What problem does it solve?
Finding a home online still mostly means setting filters (price, beds, baths, radius) and scrolling photos across open tabs, a calculator and a map, even though what a shopper actually wants is rarely expressible as a filter: a backyard for a dog, a short commute to a specific office, a kitchen that has not been renovated in 20 years, or "can I afford this if I move in June". Zillow describes home search as historically meaning "setting filters, scrolling through listings, opening multiple tabs and windows, pulling up a calculator to piece together whether a place actually works."
Three large US real estate portals, Zillow, Realtor.com and Redfin, each launched a conversational, AI powered search experience within a few months of each other in late 2025 and 2026, all built on the same idea: let the shopper type or say what they want in their own words, keep the context across a session instead of resetting on every query, and turn the answer into a next step (a comparison, a tour, a connection to an agent) rather than another list of links.
How does it work?
- Interpret the question. A shopper asks in plain language, for example "Find similar homes within my budget that are closer to light rail" or "Can I afford this apartment if I move in June?" (Zillow's own examples), and the system determines what is really being asked: a search, a comparison, an affordability question or a request to act.
- Route to the right capability. Zillow describes a multi agent architecture where a central agent invokes domain specific "skills" (property search, financing, valuation) rather than relying on one model to answer everything directly; other portals combine natural language parsing with structured search filters underneath.
- Ground the answer in live listing data. Results and answers come from the portal's own, continuously updated listing inventory, price history and, on Zillow, computer vision over listing photos to reason about layout, light and condition, not from the model's general knowledge.
- Keep context across the session. Follow up questions ("I think I can stretch to $650K now, what does that open up") update the same search instead of starting over, and some portals persist preferences across return visits.
- Turn conversation into action. The assistant offers to book a tour, apply for a rental or connect the shopper with a local agent or property manager, so the conversation becomes the entry point into the transaction rather than a dead end.
- Screen for fair housing. Zillow describes a "Fair Housing Classifier", a guardrail that evaluates both the shopper's questions and the assistant's answers to help prevent discriminatory steering.
- Audience
- Customer facing
- Autonomy
- Supervised agent
- Adoption
- Early adopters
- Channels
- Web chat, Mobile app, API and system to system
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
No public deployment has disclosed a measurable outcome yet.
Value drivers: Customer experience, Revenue growth, Speed and cycle time, Inclusion and access.
Indicative value
A real estate portal with 10 million monthly visitors
USD 600,000 to USD 16.2 million
Extra lead value from conversational search per year
How this is calculated
Formula: visitors * assistantShare / 1000 * extraConversionPoints * valuePerAction * 12. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Monthly visitors who search for a home visitors, visitors per month | 10,000,000 | 10,000,000 | The reference portal. |
| Share of visitors who use the conversational assistant instead of filters alone assistantShare, fraction of visitors | 0.05 | 0.15 | Editorial assumption for an early stage rollout still in beta on the largest portals as of 2026, replace with your own adoption data. |
| Extra tour requests or agent connections per 1,000 assisted sessions, versus filter only search extraConversionPoints, extra actions per 1,000 sessions | 5 | 15 | Editorial assumption; none of the portals in this page's evidence disclose a conversion lift, so treat this as a hypothesis to test with an A/B rollout, not a benchmark. Assumes one assisted session per assisted visitor per month. |
| Value of one extra tour request or agent connection to the portal valuePerAction, USD per action | 20 | 60 | Editorial assumption, replace with your own referral or lead value. |
What it leaves out: Entirely a modelled hypothesis: no portal in this page's evidence has disclosed a conversion or revenue result for its AI search assistant, which launched in beta in late 2025 and 2026. It leaves out the cost of the AI, the value of consumer trust and the risk that a bad or biased answer drives a shopper away.
Who already uses it?
