// AI AUTOMATION
AI Chatbot for Real Estate: What It Should Do, and What It Shouldn't
Property enquiries go cold in hours, and most of them arrive when the office is shut. Here is what a real estate chatbot should handle, what it must leave to an agent, and how to tell a good one from a liability.

Researchers writing in Harvard Business Review audited 2,241 US companies by sending each a test web lead. Only 37% replied within an hour, and the average reply took 42 hours. A companion study of 1.25 million leads found that firms which tried to reach a prospect within an hour were nearly seven times as likely to qualify the lead as firms that waited even one hour more. Now think about when people browse property: in the evening, at the weekend, on the sofa. That gap is the real case for an AI chatbot for real estate.
The case against is just as real. A bot that quotes the wrong price, promises a parking space the flat doesn't have, or tells a buyer the seller will "definitely accept" a lower offer creates problems no agent wants to clean up. This guide covers the jobs a property chatbot does well, the ones it must never touch, the data it needs, the Bulgarian details most guides skip, and what it costs.
What a real estate chatbot does that a contact form can't
A contact form asks the visitor to commit before they get anything back. Name, phone, message, then wait. Plenty of people aren't ready for that. They have a rough idea, a budget and a few questions, and they want answers before an agent starts calling them.
A chatbot reverses the order. The visitor asks first and gets something useful straight away: three matching listings, the service charge on a new build, whether pets are allowed. Contact details come later, when the visitor has a reason to share them, such as booking a viewing. It sounds like a small change. In practice the enquiries you receive are warmer, and the agent who calls back already knows what the person wants.
It also answers at 11pm on a Sunday, in whichever language the visitor writes in, without making them learn your search filters. For an agency site or a property portal, that last point is often the biggest win.
Where the chatbot should live
Start with your own website, on the listing pages and the search page. That is where visitors already have a property in front of them and a question about it. Once the bot answers well there, the same engine can sit behind Facebook Messenger, Instagram or Viber, which is where many Bulgarian buyers would rather write anyway.
One limit to know early: enquiries that come through the big listing portals usually reach you by email or phone, through the portal's own form. A chatbot on your site never sees them. If most of your leads arrive that way, the quicker win may be an automation that reads portal emails, matches them to the listing and drafts a reply for the agent. That is a different project, and often a cheaper one. It falls under AI automation rather than chatbots.
Five jobs worth giving a property chatbot
Searching listings in plain language
Most visitors don't think in filters. They think in sentences: "two bedrooms near a metro station in Sofia, under €250,000, not ground floor". A good bot turns that into a real query on your live listings and returns the matches in the chat. A weak one asks the same questions your filter panel already asks, one at a time, which is slower than the filter panel.
Answering questions about a specific listing
Floor, year of construction, Act 16 status, heating, parking, service charge, distance to the nearest school. These questions fill agents' inboxes, and the answers are usually in the listing already, buried in a description nobody reads to the end. The bot should answer from that listing and say clearly when something isn't recorded.
Qualifying the enquiry
Budget, timeline, buying or renting, cash or mortgage, already selling a property or not. An agent needs these answers before the first call. A bot can collect them naturally during the conversation and pass them to your CRM with the transcript attached. The trick is restraint. Ask only what the agent will actually use, and ask it when it fits, not as a five-question gate before the visitor sees anything.
In practice it looks like this. A visitor asks about a two-bedroom flat in Mladost. The bot answers, shows two similar listings, and only then asks whether they are buying with a mortgage, because the agent will want to know before the viewing. One question, at the moment it makes sense, with a reason the visitor can see.
Booking viewings
If your agents keep calendars, the bot can offer real slots and confirm a viewing. If they don't, it can collect two or three preferred times and hand them to the right agent. Either way, the visitor leaves the chat with a next step instead of a promise that someone will be in touch.
Explaining the process
First-time buyers and foreign buyers ask the same things again and again. What does a preliminary contract involve? Who pays the notary fees? How long does the deal take? A bot can answer these from a reviewed set of articles your agency has written, and point to an agent when the question turns specific.
What it must leave to an agent
Write these limits down before you choose a tool, and test every one of them before launch.
- Negotiation. The bot never hints at what a seller might accept, never discounts, never says a price is "flexible". It only states the asking price shown in the listing.
