Good real estate chatbot conversations identify the visitor's task, ask only the detail an agent needs, and end with a real handover. A buyer viewing request should carry a property reference and preferred time; a seller valuation request should carry property type and timing. The five scripts below show the exact exchanges and the agent trigger.
These are illustrative conversations to adapt, not claims that a particular firm or product used them. They assume the chatbot can record answers, attach the transcript and route a task to a monitored queue. Test those abilities in the product you choose. A script that sounds polished but loses the property reference at handover is a bad lead flow.
Set the conversation contract before copying a script
Every flow starts with a truthful identification: “I'm an AI assistant for the property team.” Tell the visitor whether the bot can answer from a live listing, collect an enquiry, or request a call. Give “Speak to an agent” as a visible choice. If a visitor enters that phrase in free text, treat it as the same choice. Do not require them to answer a budget question first.
The bot needs a small shared set of fields. Use intent for the visitor's goal, property reference when one exists, question for their own words, contact preference for the reply channel, permission for the follow-up they agreed to, and owner for the team member or queue that receives the record. Optional fields differ by flow. A buyer might volunteer a preferred viewing period; a seller might volunteer a broad selling timescale. Keep an “unknown” value instead of forcing a guess.
Do not write a bot message that says “Booked” when the only action was sending a request. Do not let it give an unreviewed valuation, make an offer on anyone's behalf, or infer personal characteristics to decide which listings to show. Those decisions change the visitor's position and need an accountable human process. If your firm needs more detail on a property chatbot's launch rules, use the qualification and handover checklist alongside these scripts.
All five examples use fictional property references and figures. Replace them with values from a reviewed listing source. The sample words “[first name]” stand for a field the visitor chooses to provide; the bot should not invent a personal name. A direct reply should always be possible without agreeing to unrelated marketing.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowFlow 1: a buyer asks to view one listed flat
The visitor is on listing P-214, advertised at an illustrative $420,000. The page passes P-214 into the chat so the visitor need not type it again. The bot should ask for a preferred period, not display a fixed appointment unless a booking system has confirmed it. A buyer who asks about parking is also asking a factual question that may require an agent.
Bot: “I'm the property team's AI assistant. Are you asking about P-214, arranging a viewing, or hoping to speak to an agent?”
Visitor: “Can I see it on Thursday afternoon? Is there a parking space?”
Bot: “I can pass Thursday afternoon and your parking question to an agent. I can't confirm a slot or the parking detail from this chat. Would you prefer an email or a call back?”
Visitor: “Email, please.”
Bot: “Please share the email address you want the agent to use for this enquiry. You can also speak to an agent without answering more property questions.”
Visitor: “[email address]”
Bot: “I have sent P-214, your Thursday preference and parking question to the viewing team. The viewing is a request until an agent confirms it.”
Handover trigger: the visitor asks to view a specific property. The bot creates a viewing task immediately after receiving a contact route. The agent receives P-214, Thursday afternoon as a preference, the parking question, the transcript and the email permission. The first human action is to check whether P-214 is still available, verify the parking detail and offer actual viewing times. If the task creation fails, the last bot line must change to a clear fallback; it must not claim the team has received anything.
A weak variant says “Thursday at 3 is confirmed, and parking is included” because the model fills gaps with plausible details. That would create two false promises. In a test, deliberately leave the parking field empty and block Thursday's slots. The correct conversation should still capture the lead while acknowledging both uncertainties. If the visitor refuses to give an email, offer the site's ordinary contact route or a human chat rather than ending with “I can't help”.
Flow 2: a buyer wants a shortlist, not an interrogation
Here the visitor has not selected a listing. The goal is to capture property criteria in the visitor's own terms and give an agent a useful starting point. Asking for exact finance or a full household profile at this stage can feel intrusive and may add little to the first reply. Keep the branch to two criteria and one next action. The agent can discuss the rest later.
Bot: “I'm an AI assistant. I can help pass your property search to an agent. Are you looking for a particular listing or a shortlist?”
Visitor: “A shortlist. Two bedrooms and somewhere I can work from home.”
