How to Use AI for Real Estate Agents

The average agent closes nine deals a year. That is less than one a month. Median income is $59,200. In your first two years it is $8,000.

Those numbers come from NAR’s 2026 Member Profile. They matter because most AI tools sold to agents are built for a business that does not look like that. They are built for volume: hundreds of leads, long automated sequences, a team. The median real estate team closes 32 deals a year. You are not running that business, and tools designed for it will waste your money.

Here is what the job actually looks like, and where AI helps.

The numbers

	Median agent	First 2 years	16+ years	Teams
Deals closed per year	9	2	—	32
Sales volume	$2.7M	$330,000	—	$17.5M
Gross income	$59,200	$8,000	$88,500	—
Business from repeat clients	28%	0%	49%	—
Business from referrals	22%	0%	32%	—

Other figures from the same report: the typical agent is 57 years old, works 35 hours a week, and has been at their current firm for six years. Two-thirds are women.

Read the bottom two rows again. Half of a senior agent’s business comes from people they already know. A new agent gets none of it.

Problem 1: your past clients forget you

If you have been doing this a while, half your income comes from people who are not buying anything right now. They will buy in five or ten years. Until then you have no reason to call them.

So you don’t. Then they list with someone else.

Staying in touch means remembering small things. Who refinanced in March. Whose kid starts college. Who said they wanted a bigger garage. Nobody pays you for that, so it is the first thing you drop in a busy month.

What AI does here: it reads your CRM notes and writes fifteen short messages that each mention something real about that person. You read them, fix what is wrong, and send. Ten minutes.

What it does not do: write a newsletter. A newsletter is not a relationship.

If you are new, this does not apply to you yet. You have no database. Use AI to prepare for the few calls you do get instead: pull comparable sales, research the property, draft the follow-up.

Problem 2: leads sit unanswered

Someone emails about a listing while you are at a showing. By evening it is buried.

WAV Group tested this. Researchers sent fake buyer inquiries to 384 listings across 11 states through broker sites, Zillow, Realtor.com and Trulia.

What they found
Inquiries that got no answer at all	48%
Average reply time, when there was a reply	917 minutes (15 hours)
Average phone attempts before giving up	1.5
Average emails before giving up	2.08

This study is from 2013, so treat the exact numbers as old. The reason it happens has not changed.

What AI does here: it reads the inbox, writes a first reply with the property details already in it, and waits for you to approve. You send it from your phone between appointments.

Keep it as a draft. If you let it send on its own, it will eventually tell a client something untrue, in your name.

Do not let it make phone calls. On 8 February 2024 the FCC ruled that AI voices count as “artificial” under the Telephone Consumer Protection Act. That law lets people sue you directly, and 26 state attorneys general can enforce it. Companies selling “AI ISAs” demo them by cold-calling bought lists. That demo is the risky part.

Problem 3: you can no longer choose who sees your ads

This one catches people out, because it happened quietly.

In January 2023, Meta settled a fair housing case with the DOJ. It then changed how housing ads work.

Before	Now
Target by age	Not allowed
Target by gender	Not allowed
Target by ZIP code	Not allowed
Special Ad Audiences (lookalikes)	Removed
Normal delivery	Meta spreads housing ads evenly across demographic groups

Real estate advertising ran on those settings. They are gone. What you still control is the ad itself, the offer, and the page people land on.

What AI does here: writes twenty different versions of an ad and a landing page for each listing, instead of one generic ad pointing at your homepage. When you cannot pick the audience, the ad has to find them. Making twenty versions used to cost too much for a nine-deal agent. Now it does not.

One warning. Writing an ad aimed at a type of person is still illegal, even though you are no longer selecting them. “Perfect for young families.” “Great starter home for newlyweds.” “Ideal for empty nesters.” That is the behaviour the rules were written to stop.

Listing descriptions: the thing to start with

This is the easiest win. Twenty minutes of writing becomes two.

