You've got an open house on Sunday and a listing that needs to go live tonight. The seller's notes say "lovely kitchen, great light, quiet street, perfect for a young family." You paste that into a chatbot and ask for a description. Back comes four confident paragraphs, one of which mentions a "recently renovated" bathroom nobody told it about, and one phrase that, if you read it twice, tells a certain kind of buyer they're the right kind.
Both of those are your problem now. AI prompts for real estate agents work when you separate two jobs: the model turns verified facts into readable copy, and you remain the person who verifies facts and decides what is published. The templates below are built around that split.
These are untested templates. I haven't run them against your MLS or CRM, and the worked example is labelled illustrative. Treat the first few uses as drafts to tune.
What is the one rule for AI listing copy?
Describe the property, not the people. That's the working version of a legal standard. The federal advertising rule, 24 CFR 100.75, makes it unlawful to publish an advertisement for the sale or rental of a dwelling that shows a preference, limitation or discrimination based on race, color, religion, sex, handicap, familial status or national origin. It applies to written and oral statements, and it names "words, phrases, photographs, illustrations, symbols, or forms" as the kinds of things that can suggest a home is available or unavailable to a group.
Two things follow. First, the test is about what the copy signals, so "perfect for a young family" can be a problem even though it sounds friendly. Second, the model doesn't know any of this by default. It learned from a huge amount of older listing language, and phrases like "ideal for families" are common in it. You have to tell it.
I confirmed the regulation text on the Cornell LII copy on 2026-10-08. HUD's own 2024 guidance on advertising through digital platforms is also worth reading if you buy targeted ads, but its PDF returned a 403 for me, so I'm only pointing at it, not quoting it. Check HUD's site for the current version. State and local laws often add protected classes. And once more: this post is not legal advice, so ask your broker or compliance contact about your market.
Listing description prompt: facts in, copy out
The failure that costs you the most is invention. A model asked to make a kitchen sound great will give it quartz counters. So build the prompt around a fact sheet and a hard "nothing else" rule.
You are drafting a property listing description for an MLS and a portal page.
FACT SHEET (use ONLY this; do not add features, upgrades, dates or measurements):
- Address area (no street number): [NEIGHBOURHOOD / CITY]
- Type: [HOUSE / CONDO / TOWNHOUSE], [BEDS] bed, [BATHS] bath, [SQFT] sq ft
- Lot / outdoor: [FACTS]
- Updates with years: [e.g. "roof 2021 per seller", or "none known"]
- Layout notes: [FACTS]
- Nearby, with distances I measured: [e.g. "park, 0.4 mi", "station, 12 min walk"]
- Terms: [PRICE, HOA, TAXES if appropriate]
Write:
1. A headline under 70 characters.
2. A 120-150 word description in plain, concrete language.
3. A bulleted list of 6 features, each traceable to the fact sheet.
Rules:
- Describe the home and its surroundings only. Do not describe, imply or
mention the type of person, family, age group, religion, nationality or
background suited to the home or area.
- Do not use "perfect for", "ideal for", "great for" followed by a type of
person. Say what the house has instead (for example "three bedrooms on
the upper floor", not "great for a growing family").
- No superlatives I can't prove ("best", "stunning", "must-see").
- No references to schools, safety, crime or the "feel" of the community.
- If a detail is not in the fact sheet, leave it out. Add a final section
"NOT IN FACT SHEET" listing anything I should confirm before publishing.
Run it, then do the review pass below before anything is posted. For the voice, how to prompt AI to sound human covers stripping out the generic brochure tone.
A review prompt for your own draft
Models are better at checking text than at avoiding every trap while writing it. Use a second pass, ideally in a fresh chat so it doesn't defend its own draft.
Review this listing text as a careful editor.
For each sentence that (a) states a fact not in the fact sheet, (b) suggests
a preferred or unwelcome type of buyer or tenant, or (c) describes the
people, culture or character of the area rather than its physical features,
quote the sentence, say which problem it is, and suggest a neutral rewrite.
Do not approve the text; list only concerns. If you find none, say "no
concerns found" and list what you could not judge.
FACT SHEET: [PASTE]
LISTING TEXT: [PASTE]
The "do not approve" line matters. A model asked "is this okay?" tends to say yes. This is an aid for catching obvious slips, not a compliance check, and it will miss things.
Neighbourhood summaries without steering
Buyers do want to know what the area is like. The safe and useful way to give it is through facts you can point at. The risky way is adjectives about the residents.
Write a 100-word area summary for buyers of [PROPERTY TYPE] in [NEIGHBOURHOOD].
Use only the data below. For each statement give the data point it came from.
DATA I VERIFIED:
- Walk/transit: [e.g. "Metro line X, 8 min walk"]
- Parks and open space: [NAMES, DISTANCES]
- Shops and services: [NAMES, DISTANCES]
- Commute times to [EMPLOYMENT CENTRES] at [TIME OF DAY], source: [MAPS APP]
- Recent sale price range for similar homes, source: [MLS PULL, DATE]
Exclude: opinions on schools, crime or safety; descriptions of residents,
demographics or "community feel"; religious institutions as selling points;
comparisons to other neighbourhoods based on who lives there.
Add one line telling buyers to check official school and public-safety
sources themselves.
Notice the commute line. A time and a time of day gives a buyer something to check. "Convenient location" gives them nothing.
