AI for Real Estate Agents: The Five Steps I Taught a Room of REALTORS

I spent an hour and a half in front of a room of REALTORS at the Southland Regional Association of REALTORS talking about artificial intelligence. Safety protocols, workflow, the whole thing. Some of it was boring. Boring is fine. Boring is what keeps you out of trouble.

But the part that actually mattered was not a tool, and it was not a prompt. It was the order you do things in. Almost everybody I talk to is doing the right things in the wrong sequence, and that is why it is not working for them.

Before any of that, though, we need to deal with what is being sold to you right now.

The system somebody is selling you does not exist yet

There is a version of this you have seen advertised. You sit down, you say "I am a real estate agent, make it work," and the machine goes and builds your business while you go fishing.

That system is coming. I do not know how many years out it is. Right now, it is not that.

And somebody is promising you it already is.

The pitch, decoded

The pitch usually sounds like this. Pay ten thousand, twenty thousand, thirty thousand dollars, one time, and they install the system. Sometimes there is a better-sounding version: you pay nothing at all until it produces for you.

That second one is the one I want to take apart, because it is the one that works on good people.

Here is the mechanism. You are not paying them anything up front, and that is technically true. What you are doing is paying Meta. You are paying Google. You are paying LinkedIn. You are funding the ad spend directly, on your card, to the platform.

They do the setup. And the setup is not nothing, but it is also not the heavy lifting they describe. It is a couple of five-pound dumbbells.

Then the ads run. And ads produce leads. That is what ads do. If you advertise on a pay-per-click platform, in any of these channels, you will get leads. The quality is another conversation, and for real estate, insurance, attorneys, and mortgages, the cost per click is steep, because everybody in those verticals is bidding against everybody else.

But the leads come in. And the moment they do, the promise has been fulfilled. The guarantee has been satisfied. Now the card you have on file gets charged.

Leads are not clients

This is the sentence I want you to write down.

The guarantee is almost never written against closed business. It is written against leads, or bookings, or qualified calls. Not money in your account. Not a signed listing agreement. People who clicked.

So you can spend real money on ads, hit whatever the contractual definition of "results" is, owe the full fee, and have closed nothing at all. The risk reversal is theatrical. The money still leaves.

If somebody puts one of these in front of you, ask two questions and you will know everything you need to know:

1. What is the monthly ad spend requirement, and why that specific number?
2. What is the exact contractual definition of "results"?

If the answer is vague, or if it gets redirected to a call, or if you have to speak with a manager who turns out to be a very good salesperson, you have your answer.

Nobody selling you AI cares about your license

I watched something like fifty-nine of the top videos on artificial intelligence for real estate agents. That was the research for this talk, and it taught me a lot, most of it discouraging.

Nobody cares about your license. Not one of them. They could not care less that you carry a fiduciary duty, that you have continuing education requirements, that there is a Department of Real Estate that can take your livelihood away.

What they care about is selling you something. And the thing that gets me is that there does not appear to be a conscience attached to it. They know they are selling a false bill of goods. They know it will not do what they said. And they do it anyway, because they can monetize it on the way out the door, and repercussions later are somebody else's problem.

We all want the extra leads. That is the lever they pull.

So let us do the real thing instead.

Step 1: Pick one model, and actually pick it

The first question in every room is always the same. Which one should I use?

My answer is a question back. Which one do you like? Have you played with any of them?

The main options are ChatGPT from OpenAI, Claude from Anthropic, Grok from xAI, Gemini from Google, and Copilot from Microsoft. There is also a Meta assistant inside Facebook. I think that one is a waste of your time right now, though I said the same about plenty of things that later ate the world, so check back.

They are all constructed a little differently. The personalities differ. That is not marketing, it is a real consequence of how each one was trained. Anthropic built theirs on something they call constitutional AI, and it does behave differently as a result.

Right now they are running neck and neck. This week one model is on top because of some change. A couple of days later a different one leapfrogs it. It really is a playoff. Do not agonize over picking the eternal winner, because there is not one.

