AI Home Tours: What Buyers Actually Ask at 11pm

The Question Nobody Is Awake to Answer

Buyers do not shop for homes during business hours. They shop in bed, on a phone, after the kids are down, with a dozen tabs open and a spreadsheet they will never finish.

Somewhere around 11pm they hit a question the listing page does not answer. How old is the roof. What are the taxes actually going to be. Is that addition on the same HVAC system. And there is nobody to ask.

That question is rarely trivial. It is usually the one that decides whether the listing stays on the shortlist or gets closed and forgotten by morning. Listings lose to silence more often than they lose to a better house.

An AI home tour exists to answer that question in that moment, from the property record, without pretending to be a licensed professional. This post covers what buyers actually ask, which questions an AI should answer, which ones it must refuse, and which ones belong to a human every time.

The examples below are drawn from a research set of 385 real queries harvested from Google autocomplete, People Also Ask panels, and related searches. These are queries people actually type. None of it is a hypothesis about buyer behaviour. It is the phrasing buyers use when they think nobody is listening.

Five Things Buyers Ask After Hours

Sort the questions and they fall into five groups. The groups matter because each one has a different risk profile and a different correct handler.

1. Property facts

These are the questions a listing sheet should answer and often does not, or answers in a format nobody reads on a phone.

• "how old is the roof"
• "what year was this house built"
• "square footage above grade vs total"
• "is the basement finished"
• "what kind of heating does it have"
• "how old is the furnace and water heater"
• "does it have central air"
• "is there a garage and how many cars"

Every one of these has a factual answer sitting in the listing record, the disclosure packet, or the tax assessment. The buyer is not asking for judgement. They are asking for retrieval.

2. Cost of ownership

The second wave, and the one that quietly kills more shortlists than any other, is the cost of the house after the price.

• "what are the property taxes on this house"
• "how much is the HOA fee and what does it cover"
• "average utility bill for a house this size"
• "is this in a flood zone"
• "how much is insurance going to be here"
• "are there any special assessments"

Some of these are retrievable facts. Some, insurance in particular, are estimates that vary by carrier and buyer, and an AI should say so rather than produce a number.

3. Neighborhood and location

This is where the questions get dangerous, and we will come back to it in the next section.

• "how far is this from downtown"
• "what is the commute to the hospital"
• "is there a grocery store nearby"
• "how close is the highway and is it loud"
• "what school district is this in"
• "is this a good neighborhood"

Notice the mix. Distance to a hospital is a mapping question. "Is this a good neighborhood" is a fair housing question wearing a casual disguise.

4. Process

Buyers who are getting serious start asking about mechanics, usually before they are willing to call anyone.

• "how do I make an offer on this house"
• "how long has this been on the market"
• "do I need a pre approval first"
• "what happens after an offer is accepted"
• "can I see it this weekend"
• "is the seller looking at multiple offers"

These are the highest-value questions in the whole set, because they are buying signals in question form. Some can be answered generically. Some cannot be answered by software at all.

5. Comparison

The last group is the buyer doing the agent's job badly, on their own, at midnight.

• "how does this compare to the one on the next street"
• "why is this cheaper than similar houses"
• "is this priced high for the area"
• "what did this last sell for"

Sale history is a record. Whether a price is fair is a professional opinion, and the moment an AI offers one it has stepped into work that belongs to a licensed person.

The Three-Column Rule: Answer, Refuse, Hand Off

This is the part worth printing out. Every question a listing AI can receive belongs in exactly one of three buckets, and the bucket has to be decided before the question ever arrives, not improvised in the moment.

The refuse column is where fair housing lives

The middle column is not a limitation of the technology. It is the point of the design.

"Is this a good neighborhood for families" reads as friendly small talk. It is an invitation to steer, and any answer that ranks an area by the people in it is a fair housing problem regardless of whether a human or a model produced it. The same is true for safety questions, "people like me" questions, and anything asking the system to compare areas on characteristics that map to protected classes.

A well-built system does not answer these and then add a disclaimer. It declines the framing and redirects to objective, sourceable information the buyer can evaluate themselves: commute times to places they name, published district assignments, public data sources they can go read. Our AI policy sets out the operating rule directly: offer the same categories of information to similarly situated users, and avoid directing people toward or away from areas based on protected characteristics.

A refusal is also not a dead end. The good pattern sounds like: "I cannot rate neighborhoods, and I would not be a good source if I tried. I can tell you the commute to any address you give me, the district assignment on record, and where to find the published local data. Want me to pull any of those?"

The deeper treatment, including how to write refusal language that does not sound evasive, is in our post on fair housing compliant AI.

The handoff column is where the money is

Notice what is in the third column. Negotiation. Legal wording. Financing. Investment judgement. Inspection strategy.

These are not questions the AI failed to answer. They are questions whose arrival means the buyer has moved from browsing to transacting. Every one of them should trigger a route to a person, and the routing is a feature, not a fallback.

Grounded Answers and Why Model Recall Is a Liability

There are two ways a system can answer "how old is the roof."

It can look up the roof field in the listing record, the disclosure, or the permit history, and report what it found, including reporting that the field is empty. Or it can generate a plausible-sounding answer from what a model absorbed about houses in general.

The second one is not a feature. It is an unattributed statement about a specific property, made in the listing brokerage's voice, that nobody can trace to a source. That is a disclosure problem with a chat interface on it.

