Friday, 10:40 p.m. A couple scrolling Zillow in bed taps "Contact agent" on your three-bedroom listing on Maple Court. They type: "Is this still available? Can we see it Saturday?" You are at your kid's game, then asleep. At 8:15 a.m. you reply. By then they have already booked a showing with the agent who answered at 10:43. Nobody did anything wrong. You just lost to speed.
That is where AI for real estate earns its money, and it is not where most people start. Most agents first try AI on listing descriptions. That is useful, but it saves you twenty minutes. Answering a portal lead in under a minute, asking the right three questions and following up with the people who went quiet is what moves closings. This guide goes through the work of a residential agent step by step: what I would hand to AI, which tool, what it costs, and where it gets you in trouble with Fair Housing rules or your MLS.
AI for real estate agents: the map of your week, task by task
I built this table around a solo agent or a small team doing 15 to 40 transactions a year, mostly resale, some rentals. Times are rough estimates from agents doing this work every week, not a study.
| Task | Time without AI | What AI does | What stays with you | What I would use |
|---|---|---|---|---|
| First reply to portal and website leads | Whenever you see it | Answers in seconds with real listing data and asks 2 or 3 questions | The showing, the relationship | An AI agent over SMS and website chat (see the agent guide) |
| Lead qualification | 10 to 15 min per lead | Pre-approval status, timeline, areas, must-haves | Reading the person, deciding where to spend time | Prompts in ChatGPT or Claude, or the agent |
| Follow-up with leads who went quiet | Usually never done | Writes a 4-message sequence that sounds like you | Picking who gets a phone call | Your CRM plus AI-written messages |
| MLS description and Zillow listing text | 30 to 45 min | Draft from your notes, within the character limit | Every fact, every number, the Fair Housing check | ChatGPT or Claude |
| Social posts and short video scripts | 1 to 2 h a week | Captions, hooks, a 30-day calendar | Filming, your face, your voice | ChatGPT plus Canva |
| Listing presentation and CMA write-up | 1 to 2 h | Turns your comps into a clear story for the seller | Choosing the comps and the price | Claude with your CMA export |
| Objection replies | Minutes, but under pressure | Three ways to answer "the price is too high" | Tone, timing, the actual negotiation | Any chat assistant |
| Explaining contracts, inspection reports, closing costs | 30 to 60 min | Plain-English summary for the client | Anything that sounds like legal or tax advice | Claude with the PDF (redacted) |
| Valuation and pricing | 1 to 3 h | Almost nothing you can trust | Everything | Your MLS data and judgment |
Look at the last row. AI does not know what the house on Maple Court is worth. It does not see your MLS, it does not know the seller replaced the roof, and if you ask, it will still give you a confident number. Pricing stays with you.
Listing descriptions: fast, but locked to your notes
This is where everyone starts, so start right. The rule is simple: the AI may only use facts you give it. If you paste "3 bed, 2 bath, about 1,600 sq ft, updated kitchen," a careless prompt will happily add "quartz countertops," "hardwood throughout" and "steps from top-rated schools." Two of those may be false, and the third is the kind of claim you do not want in your MLS remarks.
The prompt I would use:
Write an MLS public remarks description for this listing. Use ONLY the facts below. If a detail is not listed, do not mention it. Do not describe who the home is "perfect for." Describe the property, not the buyer. No claims about schools, safety or the neighborhood's residents. Maximum [CHARACTER LIMIT OF MY MLS] characters, including spaces. Then give me a second, shorter version for Zillow and social, under 400 characters. Facts: [ADDRESS AREA, BEDS, BATHS, SQ FT AS MEASURED, LOT, YEAR BUILT, UPDATES WITH YEARS, HOA FEE, GARAGE, WHAT IS NEARBY BY NAME AND DISTANCE]
I ran it with a made-up listing (3 bed, 2 bath, 1,580 sq ft per county records, kitchen redone in 2023, new HVAC 2024, fenced backyard, one-car garage, 0.4 miles to Greenway Park, no HOA). Real output, first version:
Updated 3-bedroom, 2-bath ranch with 1,580 sq ft per county records. The kitchen was fully redone in 2023 with new cabinets and appliances, and the HVAC system was replaced in 2024. Open living and dining area, a primary suite with its own bath, and a fenced backyard with room for a garden. One-car attached garage. No HOA. Greenway Park is 0.4 miles away. Schedule your showing today.
Short, accurate, nothing invented. It is not poetry. MLS remarks do not need poetry, they need to be true and to make a buyer want to see the photos.
Two more checks before you publish. First, the square footage source: say "per county records" or "per appraisal" if that is what you have, because square footage disputes are one of the classic complaints against agents. Second, read it once with Fair Housing eyes, which I cover below.