Pick the one whose personality you can stand talking to every single day. That is the actual criterion, and nobody says it out loud.

One warning about starting over

Here is something people do not think about.

If you have already spent a couple of months on a paid plan treating one of these like a Google search box or a vending machine, that model has a couple of months of memory built on that behavior. Simple thing in, simple thing out. You are not winning any prizes with what it produces, because you are not using a fraction of what it can do.

That memory is strong. And it is memory of a shallow relationship.

In that specific case, it might actually be worth starting fresh somewhere else, with the intention of doing it right from day one. Not because the first model was bad. Because the relationship you built with it was.

Step 2: Commit

Once you have picked, commit.

Twenty dollars a month. Exchange rings at the beach.

I am not being cute about that. Here is why it matters. The free tiers do not carry memory in any serious way. And memory is the entire game. Everything I am about to tell you in steps three, four, and five is an investment in a system that remembers. If it forgets, you did the work for nothing.

Test on free. Then pay. Do not build a long-term relationship on a system that resets.

Step 3: Teach it who you are, and talk, do not type

This is the step almost everybody skips, and skipping it is why their results are mediocre.

Most people open the box and immediately ask it for something. Write me a listing description. Give me a social post. That is the vending machine relationship, and the output reflects it.

Instead, before you ask it for anything, tell it who you are.

I am Connor. I am from here. I grew up here. My parents were like this. I spent twenty years at LAPD. I sell real estate in the Santa Clarita Valley.

Do not type this. Say it.

Typing your background is a bad way to convey it. You compress. You get bored. You leave out the texture that makes it useful.

Talking gets you there faster and it gets you there richer. I am a fine typist, for the record. My high school football coach made all of us meatheads take typing, because he was dating the typing instructor and she passed us. It turned out to be one of the most valuable skills I ever picked up by accident.

And I still would rather talk.

Every one of these systems has a microphone button. Your computer has dictation built in, F5 on both Mac and PC. Your phone can transcribe. There are dedicated tools that do it better. I use Wispr Flow, and I have a mouse button mapped to it, so I click the wheel and start talking on whatever machine I am sitting at. It even catches you correcting yourself mid-sentence and fixes the text accordingly.

Loom does automatic transcription if you record video. Canva will do some. Most large language models can now take an audio file directly, or an MP4, and pull the transcript themselves.

The point is not the tool. The point is getting your actual voice, at length, into the model.

What to tell it, and what not to

I get pushback here. People say they do not want a company knowing everything about them.

Fair. So do not go down the rabbit hole of whatever ails you. This is business. You do not owe it your personal history.

Talk about your business. Your structure. Your background as it bears on your work. Twenty years at LAPD is relevant to how I do this job, because a good police officer, the kind who actually wants to help people rather than hurt them, carries something into a business like real estate that matters.

And the model will find that. It will say, given this background, have you considered representing veterans? Have you thought about focusing on law enforcement families? Teachers? Studio people, since you are in Los Angeles County?

That is the machine doing something genuinely useful, and it can only do it if you told it who you are.

Your alignment is your niche

This is the part of the seminar that landed hardest, so let me put it in writing.

There are people who will want to work with you instead of me. No matter what I do. They will like you better. They will align with you better. They will find you more approachable than Connor MacIvor.

That is not a problem. That is the entire point. Those are your people.

So do not be too concerned about what everybody else is doing. People fall in love with a personality. Be specific enough to have one.

I am aligned a certain way. I believe in God. I believe in Jesus Christ and I trust Him as my Lord and Savior. I do not think we evolved, I think we were created. I support the death penalty. Politically I land where I land.

Every one of those is a trigger for somebody. They cause people to go sideways. And I am not putting them out there to fight about them, because I do not fight about them. I put them out there because the people who are aligned with me can find me, and the people who are not can go find somebody who fits them better. Everybody wins in that arrangement.

