Conversational AI for real estate: how it works and how to set it up

Conversational AI for real estate succeeds or fails on conversation design: what the AI asks, in what order, and the exact point it hands over to a person. This article answers one question. How do you write the conversations themselves for buyers, sellers, landlords and tenants so that each one ends in a booked viewing or valuation rather than a polite chat that goes nowhere?

Design every conversation around the next commitment

The common failure is an AI that answers questions well and books nothing. A buyer asks about parking, gets a correct answer, says thanks and leaves. The conversation was pleasant and the opportunity was lost.

Every real estate conversation has a natural next commitment. For a buyer it is a viewing. For a seller it is a valuation appointment. For a landlord it is a rental appraisal. For a tenant it is a viewing followed by an application. Write each conversation so that every answer moves one step towards that commitment, and so the AI offers it no later than its second or third message.

This is a point of view, and it has a cost. Some enquirers want information only and find an early offer pushy. You accept that cost because the alternative, an AI that answers everything and offers nothing, produces no measurable return.

The four conversation types

Write separate instructions for each. They overlap less than you would expect.

Buyer enquiries

The buyer has usually seen a specific property. Answer the question they asked first, in one or two sentences, from the listing data. Then ask one qualification question at a time: whether they have a property to sell, whether they have spoken to a lender or mortgage adviser, and when they want to move. Then offer two or three specific viewing slots, not a link to a calendar and a request to choose.

If the property is under offer, say so plainly and offer similar listings, then a viewing. Never pretend a sold property is available to start a conversation.

Seller and valuation enquiries

Sellers are cautious and often comparing offices. Do not ask for their whole situation up front. Confirm the address, ask what prompted the enquiry and offer a valuation appointment. Leave fees, price expectations and marketing plans to the person who attends. An AI that quotes a fee or suggests a value in chat removes the reason to meet.

Landlord enquiries

Landlords want to know whether you can manage their property and what it might let for. The AI should confirm the property type and location, ask whether it is currently tenanted, and offer a rental appraisal visit. Questions about fees or compliance go to a person, because the answer depends on the property.

Tenant enquiries

Tenants move fast and enquire with several offices. Answer availability, the date it is available from and the viewing options. Ask only what you need to book: preferred viewing time, move date and how many people will live there. Leave affordability and referencing to the application stage, handled by staff.

Hand-over rules

The second thing most designs get wrong is the hand-over. An AI that hands over too often adds no value; one that never hands over frustrates clients. Write explicit rules.

Hand over immediately when the person asks for a human, mentions a complaint, raises a legal or financial question about their own situation, wants to make or discuss an offer, or mentions something sensitive such as a bereavement or divorce. Hand over after two failed attempts to answer the same question. Hand over when a known client of a specific agent gets in touch, routed to that agent.

When the AI hands over, it should tell the person who will reply and roughly when, and the receiving agent must see the full conversation before replying. A hand-over that makes the client repeat themselves undoes the good work of the first few messages. For how routing to the right agent works at team level, see assigning and reviewing leads in a real estate team.

Tone and disclosure

Write the AI's tone like your best receptionist: short sentences, local knowledge, no exclamation marks, no forced enthusiasm. Avoid language that sounds like a sales script, such as "amazing opportunity" or "act fast". Real estate clients are spending a lot of money and respond to calm competence.

Disclose that it is an AI in the first message. Something like: "Hi, I am the office's AI assistant. I can answer questions about this property and book a viewing, and a member of the team is available if you would prefer." Clear disclosure prevents the awkward moment when a client realises later.

Setting it up

1. Write one page per conversation type covering purpose, questions in order, the commitment to offer and hand-over triggers.
2. Go to AI Agents, then Conversation AI, click + Create Bot and choose Prompt Based Bot for buyer and tenant enquiries.
3. In Bot Settings, paste the matching instructions into the bot's prompt.
4. In Bot Training, add your current listings and approved answers.
5. In Bot Goals, add the goal prompts, turn on Enable Appointment Booking, choose Multiple Calendars and select the viewing calendars.
6. Under Setup Your Actions, open Human Handover, enable the scenarios and choose the user who receives the conversation.
7. Create a second bot the same way for seller and landlord enquiries, with its own instructions and the valuation calendars.
8. In each bot's Deploy tab, enable Chat Widget, assign your widget, click Update, then Save. Add SMS, WhatsApp and social channels once the answers are reliable.
9. Run each bot in Suggestive mode first, read every conversation for two weeks, and move to Auto-Pilot once the answers hold up.

Channel differences

The same conversation reads differently on each channel. On web chat, people expect an instant reply and will accept three or four short messages. On SMS, keep each message short, avoid links in the first reply and never send more than two messages in a row without a response. On WhatsApp you can share photos and floor plans where your listing data includes them. On social messages, people are more casual and often less ready to commit, so offer a viewing but accept a request to keep in touch. More on the text channel in real estate SMS marketing as a conversation channel.

What to measure

Count booked commitments per conversation type, not conversations handled. Track hand-over rate by reason, because a rising rate for one reason usually points to a gap in the knowledge base. And sample conversations weekly for tone, since a correct answer delivered badly still loses the client. For the brokerage-wide view of where these agents sit, read AI across a real estate team, and for the single realtor's perspective see AI for realtors.

Frequently asked questions

Can conversational AI answer questions about price?

It can state a listing price and any published details. It should not advise on what to offer, what a home will sell for or whether a price is fair, because that is advice that belongs to a licensed or qualified person.

What if the AI gets a question it cannot answer?

It should say it will check with the team, hand over and tell the person when to expect a reply. Then add the answer to the knowledge base so it can answer next time.

How many questions should the AI ask before offering a viewing?

No more than two or three, one at a time. Long qualification before a booking makes enquirers feel interrogated, and many leave to book elsewhere.

Does conversational AI work out of hours?

Yes, and that is where much of its value sits. Evening and weekend enquiries get a reply and a booked slot straight away instead of waiting for Monday morning.

The journey this guide describes