The 10:40 p.m. lead: speed to lead with AI
Most portal leads in the US come by email notification, text or the portal's app, and people contact several agents for the same home. The one who answers first, with a real answer, usually gets the showing. You cannot answer at 10:40 p.m. every night. An AI agent can.
What a good one does: reads the message, looks up the listing in your sheet or your IDX feed, answers with real facts ("Yes, it's active. HOA is US$ 0, the garage fits one car"), asks two or three questions (pre-approved? timeline? selling a home first?) and offers two showing times from your calendar. Then it sends you a summary: hot, warm or cold.
The step-by-step build, the instructions, a real test conversation and the monthly cost (around US$ 40 to US$ 120) are in ChatGPT for real estate: the lead response agent. Short version: below about 30 leads a month, you do not need it. A good set of saved replies and an alert on your phone will do.
One thing that is different in the US: texting. To text leads from a regular 10-digit business number through a provider like Twilio, you need A2P 10DLC registration, or carriers filter your messages. And under the TCPA, automated texts to people who did not give consent are a real legal risk. Replying to someone who just texted or submitted a form asking about a home is a different situation from blasting an old list. Keep it that way.
Qualification without sounding like a mortgage application
Asking "Are you pre-approved? What's your budget? When are you moving? Do you need to sell first?" in one text feels like a form. One question at a time, mixed with something useful, gets answers. Ask AI to write your qualification sequence with that rule:
I'm a buyer's agent in [CITY]. Write a 4-message text sequence to qualify a buyer lead who asked about [LISTING]. One question per message, each message under 220 characters. Order: timeline, financing (pre-approved or not, without asking the amount), whether they need to sell a home first, must-haves. Each message gives something useful before asking (a fact about the home, a tip). Friendly, not pushy. No emojis.
The financing question matters. You want to know whether they have talked to a lender, not their income. If they have not, your move is to connect them with one or two lenders, not to estimate what they can afford.
Another US detail: since the NAR settlement changes in August 2024, MLS participants working with a buyer need a written buyer agreement before touring a home. Your qualification sequence is a good place to mention that early, in plain words, so it does not feel like a surprise at the front door.
Follow-up with the leads who disappeared
This is the most neglected money in real estate. A lead who asked about a home in March and went quiet might buy in September. They will buy with whoever is still in their texts or inbox. AI is good at writing a sequence that does not sound like a drip campaign:
- Day 2: one useful fact about the home they asked about (price change, open house, a similar home just listed).
- Day 7: a question they can answer in one word ("Still looking in Northside, or did plans change?").
- Day 21: something about the market in their area, two lines, no chart.
- Day 45: the permission close ("Want me to stop sending homes, or keep an eye out?").
The fourth message gets the most replies. People answer "keep looking" or "we bought already," and both are useful. Ready-to-copy versions of these are in AI prompts for real estate agents.
Objections: "the price is too high," "we'll think about it," "my wife needs to see it"
AI will not negotiate for you, but it is a good sparring partner. Paste the situation and ask for three different replies: one that uses data, one that asks a question, one that offers a next step. Pick the one that fits the person. Practice them out loud before a listing appointment, and you will notice you stop freezing on the same three objections.
For sellers, the useful version is: paste your CMA summary (address removed) and ask AI to explain, in plain English, why the list price you recommend is where it is. AI is good at turning comps into a story a seller follows. It is bad at choosing the comps, so do that part yourself.
AI for a brokerage: what changes with 10 agents instead of 1
For a solo agent, AI is a personal assistant. For a brokerage or team, it becomes shared infrastructure, and three things change:
- Lead routing. The AI agent answers first and then routes the hot lead to the agent on duty, with the summary. That ends the "who gets this lead?" argument and the lead that sat unclaimed for four hours.
- One voice, many agents. A shared prompt library (listing remarks, follow-ups, objection replies) keeps quality even across agents. Put the Fair Housing rules inside every prompt so nobody forgets them.
- Compliance review. The broker is responsible for advertising. A simple routine where AI-written listing text gets a second read before it goes live costs little and avoids the expensive kind of complaint.
AI tools for real estate agents and prices in 2026
Prices from the official sites as of late 2026. They change, so confirm before you subscribe.
| Tool | What I use it for | Price |
|---|---|---|
| ChatGPT Plus | Listing text, replies, social posts, images | US$ 20/month (free plan exists, with limits) |
| Claude Pro | Long documents: inspection reports, HOA docs, contracts | US$ 20/month (free plan exists) |
| Google Gemini (Google AI Pro) | If you live in Gmail and Google Drive | about US$ 20/month, check the site |
| Canva Pro | Just-listed graphics, flyers, social templates | about US$ 15/month, check the site |
| Your CRM (Follow Up Boss, Lofty, kvCORE and similar) | Many now include AI text features; check before paying twice | varies, check the site |
| n8n or Make + AI API + Twilio | The automatic lead response agent | about US$ 40 to US$ 120/month in total |
| Virtual staging apps | Staging empty rooms in photos | per photo or monthly, check the site |
My pick for most agents: one chat assistant (ChatGPT or Claude, not both at first) plus the CRM you already pay for. Add the automatic agent only when you have the lead volume to justify it.