Approved-data grounding in practice means four things:

1. A defined source list. The tour reads from the MLS listing record, media, room details, disclosures, showing instructions, tax and assessment data, sale history, and brokerage-approved knowledge. That is the universe. If it is not in the universe, it is not an answer.
2. Empty is a valid answer. "That is not in the record, and I can ask the listing agent" is correct and defensible. A guess dressed as a fact is neither.
3. Attribution the buyer can see. Answers point back at where they came from, so a buyer can open the disclosure rather than take the system's word for it.
4. Estimates labelled as estimates. Insurance, utilities, and anything carrier-specific or buyer-specific get framed as ranges to verify, never as the number.

This is the same discipline that governs our broader AI agents for real estate work: agents retrieve from approved context, produce outputs a staff member can audit, and stop at the edge of professional judgement.

The Handoff Moment

The most useful thing an AI tour does is not answer questions. It is noticing when the questions have changed.

A buyer asking about square footage is browsing. A buyer asking how to make an offer, whether the seller has other offers, or whether they can see it Saturday is not browsing anymore. The vocabulary shift is the signal, and it is detectable.

What should happen next:

• The system offers a live human immediately rather than continuing to answer. "That is a question for the listing agent. Want me to connect you now or set up a call?"
• The full conversation goes to the CRM as context, not as a bare lead notification. Which rooms they revisited, what they asked twice, which objection they raised, what timing they mentioned.
• Timing and readiness signals get captured as fields the agent can filter on, so a Monday morning follow-up list sorts by intent rather than by timestamp.
• If a showing request comes out of it, it flows into the route planner with the other stops for that day instead of becoming a separate scheduling thread.

The difference between a chat widget and a tour is exactly this. A widget hands you an email address. A tour hands you a buyer plus a transcript of what they cared about at 11pm on a Tuesday.

What This Actually Changes for the Agent

Be clear about what this is not. It does not replace showings, it does not replace you, and it does not close anybody. Any vendor telling a brokerage otherwise is describing a product that does not exist.

What it changes is the shape of the top of the funnel:

• Qualification happens before the call, not on it. The basics have been answered. The first live conversation starts further along.
• Prioritisation gets real inputs. An agent working a long inquiry list can sort by what people actually asked instead of by who filled out a form.
• Fewer wasted showings. Buyers who would have discovered a dealbreaker in the driveway discover it at home, which is better for everyone including them.
• The after-hours gap closes. Questions asked at 11pm get answered at 11pm, which is the only time that answer was worth anything.
• Consistency across a roster. For a brokerage, the refusal rules and answer boundaries are configured once and apply to every listing, rather than depending on which agent is replying.

For luxury teams the calculus tilts further, because showing time is scarcer and the pre-qualification value per inquiry is higher.

What stays yours: the negotiation, the judgement, the read on the seller, the relationship, and every decision that carries a license behind it. The AI covers retrieval and routing. You cover the rest.

Where to Start

If you are evaluating this, the questions to ask a vendor are the ones in this post. Where do answers come from. What does it refuse. How does it escalate. Where does the conversation land.

• See how tours get built from listing data on the AI Home Tours page.
• Check the feed side first, since the tour is only as good as the IDX and MLS connection behind it.
• Review the answer boundaries and grounding rules in our AI policy.
• Look at the full platform if the tour needs to connect to CRM, events, or member systems.
• Common setup and scoping questions are on the FAQ, and cost structure is on pricing.

Or bring your MLS, your CRM, and your listing count to a scoping conversation and we will map what it takes.

Frequently Asked Questions

How much does an AI home tour cost per month?

There is no single monthly number, because cost is driven by how many listings you tour-enable, how much media and property data each tour ingests, how much conversation volume you expect, and whether voice is included alongside chat. Integration scope matters too: pulling from IDX, tax records, and your CRM is a different build than a standalone widget on one listing. Association-wide and brokerage-wide deployments price differently from single-team pilots. See the pricing page for current structure, or bring your listing count and integration list to a scoping call.

Is there an AI that will create a virtual tour for me?

Yes, tools exist that assemble a guided walkthrough from listing content you already have, rather than requiring a new camera shoot. AI Home Tours builds the tour path from approved MLS listing data, photos, room details, disclosures, and neighborhood context, then attaches a chat and voice layer on top of it. The important distinction is between tools that only generate a visual sequence and tools that can also answer questions about the property. A tour that cannot answer questions is a slideshow with better transitions.

Which AI is best for a home tour on a real estate listing?

The best one for a listing is the one grounded in that specific property's record, not a general-purpose assistant improvising from training data. Evaluate on four things: where its answers come from, what it refuses to answer, how it escalates to a human, and whether the conversation lands in your CRM. A general chatbot will happily answer a question about roof age it has no way of knowing. That is the failure mode you are buying protection against.

Can AI answer questions about a listing?

It can, and it should, as long as the answer is retrieved from the listing record, tax data, disclosures, or other approved sources rather than generated from model recall. Grounded answers cite what they are drawn from and say so plainly when a field is not in the record. Unsourced answers about a specific property create disclosure and accuracy exposure for the listing brokerage. Grounding is not a technical nicety here; it is the entire safety model.

Do AI home tours replace showings?

No, and a vendor promising otherwise is selling you something worth being skeptical about. The tour handles the self-education phase that used to happen through unanswered voicemails and email tag, so buyers arrive at a showing already past the basics. The practical aim is fewer wasted showings and better first conversations, not fewer showings overall. The in-person visit is still where the decision gets made.

How do I add an AI tour to my listing?

In practice it starts with the data feed rather than the listing page: the tour needs an approved source for listing content, media, disclosures, and showing instructions, which usually means an IDX or MLS connection. From there the tour path is generated per listing, the answer boundaries and refusal rules are configured once at the brokerage or association level, and the handoff is wired into your CRM. Individual listings then inherit the setup automatically as they come on market. Talk to us with your MLS and CRM details and we can map the sequence.