Bot: “I have two bedrooms and space for a work area. Is there a price range you would like the agent to use, or would you rather skip that?”
Visitor: “Around $450,000, but I could stretch if the layout works.”
Bot: “Would you prefer links to suitable public listings first, or a conversation with an agent?”
Visitor: “Send links first.”
Bot: “What contact route may the agent use for this shortlist? The agent will check current listings before replying.”
Handover trigger: the visitor requests a shortlist. The record should preserve “space for a work area” as free text and mark $450,000 as approximate. It should not turn “could stretch” into an unlimited budget. If the bot has a live, reviewed listing feed, it may show matching public results; otherwise it should leave the shortlist to an agent. A list generated from old web pages can contain sold properties or incorrect prices.
Illustrative agent note: “Buyer seeks two bedrooms and workable desk space; approximate $450,000, flexible for layout; asked for links first; no specific property chosen.” The human can then send a small set of verified listings and ask one useful follow-up, such as whether a separate room is needed. The bot need not ask about family status, background or any other personal cue to produce a shortlist. Fair housing rules prohibit steering, such as discouraging someone from viewing a home or directing them to one part of a neighbourhood because of a protected characteristic, so a neutral, criteria-based route is the safer design.
If the visitor answers “I don't know the budget”, keep the conversation moving. Offer the human route and a way to describe preferred features. A compulsory price field can make the bot discard an enquiry that an agent could handle. For the wider question of what property bot features to demand from a vendor, see the real estate chatbot features checklist.
Flow 3: a buyer asks about a changed listing
The listing page says P-318 is available at an illustrative $510,000, but the firm's internal record is being updated. This is a test of uncertainty, not a chance for the bot to improvise. The bot should carry the reference, capture the precise question and connect the visitor to an agent. It should not replace the listing with another property without asking the visitor.
Bot: “I'm the AI assistant for the property team. What would you like to know about P-318?”
Visitor: “The page showed $510,000 yesterday. Is that still the asking price? Can I make an offer?”
Bot: “I can't verify the current price from the information available to me. An agent needs to confirm it and discuss any offer. Would you like me to send your question and P-318 to the property team?”
Visitor: “Yes. Please call me this afternoon.”
Bot: “Please share a number for this enquiry. I can pass on your afternoon preference, but the team must confirm when someone can call.”
Visitor: “[phone number]”
Bot: “Your price and offer question has been sent to the team. No offer has been made through this chat.”
Handover trigger: a conflict in price information or an offer question. The agent's record needs the visitor's exact wording, the page reference and the time the chat occurred. The agent verifies the latest approved price and the status of the listing before replying. The bot should not collect a proposed offer amount unless the firm has designed and reviewed a formal offer process; even then, the language must make clear what counts as submission and what does not.
A useful failure test changes the listing during the conversation. Start the bot when the page contains $510,000, then update the authorised record to a different figure before the visitor asks again. The acceptable behaviour is to consult the current authorised record or say it cannot verify the figure. Repeating the earlier figure as if it were current fails the test. The chatbot promise guardrails explain how to set similar boundaries for customer-facing bots.
Flow 4: a seller requests a valuation appointment
A seller lead is often treated as a demand for an instant estimate. The useful first step is to understand what service the seller wants and route it to the right person. The bot may ask for property type and broad timing, then let an agent explain the firm's valuation process. It should not produce a confident price from a short chat or claim that a professional has assessed the property.
Bot: “I'm the agency's AI assistant. I can pass a valuation request to an agent. Are you looking to arrange a conversation about selling, or do you have a question first?”
Visitor: “I may sell my three-bedroom house. Can someone value it?”
Bot: “Yes, I can ask an agent to discuss a valuation with you. Are you thinking of selling soon, later this year, or are you still deciding?”
Visitor: “Still deciding. I want to understand the likely process.”
Bot: “I will pass on that you're still deciding. Would you prefer an email or a call to discuss the process?”
Visitor: “Email. Please don't call.”
Bot: “Understood. Please share the email address for this enquiry. I will ask an agent to reply by email about the process and a valuation appointment.”
