Short version: if you're a real estate agent adding AI in 2026, put it on follow-up first. Answering a new lead in two minutes instead of two hours is worth more than any listing description a model can write. The tools are cheap now. A phone number costs about $1 a month, transcribing an hour of conversation runs roughly 20 cents, and a recorded, transcribed, summarized call comes to about a penny a minute all in. Marketing copy is where most agents start, and it's the lowest-value thing on the list.
I should say up front that we build software that does this, so weigh what follows accordingly. I'd rather tell you what actually moved the needle than pretend I'm neutral.
What should a real estate agent use AI for first?
Follow-up. Specifically, the gap between a lead arriving and a human replying.
Every agent I've talked to knows their response time is worse than they'd like, and almost none of them know what it actually is. When you measure it, the number is usually embarrassing, and it's usually caused by something boring: the lead came in at 7pm, you were at a showing, and by the time you saw it the next morning three other agents had already called.
That's the problem AI is good at right now. Not writing your bio. Not generating a headline for a Woodland Hills listing. Catching the thing at 7pm.
Why does follow-up speed matter more than better listing descriptions?
Because a listing description competes with other descriptions, and a response time competes with silence.
If your copy is a bit worse than the next agent's, you lose a small amount of polish. If your response is four hours slower, you lose the client, and you don't find out you lost, because nobody emails to say they went with somebody else.
There's a second reason, less obvious. A lead who hears from you in two minutes is talking to you while they're still looking at the house. A lead who hears from you tomorrow has already had a night to cool off and a morning to book a showing with someone else. Same lead, completely different conversation.
What does AI actually cost a real estate agent in 2026?
Less than people expect, and the pricing has moved a lot in the last eighteen months.
Here's what the underlying pieces cost, at list, if you were buying them yourself:
- A phone number: about $1 a month
- Voice minutes: under a cent a minute
- Recording a call: about two tenths of a cent a minute
- Transcribing with speaker labels: roughly 20 cents an hour of audio
- A short AI summary of a call: a penny or two
A busy agent doing 400 calls a month, with maybe 120 of those actually connecting, lands somewhere around nine dollars. That's the whole stack: the calls, the recordings, the transcripts, and a written summary of every conversation filed against the right client.
Compare that to what dedicated dialers charge per seat and you can see why the category is getting rebuilt. The infrastructure got cheap and the pricing hasn't caught up.
What about recording laws?
This part people get wrong, and it's worth getting right before you turn anything on.
California is an all party consent state. Penal Code section 632 makes it a criminal offense to record a confidential communication without everyone's consent, with civil damages available on top. Twelve or so states work this way. The rest are one party consent, meaning you can record a call you're part of.
If you're in California and you record client calls, you need consent from the other side. An announcement at the start of the call is the usual way. It feels awkward for about a week and then nobody notices it. What you can't do is record everyone quietly and sort it out later.
Where should an agent keep a human in the loop?
Anything with your name on it that a client will read.
I'm comfortable with AI drafting a follow-up email, summarizing what was said on a call, flagging which five people to ring back today, and telling me a client I haven't touched in six weeks just looked at three listings. I'm not comfortable with it sending anything to a client without me seeing it, and I don't think you should be either.
The failure mode isn't the model saying something offensive. It's the model being confidently wrong about a price, a date, or a contingency, in writing, to somebody who will act on it. Drafts are fine. Autopilot on client communication is not.
How do you start this week?
Find out what your actual response time is. Not your impression of it. Go back through the last twenty leads and check the timestamp on the enquiry against the timestamp on your first reply. Most agents are off by a factor of five when they guess.
If that number is bad, fixing it does not require buying anything. It requires knowing which leads are sitting. Whatever you use to catch them, a CRM alert, an assistant, a piece of software, the point is the catching, not the intelligence.
The AI part earns its keep on the second problem, which is memory. Nobody remembers what a client said in March. Something that recorded it, wrote it down, and can hand it back to you before the next call is doing work no human assistant was ever going to do for nine dollars a month.
That's the honest case for it. It's less exciting than the pitch you usually hear, and it's the part that actually shows up in your closings.
De'Andre Willis Founder, Newmarket Edge