QWE AI Academy If you’re an agent writing listing descriptions, you already know the bottleneck: it’s not creativity, it’s time. Showings stacked, three contracts in motion, MLS deadline tonight. ChatGPT for real estate listing descriptions can cut a 45-minute writing job to 5 minutes – but the version most tutorials sell you will get you in trouble.
Here’s what nobody puts in their headline: ChatGPT will cheerfully write “family-friendly home in a safe, walkable neighborhood near great schools” – a sentence that contains four separate Fair Housing red flags. The model has no idea. Neither does the agent who copy-pastes it.
This guide skips the prompt-template gallery. What it covers instead: which plan to use, how to brief the model so it doesn’t hallucinate, and how to put compliance into the prompt itself – not the post-hoc review.
The scenario: 12 listings this month, zero hours to spare
Picture the agent who tried AI last spring. The output sounded like every other listing on Zillow – “step into luxury,” “nestled in,” “boasts stunning views.” Worse, the description mentioned a wine cellar the property didn’t have. They went back to writing manually.
That’s the default experience. Generic prompts produce generic copy, and ChatGPT will invent features when the brief is thin – CubiCasa’s listing guide explicitly warns to check outputs for factual mistakes for exactly this reason. The fix isn’t a better prompt template. It’s a better input.
Free vs. Plus vs. Business – the honest breakdown
The wrong tier will quietly hurt your output. Here’s where things stand as of May 2026:
| Plan | Cost | What you get for listings |
|---|---|---|
| Free | $0 | 10 messages per 5 hours on GPT-5.5, then auto-downgrade to mini |
| Plus | $20/mo | 160 messages per 3 hours on GPT-5.5, plus Custom GPTs |
| Business | See OpenAI pricing page | Same as Plus + your data is not used for training by default |
Free sounds workable. It isn’t. Free accounts get 10 messages on GPT-5.5 every 5 hours, then drop to mini until the limit resets. Drafting a listing usually takes 4-6 turns. You’ll hit the cap mid-edit – and the quality drop is real. Copy gets flatter, instructions start getting ignored.
Plus at $20/month is the practical floor for daily use. 160 messages per 3 hours on GPT-5.5. But the bigger reason to upgrade isn’t volume – it’s Custom GPTs. Those let you permanently store your brand voice, your MLS’s banned-word list, and your standard brief format. Every new listing chat starts pre-loaded.
Watch out: on Free, Go, Plus, and Pro, conversations are used to train OpenAI’s models by default, with opt-out available in settings. Pasting unlisted seller motivations or pre-market property details? Toggle training off – or use Business, where it’s off by default.
The setup that prevents hallucinated wine cellars
Agents type “write a listing for a 4-bed colonial in Westport” and hit enter. ChatGPT fills the gaps with stock phrases and invented features. The fix is a structured brief – fill it out for each property before you prompt.
PROPERTY FACTS (use only these - do not invent features):
- Address: [exact]
- Beds/Baths/SqFt: [exact]
- Year built / last reno: [exact]
- Confirmed features: [bullet list - only what you've verified]
- Recent upgrades w/ dates: [bullets]
- Lot size + orientation: [exact]
CONTEXT:
- Target buyer: [first-time / downsizer / investor / etc.]
- Price tier vs. neighborhood: [at, above, below comp]
- What the listing photos emphasize: [kitchen, view, etc.]
CONSTRAINTS:
- Max 150 words for MLS public remarks
- Active voice, no clichés (list below)
- Fair Housing: describe the property, not the buyer
- Banned phrases: [paste your MLS's flagged list]“Use only these – do not invent features” matters more than it looks. Without it, the model pattern-matches to other listings it’s seen and adds plausible-but-fictional details. With it, hallucinations drop sharply. That said – always cross-check numbers in the output against what you actually typed in. Square footage, distances, school district names. If you didn’t include it, ChatGPT may have guessed.
There’s a broader question here that doesn’t get asked enough: if AI copy can convincingly describe a property it’s never seen, what does “good” listing copy actually do? Probably less about describing features and more about sequencing them – leading with the detail that makes a specific buyer stop scrolling. That part is still human judgment.
