By Michele Di Sei · 10-Nov-2026
|
Key data
|
More than two in three buyers who type into the AI on an agency's website never ask it a second question. Those who do are four times more likely to leave their details, call or write to the agency.
I see plenty of agents doing exactly the same thing with the buyer sitting right in front of them.
The data comes from Propelos and hadn't been published before.
The question I left open in September
In “Why Do We Ask for One House and Fall in Love With Another?” I wrote about something that happened to me for years as an agent. The buyer would hand me a perfect list: four bedrooms, a good school, a pool, under 1.5 million. I'd find houses that ticked every box. They turned them all down. After five, ten or twenty viewings, we'd walk through a door that met half the list and, five seconds in, they'd say: “This is the one.”
The list fits in a filter. What makes people choose almost never does.
I ended that article with a question. Now that buyers can talk to an AI the way they'd talk to a person, will they finally say how they want to live?
I put it to Mihnea Gămulescu, founder of Propelos, over a couple of conversations in Marbella. His company makes the AI search tool some agencies have on their websites: buyers type in their own words and the AI shows them homes from the agency's listings. I asked him for real data. He sent me aggregated data from thousands of sessions, from late February to early October 2026.
Before you read it, bear the profile in mind. All five cases in the report are buyers from outside Spain, and those who set a budget cap put it, at the median, at one million euros. The typical Spanish buyer is different. But I recognise the way they write from any first meeting at the office.
How they write: three words
The median search is three words long. One in four is a single word, almost always the name of a town. Only one in eight reaches ten.
A buyer from Ireland, on a phone, started with “3-bed” and a well-known stretch of coast. Then added “apartment”. Then a 1.5 million cap.
What they ask for is pure filter. 63% of searches name a place, 32% a property type and 19% a budget. A feature of any kind shows up in only 14.5%, and when it does it's usually a pool, sea views or a terrace.
They have a tool in front of them that lets them say everything, and they fill in boxes. Years of property portals have taught them to search that way.
When they carry on, they write in shorthand. “Max 450,000.” “With a pool too.” “Show me the rest.” 63% of follow-up questions are four words or fewer, and they only work because the AI remembers what came before.
The second question changes the session
32% ask a second question. One in eight gets to three or more. And those who reach five or more open about seven times as many listings as those who stop at one.
The more they ask, the better they explain themselves. In sessions with three or more searches, the longest has a median of six words. The first, three.
Take this case (the wording is summarised). A buyer from Norway opens with “3-bedroom apartment”. Eight questions and 21 minutes later, they are writing what they're really after: two bathrooms, a big balcony, a big communal pool, a small complex rather than one with 200 or more flats, between 400,000 and 700,000 euros, in a real town, with restaurants, where people live all year round. They opened ten listings.
A town where people live all year round. That isn't a requirement. It's a way of living. And it wasn't in the first sentence.
Another buyer, from Slovakia, asked for a two-bedroom flat with sea views, up to 700,000 euros, off-plan. They came back 80 minutes later and changed course in plain words: a resale, not a new development. What they really wanted wasn't in the first question either.
Your team does the same
The “four times” needs reading carefully. It's a correlation: maybe the second question creates the interest, maybe it only brings it to the surface, because serious buyers ask more. The data can't tell the two apart.
When you're sitting across from a buyer, it makes no difference. Without the second question you don't know which of the two you've got.
The agent who stops at “three bedrooms, two bathrooms” looks for houses that tick the boxes, and the buyer falls for none of them. The one who asks “and what do you need three for?” finds the home office, the in-laws who come for two months a year, or the divorced father who wants his kids to feel this is their home too.
I've seen it since 2012: most agents do with the buyer what the buyer does with the AI. They stop at the first question. And then they show ten houses.
Before the conversation, the recommendation
This week, in a coaching session, a real estate business owner I work with told me two new clients had come to him. They had asked ChatGPT who to work with, and ChatGPT had recommended him. They didn't even go through Google.
The company that shared this data also works on getting AIs to recommend agencies. That changes the order: today the buyer walks through two doors. First, an AI names you when someone asks who to work with. Then someone on your team asks the second question when they arrive.
If you run an agency
Among buyers from other European countries, 56% of sessions start outside weekday office hours (9:00 to 18:00, local time). 27% start at the weekend.
And they write in their own language. 70% of full-sentence searches from Sweden are in Swedish; from the Netherlands, 69% are in Dutch.
Picture your website at eleven on a Saturday night, with a buyer typing in Swedish. That's where AI makes sense, because it's the only part of your agency that's open. But the sale is still closed by a person, and on Monday that buyer reaches an agent on your team with a three-word list.
At your next team meeting, pick five active buyers and ask each one's agent what the buyer wants it for. What for, in the buyer's own words. Every buyer the agent can't answer for is their next call, before they show that buyer one more house.
The list a buyer gives you is the one that fits in a filter. Your team's job is to find the other one.
Frequently asked questions
How many buyers ask a property website's AI a second question?
32%, according to aggregated data from thousands of sessions on Propelos' AI search tool on agency websites in Spain, from late February to early October 2026. Those who ask are four times more likely to leave their details, call or write to the agency. It's a correlation: the data doesn't show whether the second question creates the interest or only brings it to the surface.
How do buyers search for a home when they type to an AI?
The way they would on a portal. The median search is three words long: 63% name a place, 32% a property type and 19% a budget, and only 14.5% mention any feature. The more they ask, the better they explain themselves: in sessions with three or more searches, the longest has a median of six words and the first, three.
What can an agency owner or manager do with this data?
At the next team meeting, pick five active buyers and ask each one's agent what the buyer wants it for, in the buyer's own words. Every buyer the agent can't answer for is their next call, before they show that buyer one more house.
Data: Propelos AI Property Finder. Searches typed by buyers themselves on agency websites in Spain, from late February to early October 2026; the opening-hours figure covers late March to early October, with Europe on summer time. Figures are aggregated across agencies: no agency or buyer can be identified. Durations are minimums. In the cases, buyers' words are summarised and places are described in general terms. Thanks to the company's founder for sharing them.
Further reading: a guide to artificial intelligence for real estate agencies and what training an agency director needs.
Michele Di Sei, coach for real estate leaders.