Where AI gets real estate wrong
- It invents features. Granite you do not have, "renovated" bathrooms that were only painted, a pool that is actually a community pool.
- It invents numbers. Property taxes, HOA fees, school ratings, commute times. Never publish a number you did not supply.
- It describes people instead of property. "Perfect for a young family," "ideal for empty nesters," "quiet, exclusive neighborhood." This is where Fair Housing problems start.
- It does not know your local rules. Your state's disclosure forms, your MLS's photo rules, your local transfer taxes. Ask it to explain a concept, not to tell you what applies in your county.
- It sounds like everyone else. "Welcome home to this stunning gem" is in a thousand listings. Ask for plain, specific sentences and cut the adjectives.
Common mistakes when agents start with AI
- Pasting client documents without redacting. Pre-approval letters, bank statements and Social Security numbers do not belong in a chat tool. Remove names and numbers first, and turn off training on your data in settings.
- Publishing the first draft. Read every AI-written listing against your notes, line by line, before it hits the MLS.
- Using AI photos without saying so. Virtually staged or edited photos generally need to be labeled; many MLSs have specific rules. Check yours, and keep the unstaged photo in the set.
- Automating texts to old lists. Replying to an inquiry is one thing. Texting 2,000 old contacts with an automated tool without consent is how agents end up in TCPA trouble.
- Letting the bot negotiate. An AI agent answers questions and books showings. Price, terms and anything that sounds like advice go to you.
Fair Housing, NAR and MLS rules apply to AI text too
No rule says "AI-written ads are exempt." The Fair Housing Act prohibits advertising that indicates a preference, limitation or discrimination based on race, color, religion, sex, disability, familial status or national origin, and many states and cities add more protected classes. HUD enforces it, and HUD has published guidance on how it applies to advertising on digital platforms, including ads delivered by algorithms. The words come out of AI, but your name is on the listing.
The practical fix is to put the rule inside the prompt: "Describe the property, not the buyer. Do not mention who the home is ideal for. No references to schools' quality, safety, churches, or the people who live in the area." Then read the result looking for those things.
If you are a REALTOR®, the NAR Code of Ethics also requires that your advertising gives a true picture, and your MLS has its own rules on accuracy, photos and what can go in public remarks. And state license law usually requires your name and brokerage on advertising, so add them to your social captions too.
Where to start this week
- Pick one assistant (ChatGPT or Claude) and pay for one month.
- Rewrite your three active listings with the "only these facts" prompt and compare with what is live.
- Write your saved replies for "Is this still available?", "Can I see it Saturday?" and "What's the HOA?" with AI, and load them in your phone or CRM.
- Write a 4-message follow-up sequence and send it to ten leads who went quiet in the last 90 days.
- Count your leads for a month. Over 30, read the lead response agent guide. Under 30, keep the manual routine and the prompt library.
AI for real estate does not replace the part where you walk a buyer through a home and notice they went quiet in the kitchen. It replaces the 10:40 p.m. reply you were never going to send in time.
Want to skip this part?
Ready-made AI agents for real estate agents: nothing to build, no prompts to write, they already work for you.
- Answers Zillow, text and website leads in seconds, even at 10 p.m. on a Friday, with your listing data
- Asks about timeline, lender and home to sell, and offers showing times from your calendar
- Follows up with leads who went quiet and hands you only the summary of the hot ones
Frequently asked questions
What is the best AI for real estate agents?
For most agents, ChatGPT or Claude at US$ 20 a month covers listing text, replies and follow-up. Claude handles long documents like inspection reports better. Check whether your CRM already includes AI features before adding another subscription.
Can AI write my MLS listing description?
It can write the draft, but only from facts you supply, and you must check every line. AI tends to invent features and numbers, and a false square footage or upgrade claim in the MLS is your responsibility.
Does the Fair Housing Act apply to AI-written listings?
Yes. The law looks at what the ad says, not who wrote it. Tell the AI to describe the property, not the buyer, and read the text for phrases about who the home is ideal for, schools' quality or the neighborhood's residents.
How much does AI cost for a real estate agent?
Starting costs about US$ 20 a month for one chat assistant. An automatic lead response agent with n8n or Make, an AI API and Twilio runs about US$ 40 to US$ 120 a month depending on volume.
Will AI replace real estate agents?
Not the part buyers and sellers pay for: pricing judgment, showings, negotiation and holding a deal together through inspection and appraisal. It replaces slow replies and repetitive writing.
Can I paste client documents into ChatGPT?
Remove names, account numbers and Social Security numbers first, and turn off training on your data in settings. Pre-approval letters and bank statements should not go into a chat tool as they are.
