Real Estate: AI is reaching the buyer before the agent does
The most consequential AI change in residential real estate right now is not inside the brokerage office. It is at the front of the search, where people first look for a home. Since March 2026, Zillow, Realtor.com and Redfin have moved home search into conversation: a buyer describes what they want to an AI assistant, sometimes inside ChatGPT, and gets answers on price, neighbourhoods and listings before any agent is involved. The software that shapes a buyer's shortlist is moving upstream of the agent.
Where AI actually stands today
Start with the honest baseline: the headlines run ahead of the reality. AI is now ordinary equipment for an agent, not an experiment. In an early-2026 survey by Delta Media reported by Inman, about 97% of brokerage leaders said their agents use AI, mostly for the unglamorous work: writing listing descriptions, drafting emails and social posts, and tidying up the contact database (the CRM, where an agent keeps client records). McKinsey estimates generative AI could add $110 billion to $180 billion in value across real estate, mostly from faster paperwork and pricing, but notes the industry is among the slowest to adopt new technology, and that the value appears only when a firm changes how it works, not when it buys a subscription.
The search moved upstream
The change that matters most is in discovery, meaning how a buyer finds a home. On 25 March 2026, Zillow launched AI mode, which lets a shopper search by describing what they want in plain language instead of setting filters, then schedule a tour or ask to be connected with an agent (Zillow; Inman). Five days later, on 30 March, Realtor.com launched an app inside ChatGPT: a buyer asks about affordability and neighbourhoods inside the chatbot, and Realtor.com then routes the high-intent ones back to its own site to see listings and reach an agent (Realtor.com; Inman). Zillow had placed the first such app in ChatGPT in late 2025, and Redfin followed earlier in 2026. In May 2026, Google began showing home listings in US mobile search again in selected markets, drawing on agent-supplied data through partners including eXp (HousingWire).
The common thread: the first conversation a buyer has about a home is increasingly with a machine, on a platform the agent does not own.
The agent's own tools filled up too
Underneath the portals, the tools an agent uses daily are getting the same treatment. On 1 April 2026, Lofty (the Phoenix company formerly called Chime) launched Homeowner Agent, which scans an agent's own CRM and flags who is likely to sell, who owns a property they do not live in, and who may be heading toward foreclosure, then sends those owners reports on their home's estimated value and equity (Lofty; Inman). The pitch: the listings you could win are already in your database, and the AI surfaces them. On 12 May, Cotality launched its Broker Listing Exchange with Keller Williams and HomeServices of America, where a broker enters a listing once into a brokerage-controlled system and its AI checks the data against industry standards before sending it out to the MLS (the shared database agents use to list and find homes) and the portals (HousingWire). Prospecting and listing admin, the routine parts of the job, are the first to be automated.
What it means for you
Two things change for a working agent or broker.
First, your leads. For two decades the portal was where buyers searched and the agent was who they called. Now an AI assistant sits in between, deciding which homes a buyer sees and when a human is introduced. If that introduction comes later, after the buyer has already narrowed the field, the agent has less say over the shortlist and arrives further down the process. The portals say plainly that they route "high-intent" buyers to agents, which means the portal, not you, now defines high intent.
Second, what you are paid for. The tasks AI takes first, listing copy, prospecting, data entry, and the comparative market analysis (the priced-out estimate of what a home should sell for), used to fill an agent's day and justify junior staff. What is left is the part the software cannot do: knowing a street, reading a seller's real motivation, holding a deal together when it wobbles, and being the name a client trusts with the biggest transaction of their life. This is the argument in defensibility in the AI era: when a tool becomes cheap and common, value moves to judgment, relationships and local knowledge that do not copy. The same shift just reached law, where AI got cheap inside the software lawyers already use, which we covered in this week's Industry Pulse on legal.
What to do now
Treat the AI portals as a referral source you do not control, and lower your dependence on them. The agents least exposed are those whose business comes from past clients and their own network, not bought leads. Lofty's tool makes the point even if you never use it: the cheapest listings to win are the ones already in your database, so use it.
Pick one or two routine tasks, listing descriptions or a first-draft market analysis, and let AI do them faster while you check the output. Measure how often it is wrong before it reaches a client. The error rate, not the demo, tells you what the tool is worth.
Be careful where AI sets prices. The clearest legal warning in real estate today concerns rent. RealPage, whose software recommended rents to landlords using competitors' private data, settled an antitrust case with the US Justice Department in late 2025 and must stop using it; New York has banned algorithmic rent-setting outright, and RealPage is suing to overturn the law (ProPublica; Reed Smith). If you manage rentals, letting an algorithm set prices off shared, nonpublic numbers is now a legal exposure, not only an efficiency. The wider lesson holds across the job: an AI recommendation does not move responsibility off whoever acts on it.
The honest read for mid-2026: AI has not changed what a good agent does, but it has changed where the customer starts and who they meet first. Brokerages treating it as a cheaper way to handle admin are getting the small win; the ones working out how to stay the buyer's first call, now that the first call is often a chatbot, are working on the real one. As we put it in The model wasn't the moat, the advantage was never the software. It is the relationship the software routes around.