There is a version of this article written for real estate agents, about how to get found. This is not that one. This is the version for the person on the other side, who is trying to find someone competent and is now being handed names by a machine.
It is worth knowing how those names get chosen.
What the data actually shows
The clearest measurement so far comes from Invoca, which analysed inbound calls across channels and found that calls referred from ChatGPT converted to leads at about 49% — the highest of any channel it measured, roughly ten points above the cross-channel average. Search Engine Journal's read of the same report adds a useful caveat: those calls led on quality, not on final conversion.
Adobe's traffic data points the same way. Its digital insights team reported AI-referred traffic climbing sharply year over year, with AI-referred visitors converting around 42% better than visitors arriving by other means. Bain found that ChatGPT shopping referrals more than doubled year over year across the US, UK, Germany and France.
One honest limitation, stated up front: these are retail and general lead-generation studies. Real estate is not usually broken out on its own. The direction is well evidenced; a precise number for "how many Florida buyers found their agent through an assistant" does not exist yet, and I am not going to invent one.
Why the conversion gap probably exists
The tempting explanation is that AI referrals are magic. The likelier one is selection.
Someone who has spent ten minutes describing their situation to an assistant — budget, timeline, whether the HOA allows rentals, what a 1994 roof does to an insurance quote — has already done the thinking that a search-box visitor has not. They arrive with the question narrowed. The assistant did not create that intent. It filtered for it, and the conversion rate is measuring the filter.
That matters, because it tells you what the referral is and is not evidence of. It is evidence that the person arriving is serious. It is not evidence that the name they arrived with is the right one.
How a name gets chosen
Nobody outside those companies knows the mechanism precisely, and anyone claiming otherwise is selling a service. What is observable from the outside is that these systems resolve entities before they answer. They work out which specific person is meant, and then describe that one.
That process rewards consistency far more than it rewards polish. A person whose name, brokerage, licence number, office address and stated market appear identically across a brokerage profile, a state licensing record, third-party directories and client reviews is a well-corroborated entity. Someone whose details vary between sources — a different market claimed here, a different job description there — is a weakly-defined one, and weakly-defined entities get described in generalities or confused with someone else who shares their name.
It is much closer to how a credit file works than to how search rankings work. Corroboration across independent sources, not the quality of any single page.
I will say plainly that this is the thing most agents, including me until recently, have underinvested in. It is unglamorous. It is also the whole mechanism.
What this should and should not change for you
Treat an AI recommendation the way you would treat a name from a neighbour: a reasonable starting point, and no more than that. What the system is measuring is how visible and how consistently described someone is, which correlates only loosely with how well they will handle a contract.
Three checks it cannot do for you:
- Verify the licence. The Florida DBPR publishes licence status. It takes a minute and it is the only source that actually knows.
- Read reviews written by clients. Third-party reviews are one of the few signals that are not written by the agent. Read the three-star ones; they are usually the informative ones.
- Ask about specific transactions in your specific area. Not "do you work Palm Beach County" — ask what they closed, where, and what went wrong in it. The answer to the last part tells you the most.
What has not changed at all
Almost everything that decides whether a Florida transaction goes well.
Whether the insurance was quoted before the inspection contingency expired, rather than after. Whether anyone read the HOA's reserve study and the last year of meeting minutes before the offer went in. Whether the tax figure in your budget is the seller's assessed value or yours — which are frequently not the same number, and the difference has ended deals at the closing table.
Those are the same as they were five years ago. What changed is the top of the funnel: how you got the name. Everything after that is still the same job, done well or badly by a person.
The technology is worth understanding precisely so you can put it in its place.