answer. A page that says “as of June 2026, the Marley Park HOA dues are 192 a month” reads as current and specific to a model. A page that says “low HOA dues” reads as marketing and gets ignored. Refresh the dates quarterly and the page keeps earning citations while your competitors’ static neighborhood pages decay.
Wire your own site to answer in plain language
The portal gave the buyer a conversation. Give them one too, grounded only in your data so it never hallucinates a price. You do not need an engineering team. A retrieval setup over your own documents costs a weekend and under 60 dollars a month.
The stack: export your listing facts, neighborhood notes, and closed-comp history into plain text files. Load them into a retrieval tool like Chatbase or CustomGPT, both of which run 40 to 150 dollars a month and answer only from documents you upload. Drop the widget on your site. Now a buyer on your page at 11pm asks “how is the commute from this house to downtown Phoenix” and gets your answer, not a generic one. Set the system prompt to refuse anything outside your data and to hand off to your contact form when it does not know.
One small team in Mesa did a manual version of this and cut their after-hours form abandonment by routing every unknown question to a same-night text. The lesson holds: the tool’s job is not to close. It is to keep the buyer in your answer instead of bouncing back to Realtor.com. Track one number to know if it works: the share of site visitors who ask the bot at least one question. If that climbs past 15%, your page is doing the job the portal wants to take from you.
Watch what the vendors ship, then decide what you build
Your CRM vendor is racing the same direction, and some of this you should rent rather than build. In late May, Realty ONE began rolling out ZONE Pro, an agent platform with AI workflows baked in. Lofty, formerly Chime, shipped a Homeowner Agent product in April that mines your existing contacts for likely-seller signals. Cloze unveiled Cloze Forge, which lets brokerages spin up branded tools wired to their own data without a developer.
The tradeoff is real and worth naming. Vendor tools are faster and supported, but they answer from the vendor’s structure, not yours, and your proprietary local knowledge gets flattened into their template. The schema feed and the document chatbot above are yours to keep and move between brokerages. A reasonable split: rent the seller-signal mining from Lofty or whatever your CRM offers, and own the buyer-answer layer yourself because that is your moat. Do not pay a vendor to template the one thing that makes you distinct.
The signal to watch over the next 30 days is whether RealAssist exits limited beta and whether Zillow answers it. If both portals are answering buyers conversationally by July, the early-conversation window closes fast, and agents without their own grounded answer layer will meet buyers later and colder than they do today. The agents who shipped a listing FAQ feed and a document-grounded chatbot this month will be the only local source an AI can cite when the question gets specific.
We build these answer assets for operators who would rather run their business than wire schema by hand. If you want the listing FAQ feed and the grounded site chatbot done for you, that is what we do at Atlas Unchained.