The quality depends on what you give it. Put four things in the prompt:

1. Real details. Square footage, year built, beds, baths, lot size. Say “quartz counters, white oak floors,” not “updated kitchen.” Include dates of any work done.
2. What the buyer is buying, not who they are. “Single-level layout.” “Ten minutes from the hospital.” “Garage fits a workbench.” Never describe a type of person.
3. Your MLS’s banned word list, pasted into the prompt. The model follows rules you give it. It knows nothing about rules you don’t.
4. A character limit, matching your MLS field.

Then read what comes out.

Ask any AI for a warm listing description and it will often write “great for families.” That phrase is a fair housing violation. The model is not broken. It is writing something that sounds nice, and nice is not the standard here.

OneKey MLS publishes its banned list. It is longer than most agents expect:

Category	Examples they flag
Families and children	“great for families,” “no children,” “newlyweds”
Age	“adults only,” “golden agers”
Disability	“not suitable for wheelchair”
Income source	“no Section 8,” “must be employed,” “verifiable income”
Race, religion, nationality	direct references of any kind
Other protected groups	sexual orientation, military status

OneKey tells the agent and the broker, then deletes the wording. If a complaint goes further, it is your name on the listing, not the software’s.

Photos: what you may and may not change

Virtual staging is allowed. Inventing features is not. NorthstarMLS spells out the line:

You may change	You may not change
Furniture	Walls
Décor	Flooring
Crop, exposure, colour balance	Doors and windows
Light retouching	Roof, siding, ceilings
	Fireplaces
	Driveways and grading

Two more rules from the same guidelines: label the photo as virtually staged or AI-enhanced (in the caption, the filename, or on the image), and include an unedited “before” photo of every room you staged.

Watch the tool. Tell an AI to “brighten and modernise” a kitchen and it may swap the floor without telling you. Compare against the original before you upload.

These are one MLS’s rules. Check yours.

Never use AI to pick people

Drafting and summarising: fine. Choosing between human beings: not fine.

In 2024, SafeRent Solutions paid $2.28 million to settle a class action. Its tenant-scoring software was accused of giving lower scores to Black and Hispanic applicants and to people using housing vouchers, breaking the Fair Housing Act. The lawyers called it the first case of its kind.

So do not use AI to score or rank buyers or tenants, to describe neighbourhoods, or to decide your position in a negotiation. “The software decided” has been tried as a defence and it failed.

What to use it for, at a glance

Task	Use AI?	The rule
Listing descriptions	Yes, then check	Fair housing wording
Past-client follow-up	Yes, as drafts	Read before sending
First reply to a new lead	Yes, as drafts	Read before sending
Ad copy and landing pages	Yes	No demographic targeting in the words
Summarising comps	Yes	You pick the comps
Reading contracts for odd clauses	Yes	Broker still reviews
Virtual staging	Yes, with limits	Furniture only, label it, keep a “before”
Calling leads with an AI voice	No	TCPA
Scoring tenants or buyers	No	Fair Housing Act
Describing a neighbourhood	No	Steering

Getting started

Do listing descriptions for a month. Only that. Check every one against your MLS word list before it goes live.

Keep a note of what AI helped write and how you checked it. If a complaint ever comes, someone will ask what your process was. “I check every description against the banned list” is an answer.

Ask your broker what the firm’s AI policy says. Most firms wrote one in the last two years.

Common questions

What should a solo agent use AI for first? Listing descriptions, then follow-up with past clients. At nine deals a year, the win is keeping old relationships alive, not handling more leads.

Can AI write my MLS listing? Yes. Check it against your MLS’s banned word list before publishing. The listing is yours no matter what wrote it.

Is AI virtual staging allowed? Usually, if you label it. Most MLSs limit it to furniture and décor and want an unedited photo too. Rules differ, so check yours.

Can I use an AI voice to call leads? Treat it as regulated. The FCC ruled in February 2024 that AI voices count as artificial under the TCPA. Get legal advice before you switch one on.

Can AI help with Facebook ads? With the words, not the audience. Age, gender and ZIP targeting were removed for housing ads in 2023. Use AI to write more versions.

Can AI screen tenants? No. That has already cost one company $2.28 million, and the liability lands on whoever uses the tool.