3 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Realtor.com
United States · Real estate · 2026
Realtor.com, operated by Move, Inc., a News Corp subsidiary, launched an AI powered natural language search feature in October 2025 that interprets listing details and lets shoppers combine their own words with more than 300 searchable terms and tags. On June 2, 2026, it launched RealAssist AI, built with Google's Gemini and Google Cloud, an AI first search experience that keeps context across sessions, maps commute routes and connects shoppers to a local agent. As of the launch, RealAssist AI is in a beta available to a select group of logged in users, with full availability rolling out afterward.
No outcome disclosed.
Redfin
United States · Real estate · 2026
Redfin, part of Rocket Companies, announced conversational search on Redfin.com in November 2025, letting home searchers ask questions and refine results in plain language rather than filters, and extended it to its iOS app. In the week of February 6, 2026, Redfin launched an app inside ChatGPT so users can search Redfin's listings, data and market insights directly within ChatGPT, carrying context across follow up questions about neighborhoods, commute times or price history. Redfin says it remains the only brokerage that lets people hold an ongoing, back and forth conversation directly in the search bars on its own website and mobile app.
No outcome disclosed.
Zillow Group
United States · Real estate · 2026
Zillow launched Zillow AI mode on March 25, 2026, a conversational experience that lets buyers and renters ask questions in natural language, such as affordability or neighborhood questions, and take actions like booking a tour or connecting with an agent. It uses a multi agent architecture where a central agent invokes domain specific skills (search, financing, valuation) and includes a "Fair Housing Classifier" guardrail that checks both questions and answers. The rollout is a beta available to a limited group of users, with a wider rollout planned during 2026.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Live, structured listing inventory with price history and status
- Neighborhood, school and market data linked to each listing
- The portal's own affordability or valuation model, callable as a tool
- A labelled set of fair housing steering scenarios to test the guardrail against
Systems to integrate
- Listing and MLS data feed
- Affordability, mortgage and valuation calculators
- Tour scheduling and agent or property manager routing (CRM)
- Map, commute and neighborhood data providers
Complexity: High
Search itself is not the hard part; a portal already has structured listing data and filters. The work is in tool or function calling across several specialized skills (search, affordability, valuation, scheduling), maintaining conversational context across a session and return visits, a fair housing guardrail tested on real steering scenarios, and, for the richest experiences, computer vision that reads listing photos rather than only structured fields.
- 1
Start with search and comparison
Natural language search and side by side comparison of listings rely only on data the portal already has, so they are a natural starting point. Zillow says early testing questions focus on affordability, neighborhood insights and comparing homes with specific features, so plan to bring affordability reasoning in early rather than treating it as a later phase.
- 2
Wrap calculators and data as tools, not model guesses
Affordability, mortgage payment and valuation numbers must come from the portal's own calculators and models, called as tools, the same way a home loan assistant should never let the model compute a repayment itself.
- 3
Build and red team the fair housing guardrail before launch
Write test scenarios that probe for steering, discouragement or disparate treatment by protected characteristic, run them on every model and prompt change, and keep a human review path for edge cases.
- 4
Make the conversation lead somewhere
Every qualified conversation should end with a concrete next step: a tour booking, an application, or a warm handover to a local agent or property manager with the conversation attached.
- 5
Measure adoption honestly before claiming a lift
Track how many visitors use the assistant, how many searches it actually resolves without a filter fallback, and run a controlled comparison against filter only search before crediting it with any conversion gain.
Guardrails
- Search results, price history and valuations only from the portal's own live data, never generated
- A fair housing classifier or equivalent guardrail on both the shopper's question and the answer
- Every affordability or valuation number computed by a calculator tool, labelled as an estimate
- Clear handover to a licensed agent or property manager for advice or a transaction
KPIs to instrument
- Share of visitors who use the assistant instead of, or alongside, filters
- Tours booked, applications started and agent connections per assisted session
- Guardrail trigger rate and false positive or false negative rate on the fair housing test set
- Session return rate and whether context correctly carries across visits
Human in the loop
Licensed agents and property managers own advice, offers, applications and every step of the actual transaction; the assistant informs and schedules. Zillow says AI mode is designed to function within established industry regulations and professional standards, "supporting, not replacing, licensed expertise in the transaction." A product and fair housing team reviews a sample of conversations and every guardrail change.