- Valuations. "What is my flat worth?" is a lead for an agent, not a question for a language model. The bot can collect the address and basic details and book a valuation call.
- Legal and tax advice. It can explain what a preliminary contract is in general terms. It should not tell a specific buyer whether a specific property has a clean title or what tax they will owe.
- Anything not in the data. If the listing doesn't mention a parking space, the answer is "that isn't listed, an agent can confirm", not a confident guess. A bot that invents features is worse than no bot.
- Steering people. A bot that filters by price, area and size is helpful. One that nudges people away from areas based on who they are is a legal and ethical problem. Keep the criteria to the property, never the person.
This is why the architecture matters. A bot built on retrieval-augmented generation answers only from documents and data it has just looked up, and can be told to refuse when it finds nothing. A general chatbot with your website pasted into its instructions will fill gaps with plausible text. We compared the two in RAG chatbot vs regular chatbot.
The data behind a good property chatbot
Most vendors promise a bot "trained on your website". For a property business that is not enough, because a property site changes every day.
Live listings, not a snapshot
Listings go under offer, get reserved, change price and disappear. A bot that reads a copy of your site from last month will keep recommending a flat that sold three weeks ago, which is the fastest way to lose a buyer's trust. It needs to query listings live, through your CMS, your listings API or a feed that updates at least daily, and it needs to respect status fields such as reserved, sold and rented.
The bot can only be as accurate as the listings it reads. If one agent writes "parking: yes" in a field and another mentions it halfway through the description, the bot will answer the first listing correctly and miss the second. Before launch, agree which facts live in structured fields (floor, area, heating, parking, completion stage, service charge) and fill them consistently. It is dull work, and it improves your site search and your portal exports at the same time.
The knowledge your agents carry
Which building has a noisy restaurant on the ground floor. Which developer is running late on Act 16. Which part of the district floods in heavy rain. Some of this should never go into a bot. Some of it, such as typical service charges in a complex or how parking works in a gated development, makes excellent answers. Write the useful parts down as short, reviewed notes and let the bot read them.
Your CRM
The bot should create a contact in your CRM with the transcript, the listings the visitor looked at and the answers to your qualifying questions. If the agent who calls back has to ask everything again, the bot has only added a step.
The Bulgarian details most guides skip
Most guides to real estate chatbots are written for American agents, so it misses the things that break bots on Bulgarian property sites.
- Two languages, often in one conversation. Coastal, ski-resort and Sofia listings draw enquiries from foreign buyers and tenants who write in English or another language. Bulgarian buyers sometimes type in Latin letters on a phone set to English. The bot has to answer in the visitor's language and still match "Lozenets", "Лозенец" and "lozenec" to the same district.
- District and complex names. Listings mix official district names, informal neighbourhood names and complex names. Search has to treat them as the same place or it will tell a buyer you have nothing in an area you cover.
- Prices after the euro switch. Bulgaria adopted the euro on 1 January 2026. Older listings, PDFs and agent notes may still carry prices in leva. Decide which figure the bot quotes and make sure it never mixes the two in one answer.
- Local terms. Act 15 and Act 16, preliminary contract, notary deed, "с ДДС" or "без ДДС" on new builds. A bot that hasn't seen these handles them badly in both languages, and foreign buyers need them explained plainly.
We dealt with most of these on the Houzez AI Assistant, the bilingual bot we built for the Houzez property platform. It turns a plain-language brief into a query on the live listings API and returns matching properties in the chat. It picks Bulgarian or English from the visitor's first message, with no language switch to click. It was deliberately kept narrow: understand property searches, answer questions about the process, and nothing else. No small talk and no pushy lead capture.
Off-the-shelf tool or a custom build?
There are plenty of chatbot platforms with a real estate template. They connect to a website, read its pages, show a chat window and charge a monthly fee. For a small agency with a few dozen listings and mostly simple questions, one of these is a sensible place to start.
They tend to struggle in three places: live listing search, Bulgarian-language retrieval and CRM connections beyond the big American systems. Before you sign up, test the tool on your own site with the questions below.
- Ask for a listing you know sold last week. Does it still recommend it?
- Ask in Bulgarian, then in Latin-letter Bulgarian, for a district you cover. Does it find the same listings both times?
