AI saves real estate agents the most time on the writing around a deal: listing descriptions, lead follow-up, market summaries from comps you pull, client updates, and transaction checklists. It doesn't replace your MLS data, your judgment on price, or your broker and attorney on anything legal. This guide walks through each stage of a deal, from new listing to closing, with the prompt to use, an example of what you get back, and what to check before it goes to a client. It finishes with the two compliance areas AI makes riskier: fair housing language and edited listing photos.
Using ChatGPT? Our ChatGPT for real estate agents guide covers the setup there. This page focuses on Claude and the workflow itself.
TL;DR
- Best first wins: listing descriptions and lead follow-up. High frequency, pure writing, same-day payoff.
- You supply the facts. Give Claude the property details, comps, and dates; it writes around them. Tell it to flag anything missing instead of guessing.
- Web search helps, but isn't your MLS. Claude can look things up and cite sources, but comps, prices, and days on market should come from your MLS.
- Check every listing for fair housing. Describe the property, not who should live there. The Fair Housing Act applies to your ads whether you or an AI wrote them.
- Label edited photos. Virtually staged or AI-enhanced photos need to be disclosed under most MLS rules and the NAR Code of Ethics' "true picture" standard.
- The free Claude Cowork course teaches the overall approach.
Set up once: give Claude your business
Before any workflow, spend 15 minutes on context so the output sounds like you:
- Create a Claude Project called "My real estate business" and add your bio, your market area, your brokerage's name and required disclaimers, three listing descriptions you're proud of, and a couple of emails in your voice. Every chat in that project starts with all of it. (Claude Projects explained.)
- Make a project per active listing if you like: the property sheet, disclosures summary, and seller notes all in one place.
- Connect your tools if you want Claude to search them: Google Drive or Microsoft 365 for documents, and Gmail to search client threads and draft replies (set sending to need your approval, so nothing goes out unread). See Claude connectors.
Stage 1: New listing
Listing description
Give Claude the facts, not a vibe. The more specific the input, the less generic the output.
Write an MLS listing description, about 150 words. Facts: 3 bed, 2 bath, 1,640 sq ft ranch, built 1962. Kitchen renovated 2024 with quartz counters and gas range. New roof 2023. Fenced backyard with mature oak. Half a mile from Riverside Park and the Main Street shops. Warm but not over-the-top. Describe the property and location only; don't describe who would want to live here. Don't add any feature I didn't list. Then give me a 2-sentence Instagram version.
MLS description
Updated 1962 ranch with the work already done. The 2024 kitchen renovation brings quartz counters and a gas range, and a new roof in 2023 means one less thing to think about. Three bedrooms and two full baths across 1,640 square feet, all on one level. Out back, a fully fenced yard sits under a mature oak. Riverside Park and the Main Street shops are half a mile away.
Renovated kitchen, new roof, fenced yard under a big oak, half a mile from Main Street. Just listed.
Before you publish: every feature matches the property sheet; square footage and year match public records; no wording about who the home is "perfect for" (see fair housing); your MLS's character limit.
Listing packet
If the documents for a listing are scattered (property sheet, HOA summary, utility averages, inspection highlights), Claude Cowork can pull them into one buyer-facing packet: "Using the files in the 418 Oak St folder, build a 2-page buyer info sheet as a PDF: key features, HOA dues and rules summary, average utilities, and recent updates. Mark anything you couldn't find with [MISSING]."
Stage 2: Leads and showings
Lead follow-up sequence
Speed and consistency win leads. Draft the sequence once and reuse it.
Write a 4-message follow-up sequence for a buyer lead who asked about 418 Oak St on Zillow. Message 1: same-day text, under 50 words, offer two showing times. Message 2: day 2 email with the listing link and one detail about the neighborhood. Message 3: day 5 text with a similar listing [I'll add it]. Message 4: day 10 check-in asking what they're looking for. Friendly, not pushy. Sign as [Name].
After a showing
Here are my notes from today's showing with the Garcias: liked the kitchen, worried about the second bathroom size, asked about the school bus stop. Write a short follow-up email that addresses each point. For the bus stop, write [CONFIRM: bus stop location] instead of guessing.
Stage 3: Pricing conversations
CMA narrative from your comps
Claude is good at turning a table of comps into a story a seller can follow. It shouldn't be the source of the comps.
Here are 5 comps I pulled from the MLS: [paste address, beds/baths, sq ft, sold price, sold date, days on market, notes]. Write a one-page seller summary explaining where their home likely fits and why. Use only these numbers. Do not calculate a list price; I'll set that. Where you'd want more data, write [NEEDS DATA].
The three closest comps, all ranches within a mile, sold between $389,000 and $412,000 in the last 90 days. The two with updated kitchens sold at the top of that range and in under 15 days. Your home's 2024 kitchen and 2023 roof put it in that updated group.
The one comp that sat for 41 days had original finishes and a 1-car garage, which suggests buyers in this range are paying for updates.
Things that would sharpen this: current active listings nearby [NEEDS DATA] and any pending sales [NEEDS DATA].