Common failure modes
- Steering by omission
- The assistant answers accurately but nudges some shoppers toward or away from certain neighborhoods through subtly different framing. Test with paired scenarios that vary only a protected characteristic.
- Stale or wrong listing data
- A sold or price changed listing is shown as available at the old price. Read from the live inventory feed, never a cached copy, and show a last updated time.
- Photo reasoning that overclaims
- Computer vision over listing photos infers a renovation or condition that is not accurate, misleading a buyer. Label vision derived claims as estimates and let a human correct them.
- A conversation that goes nowhere
- Shoppers chat but never reach a tour, application or agent. Instrument the funnel from question to action, not only message volume.
What are the risks and rules?
EU AI Act
Depends on design
For a typical portal search assistant, the default tier is limited: it falls under the Article 50 transparency duty. It becomes high risk under Annex III point 5(a) only in the narrower case where it is used by a public body or on its behalf to decide eligibility for social housing or other public assistance, which a listing search assistant rarely does. Separately, steering or discriminatory treatment in the answers themselves is unlawful under fair housing and equal treatment law regardless of the AI Act tier.
Guidance
- 42 U.S. Code 3604, discrimination in the sale or rental of housing (Fair Housing Act) (United States Congress, North America). Makes it unlawful to indicate a preference, limitation or discrimination based on a protected characteristic in any statement made in connection with the sale or rental of a dwelling, and to represent, because of a protected characteristic, that a dwelling is not available when it is. A search or chat assistant that answers on behalf of a portal or a housing provider speaks on all these points. There is no dedicated US Fair Housing Act id in this library's regulation taxonomy yet; this guidance entry stands in for it.
- Article 50, transparency obligations for providers and deployers of certain AI systems (European Union, Europe). People must be informed that they are interacting with an AI system unless this is obvious from the context.
Controls to put in place
- AI disclosure in the assistant's first response
- A tested fair housing guardrail on questions and answers, reviewed after every model or prompt change
- All financial and valuation numbers sourced from calculator tools, never generated
- Logging of conversations, guardrail triggers and handovers for review
- Clear escalation path to a licensed agent, broker or property manager
Frequently asked questions
- Which real estate portals have an AI search assistant?
- Zillow launched Zillow AI mode in beta in March 2026. Realtor.com launched a natural language search feature in October 2025 and RealAssist AI, built with Google's Gemini, in beta in June 2026. Redfin announced conversational search on Redfin.com in November 2025, extended it to its iOS app by February 2026, and launched an app inside ChatGPT in the week of February 6, 2026.
- Do these assistants report a conversion or revenue lift?
- Not yet in public sources. All three portals describe the capability and, for Zillow and Realtor.com, note they are in a beta rollout to a limited group of users, without disclosing usage or conversion figures. Treat any value estimate as a hypothesis to test, not a benchmark.
- Can the assistant tell me if I can afford a home?
- It can give an estimate from the portal's own affordability tool, such as Zillow's BuyAbility, clearly labelled as an estimate. A real affordability assessment needs a full mortgage application and underwriting by a lender.
- How is this different from a general purpose AI assistant like ChatGPT?
- A portal's assistant answers from the portal's own live, verified listing inventory and calculators. Zillow and Realtor.com say theirs are built with a fair housing guardrail specific to real estate (Zillow's Fair Housing Classifier, Realtor.com's Fair Housing compliance guardrails); Redfin's release does not describe one. Redfin's own app inside ChatGPT is one way a portal meets shoppers who are already using a general assistant, while serving Redfin's own listings and data.
How to cite this page
Blits.ai AI Use Case Library, "AI conversational property search assistant for real estate portals", last verified 29 September 2026, https://www.blits.ai/ai-use-cases/property-search-assistant-for-real-estate-portals. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 29 September 2026: First published