Building Fair Housing into the prompt itself
Every other tutorial skips this section. The Fair Housing Act prohibits any preference, limitation, or discrimination based on race, color, national origin, religion, sex, disability, or familial status – and that prohibition covers MLS public remarks directly.
The standard isn’t intent. Language that a reasonable person could read as expressing a preference for or against a protected class can form the basis of a complaint – regardless of what you meant. ChatGPT has no concept of this. It writes whatever sounds natural, and “natural” real estate copy is full of trip wires.
The phrases that get agents in trouble are the ones that sound fine. Per NorthstarMLS Fair Housing guidance: criteria like “family-friendly” or “great schools” aren’t objective property descriptions – describing a school district as good or bad can be considered steering. “Walking distance” is flagged too. The safer swap is exact mileage from Google Maps. The NorthstarMLS principle is to describe the property’s what, not the buyer’s who. HUD enforcement can carry fines historically up to $10,000 per violation, with each reprint counting separately.
So tell the model. Add this block to your prompt:
Paste this into your Custom GPT instructions: “You are writing MLS public remarks. Describe the property and its physical features only – never the buyer, occupant, or ideal lifestyle. Do not use: family-friendly, great schools, safe neighborhood, quiet street, walking distance, master bedroom, man-cave, she-shed, bachelor pad. Replace distance claims with exact mileage. If you’re unsure whether a phrase is compliant, flag it instead of writing it. Make sure no output implies a preference for any protected class.”
That instruction block does more for compliance than any post-hoc edit. Not foolproof – you still review every output – but it stops the obvious violations before they’re written.
The cliché problem (separate from compliance)
“Step into luxury.” “Nestled in.” “Boasts.” “smooth blend of modern and classic.” Not illegal. Just lazy – and they make every listing read like the same listing.
Build a no-no list and feed it alongside your Fair Housing block. Some real estate AI tools screen for 150+ flagged terms across both categories. You can replicate the basics with one text file, two lists: legal risks and tired phrases. Paste it into every prompt.
Where ChatGPT falls short
Three honest limits.
It can’t see the property. Photos help a little. The model can pick up architectural style in broad strokes. But the texture of original 1920s oak floors versus refinished oak? Still on you.
It writes confident wrong things. Square footage, lot size, school district names, distances to landmarks – if these aren’t in the brief, the model may guess. It won’t hedge. It’ll just be wrong.
It doesn’t know your MLS. Some MLSs auto-flag specific phrases. Some count contact info in remarks as a violation. ChatGPT has no idea which MLS you’re on. That local knowledge doesn’t transfer.
None of these are deal-breakers. They’re reasons to use it as a first-draft generator, not a final-copy generator.
What to do this week
Open a new chat. Paste the briefing template above with your next listing’s details. Add the Fair Housing block as system instructions. Generate a draft. Read it out loud – clichés stick out audibly in a way they don’t on screen. Fix the numbers, cut the remaining filler, ship it. Time it. If the first attempt takes longer than 15 minutes start to finish, the brief was too thin.
FAQ
Can ChatGPT actually understand Fair Housing law?
No. It pattern-matches to whatever instructions you give it. Skip the banned-phrase list in your prompt and it’ll write “family-friendly” and “great schools” without hesitation. Compliance comes from your prompt, not the model.
Should I just use a real estate-specific AI tool instead?
For 1-3 listings a month, a Custom GPT on Plus with your Fair Housing list baked in costs $20/month and works fine. At 10+ listings monthly, or if you’re part of a team, dedicated tools like Nila June screen output automatically against larger flagged-term databases – that automation starts to justify the per-listing fee. The break-even is somewhere around 5-8 listings monthly, though that’s a rough estimate based on typical per-listing pricing, not a published figure. Worth checking current rates directly.
Will my MLS know I used AI?
Most don’t care. What gets flagged is discriminatory language, factual errors, and prohibited contact info – not AI authorship.