At the end of the day, when the horses come over the hill and it is last man standing, we are all going to be back to back fighting the same demons anyway. I believe humanity aligns eventually. That is cryptic, and I know it, and I still believe it.

Put your forward-facing foot out there. People will choose you or not choose you. Both of those are useful.

The luxury market reality check

Since we are being direct, let me say the thing nobody says about luxury.

Everybody wants to be a luxury agent. Some people simply cannot pull it off, and it is worth knowing that about yourself before you spend two years on it.

The people at that level might not be you. If you can align with them, if you can learn to move in that world, maybe you fit. But nobody likes a fake, so it would have to be a genuine transformation, the kind where you fake it long enough that it stops being fake. That can work. It takes real time.

Luxury is largely a who-you-know entry point. That is the plain truth of it.

But here is the part that gives me hope for the rest of us. There is real room for the agents nobody knows, because there are very wealthy people who want to sell residential property and specifically do not want the agency that ends up on the front page talking about the experience. Everybody signs an NDA. NDAs get worked around. So some of these sellers go outside the normal channels entirely. They do not trust their producer, their scout, or their agent, and they hire their own person.

That market is open. If you are trying to sell in Nantucket, you probably need an autographed photo with the Obamas. Everywhere else, there is a door.

Meanwhile, we have agents doing extremely well serving Teamsters, electricians, nutrition companies, plumbing outfits, senior housing, and first responders. That is a plan of attack. Luxury is not the only ladder.

The spouse test

Here is how you know whether step three actually worked. It is free and it takes five minutes.

Have your AI write something real. An email to a client. A response to somebody who reached out.

Then hand it to your spouse, your significant other, a close friend, whoever knows you well. Tell them you wrote it this morning, it is about that client you mentioned, would they mind reading it and telling you what they think.

If they hand it back and say "this is fine," you are getting somewhere.

If they say "you did not write this, this is not your voice," you are not done. Go back and train it more.

And when they say "Connor, this is a great email, nice work," and you get to say that was my AI, and they say it sounds just like you? That is the whole target. That is what done looks like.

Step 4: Feed it your material

Now that it knows who you are, give it what you know.

If you are a Tom Ferry person, or a Mike Ferry person, or a Brian Buffini person, or you follow any coach at all, you have scripts. You have frameworks. You have material you paid real money for and have been refining for years.

Dump it in.

If you are one of those people who keeps good records, all the better. Years of notes on what worked and what did not in this market, with these buyers, in these neighborhoods. That is proprietary knowledge and no general model has it.

You do not even need to tell it how to use all this. Give it the material and ask it to help you build the business. It will figure out the how. Use the best model available for that particular exercise, whatever is at the top of the list that week.

AI runs on data. The reason these models have different personalities at all is the training process, and the training is the expensive part, not the data. Your data is the thing that makes yours different from the one your competitor down the hall is using.

Step 5: Your contacts and the CRM nobody has touched

Everybody has a drawer. Sometimes a shoebox. Cocktail napkins with names and phone numbers on them. Business cards from four years ago.

That is not clutter. That is the asset.

And if you already have a CRM, good, except in most cases it has not been meaningfully touched since the day it was set up. AI can make that thing come alive. It can help you sort it, segment it, figure out who has gone cold and why, and draft the outreach.

Which brings us to the guardrails, and these are not optional.

The guardrails I will keep repeating

Never let it send on its own

Do not connect it to your email and let it go.

If you have a newsletter running on a schedule, every couple of weeks, connected to your Gmail or to a system I set up for you, read it before it goes. Every send that hits a list of twenty thousand people gets read by a human being first. That human is you.

You do not have to proof every individual copy if they are identical. Read it once. Read all of it. Click the links.

Because it will tell you it is perfect and ready to send. It will ask you for the go-ahead in a very confident voice. And maybe the QR code is broken. Maybe it swapped in the wrong link. Not maliciously. It just happens.

Here is my favorite example of how these things fail. Suppose you ask it: I have a car wash three blocks from my house, I need to get my car washed, it is a beautiful day, should I walk or drive?