Handover trigger: any request for a valuation or selling advice. The record contains “three-bedroom house”, “still deciding”, “wants process information” and “email only”. The agent should reply by email, not call because that would disregard the visitor's explicit preference. The agent can explain what information and appointment are needed before any valuation is offered. If the visitor later chooses a call, the record can be updated.
The bad variant asks for a full address before explaining what will happen next, then assigns a score of “low intent” because the seller is undecided. That is a design choice, not a fact about the visitor. An undecided seller may want helpful information and a gentle reply. The bot can request location details only when necessary for the next agreed service, and the firm should explain how the details will be used. For a deeper look at the specific appointment process, see the valuation enquiry chatbot tutorial.
Flow 5: a seller has a time-sensitive question
This seller is already in a transaction and has a question about a buyer's request. It may look like a new seller lead to a simple classifier, but routing it to the general valuation queue could delay a real issue. The bot should recognise that the visitor needs the existing transaction team, stop providing advice and send the wording to the right human owner.
Bot: “I'm the AI assistant. Are you asking about selling a property, an existing transaction, or something else?”
Visitor: “My buyer wants to change the completion date. Can we agree to it today?”
Bot: “That needs your transaction team. I can't advise on whether to agree or confirm a change. Would you like me to pass your message to the team handling the transaction?”
Visitor: “Yes. Please tell them I can speak after 4.”
Bot: “What is your preferred contact route for this message? If you have an existing reference, you may add it, but you can skip that.”
Visitor: “Call me after 4 on [phone number]. Reference T-73.”
Bot: “I have sent your request, reference T-73 and contact preference to the transaction queue. The team must confirm any change.”
Handover trigger: an existing transaction, date change or request for legal or contractual advice. The bot records the exact words and reference T-73 but does not decide whether the change is possible. It must not ask the visitor to upload contracts into an unreviewed chat tool. If there is no transaction queue or the reference is unknown, route to a staffed general inbox and make the uncertainty visible to the visitor.
A second version of this example is an ambiguous message: “I need to sell quickly because the purchase fell through.” The bot should not infer that the visitor has an active transaction with the firm. It can ask one clarifying question: “Is this about a property your team is already handling, or would you like a new selling conversation?” Both answers need a human, but the owner differs. This branch prevents an urgent existing client from being buried among new enquiries without making the bot provide professional advice.
Calculate the queue load before promising fast replies
Imagine an illustrative four-agent property team that receives 60 chats in a month. Eighteen resemble Flow 1, 12 resemble Flow 2, six resemble Flow 3, 16 resemble Flow 4, and eight resemble Flow 5. Those counts total 60. Suppose 45 visitors request or accept human follow-up and 15 only ask a simple question. The team should test whether its queues can absorb those 45 handovers, rather than judging the bot by how many names it collects.
Assign one owner and one backup to each queue. Viewing and buyer shortlist tasks go to the buyer team, valuation requests to the seller team, and changed listing or existing transaction questions to a priority human review queue. In a pilot, say 43 of 45 follow-up tasks reach the intended queue, while two Flow 5 cases are mislabelled as valuations. The routing accuracy for tasks is 43 divided by 45, about 96% in this illustrative set. The two errors matter more than the percentage because they concern active transactions.
Now count human action separately. Say 37 of the 45 tasks receive a first human response during the firm's stated monitored period. The gap of eight needs inspection: two were misrouted, three arrived when the owner was absent, and three had no usable contact route. The remedy differs by cause. Change the classifier for the two transaction messages, add backup cover for the three absences, and make the contact request clearer for the remaining three. Do not simply add another automated reminder and call the queue fixed.
The team can reduce unnecessary transfers without hiding the human route. If an up-to-date listing source answers a basic question correctly, the visitor may leave satisfied. But an offer, valuation request, conflicting fact or wish to speak to someone remains a human task. A sensible test is whether the first agent message can address the visitor's actual question without asking them to repeat the entire chat. That is a better quality check than a high chatbot completion rate.
