Brokers are getting pitched AI constantly. AI assistants. AI agents. AI employees. Every vendor names theirs differently, and the names tell you almost nothing about what the tool actually does.
So... here's a better way to sort them.
There is AI that talks to the client for the broker. And there is AI that works in the business for the broker.
Two kinds. Once you can tell which one you're looking at, most of the evaluation gets easier.
The one question
Ask this of any AI feature: who reads or hears what it produces first, and can anyone stop it before it takes effect?
If the first reader is someone outside your business (a buyer, a seller, a prospect), it's the kind that talks to the client for you. If the first reader is you or your staff, it's the kind that works in the business for you.
Ignore the feature's name. Use the question.
AI that talks to the client for the broker
This is AI speaking in your name to someone outside your business. A buyer texts about a listing and the AI replies. A prospect calls and the AI answers.
Think of it as a front desk. Whatever it says, the person on the other end treats as you saying it. They judge your business by it.
A few things follow from that:
- Every output is a public act. Once it sends, the recipient has it.
- It needs boundaries. Someone has to decide what it can say, what it can never say, and when it hands off to a human. The quality of those decisions is most of the quality of the AI.
- It only knows what you gave it. Your website, your FAQs, your documents. Nothing else.
You judge it by how the lead responds. Time saved is secondary.
AI that works in the business for the broker
This is AI doing work for you. Summarizing a call. Drafting an email. Cleaning up records. Updating contacts. Setting up an automation. You give the instruction and you read the result.
Think of it as the back office. Nobody outside the business ever sees it.
Its output lands with you first, which means you get a chance to catch mistakes. But it usually has wide access to your data, and its mistakes can be quiet. A wrong value in a record, or a bulk change applied to the wrong set of contacts, may not show up until something downstream depends on it.
You judge it by accuracy and by whether it saved you the work.
Draft or act on its own
Both kinds can run in two modes. They can draft something for you to approve, or they can act on their own.
| Draft and approve | Acts on its own | |
|---|---|---|
| Talks to the client | You read each message before the lead does | The lead gets the AI's words directly |
| Works in the business | You confirm changes before they apply | Records and settings change without review |
Risk rises as you move toward the bottom right. The most exposed spot is the top right: AI that talks to the client, acting on its own. Errors reach someone outside your business and nothing stands in between.
Here's the part that surprises people. AI that talks to the client, set to draft and approve, behaves like back-office work until you press send. You're the only reader. The approval step is what changes the category.
Where the line blurs
Sometimes something that happens inside the business causes something to happen outside it.
Say AI that works in the business adds a tag to a contact. Your system is set up to send a document whenever that tag appears. The AI never spoke to anyone, but the contact now has a document in their inbox. The effect is the same as if the AI had written to them directly.
So... follow the chain to the end. Classify the feature by where its output finally lands, not by where it started.
Why this matters more for sellers
Buyers and sellers want different things from you, and they carry different risks.
A buyer inquiry is usually repetitive and low stakes: price, cash flow, location, how to get the CIM. A seller is often an owner who built the business over years. The sale may be the largest financial event of their life, and they usually do not want employees, customers, or competitors to know it's happening. Confidentiality is part of what they're hiring you for.
Now apply the framework. If AI that talks to the client gets something wrong with a buyer, you lose a low-value lead. If it gets something wrong with a seller, it might anchor them on a valuation before you've seen their financials, or send a message somewhere a spouse or employee can read it. That can cost you the listing.
So the same tool can be reasonable on the buyer side and a bad idea on the seller side. Ask the question about each.
One more point. Rules about contacting people attach to the kind that talks to the client. In February 2024 the FCC ruled that calls using AI-generated voices fall under the same consent rules as prerecorded calls. That is one reason to know which kind you're buying before you turn it on. (This is general information, not legal advice. Check with an attorney before you rely on it.)
What to do with this
Next time someone pitches you an AI feature, ask four things:
- Who reads or hears the output first?
- Can a person intercept it before it takes effect?
- If it changes something inside the business, does anything outside the business react to that change?
- Does it draft for approval, or act on its own?
Those four answers will tell you more than the feature name, the demo, or the price.