Buyer follow-up prompts that sound like you
The honest problem with follow-ups isn't writing them. It's that after the 14th showing of the month you can't remember who wanted a garage and who is relocating in March. The model can't remember either unless your notes do.
Draft a follow-up email from me to [BUYER FIRST NAME].
My notes on them (the only facts you may use):
- Looking for: [BUDGET RANGE, AREAS, MUST-HAVES, DEAL-BREAKERS]
- Timeline: [DATE / "not stated"]
- Last contact: [DATE, WHAT WAS SAID]
- Properties they saw or asked about: [ADDRESSES + their reaction, quoted]
- Open question I owe them: [E.G. "HOA fee for Unit 4B"]
Write under 120 words. Reference one specific thing they said. Answer the
open question only if I've given you the answer below; otherwise say I'm
confirming and will reply by [DATE]. Offer one clear next step.
No pressure language ("act fast", "won't last"). No guesses about their
finances or family situation.
ANSWER TO OPEN QUESTION: [PASTE or "none"]
The "quoted reaction" detail is what stops the email sounding like a template. "You mentioned the north-facing living room felt dark" lands. "I hope you're enjoying your search" doesn't.
For text and email outreach, check that your messages follow the consent and opt-out rules for your area, and keep the unsubscribe wording in your template, not in the prompt. If you're setting up inbox routines more broadly, AI email automation prompts covers that side.
Objection handling for sellers and buyers
Real objections are specific: "Your commission is too high." "It's been on the market 30 days, should we drop the price?" "The inspection found a roof issue." Give the model the actual sentence and the true facts of the situation, and ask it to separate what to say from what to find out.
Client said: "[EXACT QUOTE]".
Context: [BUYER/SELLER, STAGE, WHAT THEY'VE TOLD ME, MY LISTING AGREEMENT TERMS
IF RELEVANT].
Facts I can stand behind: [COMPARABLE SALES WITH DATES, DAYS ON MARKET, FEEDBACK
FROM SHOWINGS, INSPECTION FINDINGS].
1. Name three possible reasons they'd say this; pick the most likely.
2. Write a 3-sentence spoken reply that uses only my facts.
3. Write one question to learn the real reason.
4. Flag anything here that is a legal, tax or contract question I should
send to the right professional instead of answering.
Item 4 keeps the model from improvising an answer about disclosure obligations or tax treatment. Those are for your broker, a lawyer or an accountant.
A worked example (illustrative)
This is an invented scenario to show the shape of the process. It isn't a transcript of a real run.
Fact sheet: three-bed townhouse, 1,450 sq ft, small fenced yard, kitchen cabinets repainted 2024 (per seller), station 12 min walk (checked on a maps app), park 0.4 mi.
Seller's note: "Cosy, perfect for a young family, quiet street, great school nearby."
What you do: you leave the seller's note out of the fact sheet except for "cabinets repainted 2024". "Perfect for a young family" is a type-of-person phrase. "Quiet street" is an opinion you haven't verified. "Great school" is a judgment you shouldn't make and the model can't verify.
Draft shape you'd want back:
Headline: Three-bed townhouse with fenced yard, 12 minutes to the station
Description: Three bedrooms and 1,450 square feet, with a small fenced yard
off the back. Kitchen cabinets were repainted in 2024 (per seller). The
station is a 12-minute walk, and a park is 0.4 miles away.
NOT IN FACT SHEET: HOA fees, roof age, parking arrangement, flooring.
It's plainer than what the seller imagined, and it's true. Show the seller why: every line can be defended, and the fair-housing phrase is gone. Then you add the extras (parking, HOA) after confirming them.
Where this goes wrong
| Failure | What it looks like | Fix |
|---|---|---|
| Invented upgrade | "Renovated bath", "new roof" that nobody told the model | Fact-sheet-only rule; the NOT IN FACT SHEET section |
| Buyer-type phrasing | "Ideal for...", "great for families" | Ban the construction in the prompt; run the review prompt |
| Community adjectives | "Quiet, tight-knit, family-oriented" | Ask for distances and named amenities instead |
| School and safety claims | The model asserts ratings from stale memory | Exclude both; send buyers to official sources |
| Wrong details in follow-ups | A budget or address from another client | New chat per client; paste only that client's notes |
| Context bleed | Details from one listing appear in the next | Fresh chat for each property |
| Photo captions | Auto-generated captions describe people or "who lives here" | Caption the room and features only |
What not to automate
- Publishing. The agent signs the listing, so the agent reads the final text. A person presses the button.
- Fair-housing judgment. The review prompt catches obvious phrases. Whether a sentence is fine in your market is a question for your broker or compliance contact.
- Valuation and pricing advice. Comparables and pricing are your professional judgment. Don't hand a CMA conclusion to a model that can't see your market.
- Disclosures, contract language and legal replies. Draft your own, or use your brokerage's forms.
- Ad targeting choices. If you buy targeted ads for housing, the audience settings need a human who knows the platform's housing-ad rules.
- Auto-sending follow-ups. Wrong first names and wrong properties are expensive. Draft, then send yourself.
- Anything using clients' personal data in tools your brokerage hasn't approved. Use placeholders for names, phone numbers and financial details until you've checked the policy.
If you want to improve the quality of what goes into these prompts, giving context explains why your fact sheet matters more than your wording, and avoiding hallucinations covers the "use only this source" pattern in more depth. Owners running a small agency may also want AI prompts for small business owners, and there are more starting points in the prompt library.