With web search on, Claude can look up public information like local news or school calendar dates, with citations. It's still not your MLS: for comps, prices, and days on market, use your own data.
Market update for your sphere
Turn these monthly stats from our MLS report [paste] into a 150-word market update email for past clients. One interesting takeaway, no predictions, no jargon. End with an offer to run a free home value estimate.
Stage 4: Under contract to close
Client updates
Write a short update to my buyers: inspection is done, the seller agreed to fix the water heater, appraisal is scheduled for [DATE], and the next thing they need to do is send homeowner's insurance info to the lender by [DATE]. Calm and clear, under 120 words.
Transaction checklist
Make a checklist from contract to close for a financed buyer purchase in [state], with a column for who's responsible and a column for the deadline. Leave every deadline as [DATE from contract] for me to fill in.
Deadlines and requirements vary by state and contract, so the checklist is a structure for you to fill in from the actual contract, not a source of dates.
Explaining the process
My first-time buyers are confused about the difference between earnest money and the down payment. Explain it in 4 sentences, plain English, no jargon.
Fair housing: describe the property, not the people
The Fair Housing Act (opens in a new tab) makes it illegal to publish any ad for a home that indicates a preference or limitation based on race, color, religion, sex, disability, familial status, or national origin, and many states add more protected classes. That applies to your listing whether you or an AI wrote it.
AI can slip here because it tends to describe an imagined buyer. Watch for:
- Who it's "perfect for": "ideal for young families," "great for empty nesters," "perfect bachelor pad." Describe features instead: "three bedrooms," "single-level living."
- Neighborhood descriptions that describe people, not places. Name the park and the shops, not the residents.
- Religious or ethnic landmarks used as selling points.
Add "Describe the property and location only, not who would want to live here" to every listing prompt, then read the result with that lens before it goes out. HUD withdrew its 2024 guidance on AI in housing advertising in September 2025, but the law itself still applies. When in doubt, ask your broker.
Edited and AI-enhanced listing photos
AI photo tools make virtual staging and "enhancement" easy, and that's where agents get into trouble. The NAR Code of Ethics (opens in a new tab), Article 12, requires Realtors to "present a true picture" in their advertising, and MLSs are adding specific rules. For example, NorthstarMLS has proposed guidelines (opens in a new tab) that would require labeling virtually staged or AI-enhanced photos, include an unaltered "before" photo, and prohibit showing permanent features that don't exist.
Safe rules of thumb until you've checked your own MLS:
- Label every virtually staged or AI-enhanced photo as such.
- Include the unaltered photo too.
- Never add, remove, or change permanent features: walls, floors, windows, roof, views, landscaping.
Keep client data safe
Don't paste Social Security numbers, bank details, or full loan files into any AI tool. Summarize instead ("the buyer is pre-approved for $420,000 with a conventional loan"). If your brokerage has rules about AI tools, follow them; a Claude Team plan doesn't use your content for model training by default, which some brokerages prefer.
How to start this week
- Set up your business project (15 minutes, above).
- Write your next listing description with the prompt above.
- Build your lead follow-up sequence once and save it in the project.
- Add one stage at a time. CMA narratives and client updates are good next steps.
Frequently asked questions
How can real estate agents use AI?
Agents use AI for the writing around each stage of a deal: listing descriptions, lead follow-up sequences, CMA narratives from comps they pull, market updates, client updates, and transaction checklists. AI drafts; you supply the facts, set the price, and handle anything legal with your broker or attorney.
Is it safe to use AI in real estate?
Yes, with guardrails. Supply facts and numbers from your MLS and records, and verify everything before a client sees it. Check every listing for fair housing language, label any AI-edited or virtually staged photos, keep sensitive client data out of AI tools, and leave legal and compliance questions to your broker and attorney.
Can AI write listing descriptions?
Yes, it's one of the strongest uses. Give AI the specific property facts and your preferred tone, tell it not to add features you didn't list, and ask it to describe the property rather than who should live there. Check the result against the property sheet and for fair housing before publishing.
Can AI do a CMA?
AI can write the narrative of a CMA, explaining to a seller how their home compares to the comps, but it shouldn't pick the comps or set the price. Pull comps from your MLS, paste them in, tell Claude to use only those numbers, and make the pricing decision yourself.
Do I need to disclose AI-edited listing photos?
In most cases, yes. The NAR Code of Ethics requires a true picture in advertising, and many MLSs require virtually staged or AI-enhanced photos to be labeled, often alongside the unaltered photo. Check your own MLS's rules, and never alter permanent features of the property.
Do real estate agents need special AI software?
Usually not to start. Listing copy, follow-up, CMA narratives, and client updates all run on a general AI assistant like Claude plus your existing MLS and CRM. Real-estate-specific tools exist, but you can get the biggest time savings first without adding software.
Win back your week, starting now
Pick one stage, probably your next listing or your lead follow-up, and run it with AI this week. One real deal teaches you more than ten tool reviews.
The free Claude Cowork course covers the practical workflows with real prompts, built for operators and small businesses, not developers. It's a free download.