- Ask about something the listing doesn't mention, such as a storage room. Does it admit it doesn't know?
- Ask what the seller would accept. Does it refuse politely and offer an agent?
If the tool fails two of these, it will fail hundreds of times once real buyers use it.
A custom RAG chatbot costs more and takes longer, but it can search your live listings properly, answer only from your data, write good Bulgarian and connect to whatever CRM or listings system you run. At Ascend we start with a two-week proof of concept on your real listings, so you see measured answer quality before you commit. A production build usually follows in four to eight weeks. Running costs are the language model, billed per conversation, plus hosting, which for most agencies comes to a few hundred euros a month. Our pricing page shows how we scope it.
The rule of thumb is simple. Start with the cheapest option that passes your own test. Move to a custom build when you can name the wall you hit.
Before buyers see it
- Test it with real enquiries. Take 30 messages buyers and tenants actually sent you, typos included, and grade every answer as right, wrong or "didn't know". "Didn't know" is fine. Wrong is not.
- Say it is a bot. From 2 August 2026, Article 50 of the EU AI Act requires AI systems that talk directly with people to be designed so those people know they are dealing with AI, unless that is obvious. A short line at the top of the chat window covers it. A bot named after your best agent does not.
- Handle personal data properly. Names, phone numbers, budgets and mortgage status are personal data under the GDPR. Tell visitors what you collect and why, link your privacy policy in the chat window, and agree with your supplier where conversations are stored and for how long.
- Make handover obvious. Every conversation needs a visible route to a person, and the agent should see the whole conversation, not just a name and number.
- Read the transcripts weekly for the first two months. The questions the bot fails on show you what is missing from your listings and your site.
A soft launch helps. Put the bot on a handful of listing pages for two weeks, let one agent read every conversation, fix what goes wrong, and only then roll it out across the site. Two weeks of careful reading catches most of the embarrassing answers before a serious buyer gets one.
Then measure what matters. "Conversations handled" is a flattering number, because a visitor who gave up also counts as handled. Track viewings booked, qualified enquiries passed to agents, how many of those turned into deals, and how many wrong answers you found in the weekly review. If you also need a site that is built for search and enquiries in the first place, see our web design for real estate service.
The verdict
An AI chatbot for real estate is worth it when your enquiries arrive faster, later or in more languages than your agents can answer. It is not worth it as a novelty widget that repeats your contact page. Most of the value comes from searching live listings in plain language and handing an agent a qualified, well-documented lead while the buyer is still interested.
Keep it narrow. Let it search, answer, qualify and book. Never let it negotiate, value or guess. Then judge it the way you would judge a new junior agent: by the viewings it books and the mistakes it makes, checked every week.
Frequently asked questions
What can an AI chatbot do for a real estate agency?
It can search your listings from a plain-language request, answer questions about specific properties, collect budget and timeline, book viewings and pass the conversation to your CRM. It should not negotiate, give valuations or give legal advice.
Will a chatbot replace real estate agents?
No. It takes over the first, repetitive part of an enquiry so agents spend their time on viewings and deals. In the US, NAR's 2025 buyer survey found that 88% of buyers still bought through an agent or broker.
Can a property chatbot answer in Bulgarian and English?
Yes, if it is built and tested for it. The hard part is search, not writing: matching Cyrillic, Latin-letter and English district names to the same listings. Test any tool with your own listings in both languages before launch.
How much does a real estate chatbot cost?
Off-the-shelf platforms charge a monthly subscription. A custom RAG chatbot is a project: at Ascend it starts with a two-week proof of concept, a production build takes four to eight weeks, and running costs for most agencies come to a few hundred euros a month.
Does the chatbot need to connect to my CRM?
It should. Without a CRM connection, someone copies enquiries by hand and the agent who calls back has no context. With one, every lead arrives with the transcript, the listings viewed and the answers to your qualifying questions.
Do I have to tell visitors they are talking to a bot?
In the EU, in practice yes. Article 50 of the AI Act, which applies from 2 August 2026, requires AI systems that interact with people to be built so people know they are talking to AI, unless that is already obvious. A clear label in the chat window is the simplest way to comply.
Want to see it on your own listings?
Send us twenty enquiries your agents received last month. We'll show you how a chatbot built on your live listings would answer them, and where it should hand over to an agent.
Send us your enquiries →