Turn each script into fields, branches and a test case
A rule-based builder can implement these flows with question actions, contact properties, conditional branches and team routing. HubSpot's list of rule-based chatbot actions includes Ask question, Set contact property value, If/then branches, Send to team member and Book a meeting, and the team-member handover checks availability before it routes. That is an example of product capability, not a claim that every chatbot subscription supplies them: on that page, custom if/then conditions need a Professional or Enterprise subscription, and Send to team member needs an assigned Sales or Service seat. If your chosen tool lacks a branch, use a shorter form and a monitored inbox rather than pretending that the bot can route intelligently.
- Write the first visitor message, the expected intent and the owner for each of the five flows. Include one spelling variation and one interruption by “speak to an agent”.
- Define the minimum fields. Flow 1 needs the listing reference and viewing preference; Flow 4 needs the seller's request and reply preference. Contact details are collected only for the agreed reply.
- Write the bot's allowed answer for every property fact. If a fact is missing or stale, the allowed answer is an uncertainty plus an agent route.
- Configure the route and fallback. Test the case where the first owner is away and the case where the integration fails to create a task.
- Run the conversation in the live channel, inspect the record the agent sees, and compare it with the visitor's actual words. Correct any changed meaning.
Use a small filled-in test record to make that inspection concrete. For Flow 1, expected intent: viewing. Expected reference: P-214. Expected time: Thursday afternoon, unconfirmed. Expected factual question: parking space. Expected owner: buyer viewing queue. Expected final status: request sent, no appointment booked. If the resulting record instead says “Thursday 3 pm booking confirmed”, the test fails even if the bot's visible text sounded cautious.
For Flow 4, expected intent: seller valuation conversation. Expected timing: undecided. Expected contact: email only. Expected owner: seller enquiry queue. If a salesperson calls, the failure is in the downstream process, even if the chatbot stored “email only” correctly. The script and the team's working practice have to agree. This is why the agent view and notification route deserve the same attention as the messages the visitor sees.
One final audit checks unequal treatment. Run Flow 2 with the same property criteria in two test conversations that differ only in a personal cue, and compare the listings, available choices and handover route. Do not let the bot infer protected characteristics or steer visitors on that basis. A qualified adviser should review the firm's fair housing and data handling obligations where applicable. The test is a practical screen for obvious defects, not a legal certification.
Keep these five scripts short enough for a visitor to leave or ask for a person at any stage. Save the visitor's own words, route to an owner who can act, and make the acknowledgement match what the system actually did. Those three checks turn a plausible chatbot exchange into a usable property enquiry.
Questions about adapting the five scripts
Should every visitor answer all the questions in a script?
No. Treat the script as a set of useful branches, not a compulsory form. If the visitor asks for an agent, the bot should stop collecting details and route the conversation. If the visitor already supplied a property reference, timing or contact preference, the bot should use it rather than asking again.
Can a chatbot give a seller an instant property valuation?
A bot can collect the details needed for an agent to arrange a valuation conversation. It should not present a number as a professional appraisal unless the firm has a separately reviewed valuation service and clear wording about its limits. The sample seller flows deliberately promise a human follow-up rather than an instant figure.
What should the bot do if a listing changes during a chat?
Treat the listing record as uncertain, stop repeating the old price or availability, and pass the precise question to an agent. If the bot can read an authoritative live record, it should still show the last update and avoid claiming a viewing or offer is confirmed until the relevant booking or transaction system confirms it.
Further reads
- Real Estate AI Chatbot Use Cases: Which Three to Start With — Choose which property conversations to automate first.
- Real Estate Chatbot Pricing: What Agents Pay for AI Lead Capture — Estimate the cost of the chosen lead flow.
- Real Estate AI Chatbot Demo: A Test Checklist Before You Sign — Make a vendor demonstrate the failure cases.
- When Should an AI Chatbot Hand Over to a Human? — Set general escalation rules beyond property enquiries.
- How to Automate Buyer and Vendor Follow-Up With AI — Plan the human follow-up after the chat.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: HubSpot Knowledge Base, Choose your rule-based chatbot actions (question fields, conditional branches, team handover and meeting actions); fair housing steering regulation text, reviewed 4 November 2026.