And it says: beautiful day, only three blocks, definitely walk. It will do you good, clear your head.

And you say, but how am I going to get my car washed?

And it pauses and says, oh, I see what you mean, you are right, you will need to take the car.

That is a machine with trillions of parameters choking on something a child would get instantly. It is going to do that occasionally with your newsletter too.

Verify every number before it faces the public

If you are about to put a statistic in front of the public, and AI handed you that statistic, push back on it before it ships.

Say you want to write about how consumers feel about plumbing companies, and it tells you thirty percent of people do not trust their plumber. I have no idea whether that number is real. Neither do you. Ask it directly: are we sure this is right? Where does this come from? Is this exactly what the source says?

I hold myself to this. In another video this week I stated a figure on public approval of AI that turned out to be close to backwards, and I corrected it in writing rather than let it stand. That is the standard. It applies to me too.

Incognito is not privacy

I would not put any weight on the people telling you that as long as you click the incognito button, or switch off training data sharing, you can put sensitive client information into these systems and not worry about it.

I would not believe that for a second.

There have been subpoenas. People have done bad things, and investigators have gone through the vendor, through the company that makes the model, and gotten the information.

The only way to be genuinely unconcerned is to run your own system, in your own house, on hardware you own. That is what open source means in this context. Those machines are getting more expensive by the day, partly because memory has become a commodity and partly because the data center buildout is eating supply. But it is the only real answer if the data is truly sensitive.

For everything else, use good judgment about what goes in.

A word on intellectual property

While we are here. Be careful with images.

I have a photo of myself as an LAPD motor cop, arresting a kid on the back of my police motorcycle. It is a spoof shot from a school event somewhere in the valley. A reporter took it. Getty bought it. Getty owns it.

I tried to use my own picture and got a very nice letter telling me to stop or be sued.

That is the intellectual property problem in miniature, and it is the same argument playing out at scale over AI training data right now. Models are trained on work that belonged to somebody. The companies also extract from each other, which is what distillation means, pulling knowledge out of a competitor's model rather than going and finding the data themselves.

Whether the lawsuits change anything, I do not know. I am not an attorney. But I know the doctrine from my old job. Fruit of the poisonous tree. If you get into the car illegally, everything you find after that is void, no matter what it is. Whether that logic ever gets applied here is above my pay grade, and it is worth watching.

Human slop beats AI slop

One more thing, and this one is about your videos, not your systems.

When AI makes your videos, they are going to be perfect. No stumbles. No lags. No moment where you lose your train of thought entirely and have to admit it on camera.

Leave your mistakes in.

I do. When I misspeak, when I lose the thread, when I say something sideways, I keep it. I do not cut it out.

Because AI slop is about to be absolutely everywhere, and people are going to start missing the human version. I think humanness becomes a commodity in a very short amount of time, and the people who sanded all of it off are going to wish they had not.

Related: your thumbnails and your hooks. A little bit tricky gets the click, sure. But if the hook you offered does not get paid off inside the video, people learn that about you fast. I try not to do that. I am not hitting home runs on every video either, but I am trying to actually teach agents something.

Start here

If you do nothing else from this article, do this in order.

1. Pick one model this week and stop shopping.
2. Pay for it.
3. Record yourself for twenty minutes talking about your business and your background, transcribe it, and feed it in.
4. Dump in your scripts, your coaching material, and your notes.
5. Bring it your contacts.

Then run the spouse test, and keep going until you pass it.

Everything from the seminar is sitting on a page with no gate on it. No email required, no form, nothing to fill out. The prompts, the PDFs, and the interview prompt that has the AI pull your background out of you rather than making you write it cold.

Go get it: santaclaritaartificialintelligence.com/srar0820

And if you want to talk through where your specific business fits, get on my calendar. Fifteen to thirty minutes, no charge, no pitch.

Common questions

Connor T. MacIvor · CalDRE #01238257 · Sync Brokerage, Inc. · DRE #02031490