Using Claude to write listing descriptions — the prompt that actually works

Claude can write a good listing description in under 60 seconds. The problem is that "Claude's default" and "good listing copy" are not the same thing until you structure your prompt correctly. The pattern that works is: voice sample first, specificity constraints second, explicit exclusion list third. Everything else follows from that.

Why your first Claude listing prompt probably didn't work

Most realtors start with something like: "Write a listing description for a 2-bedroom condo in downtown Toronto with a rooftop terrace and updated kitchen."

Claude does what it's told. You get copy that technically describes the property. It also sounds like every other listing on MLS.

The model doesn't know your voice. It doesn't know your market's personality. It defaults to a register that's vaguely "real estate agent on the internet circa 2019," and that register is full of words that signal "AI wrote this" even to buyers who've never thought about AI.

The fix isn't a better tool. It's a better prompt structure.

The voice sample — put it first, every time

Before you describe the property, paste in two or three paragraphs of your own writing. It can be from a previous listing you liked, a client email you sent, a social caption. The content doesn't need to be about real estate specifically. What matters is that Claude can pattern-match your syntax, your sentence length, and your vocabulary before it starts generating.

The prompt looks like this:

Here is a sample of how I write. Match this voice, tone, and sentence structure in everything you produce in this conversation.

[YOUR SAMPLE TEXT]

Now write a listing description for the following property. Do not deviate from the voice above.

Without that voice sample, Claude writes in its own voice. With it, the output is noticeably more distinctive. This is the single biggest lever in the prompt, and most realtors skip it entirely.

Specificity constraints — tell Claude exactly what to work with

Generic prompts produce generic copy. The more specific and factual your input, the better your output.

For each listing, I recommend writing a brief "property brief" before the actual prompt. It might include:

• Exact square footage and layout (2 bed, 2 bath, den, 847 sq ft)
• Building name and street intersection
• Specific finishes that matter (integrated appliances, wide-plank engineered hardwood, soaker tub)
• Any unusual or genuinely notable feature (a terrace with unobstructed lake views, a unit that hasn't been listed since 2011, a full-sized laundry room in-suite)
• One or two facts about the immediate neighbourhood (walkable to Kensington Market, direct subway access at Spadina, directly across from Trinity Bellwoods)

You supply the facts. Claude shapes the prose. This is not optional. If you ask Claude to "describe the neighbourhood," it will hallucinate plausible-sounding context that may be inaccurate or outdated. Provide the facts you know to be true. Claude does not know your building.

The exclusion list — say what you don't want

This is the part most prompts are missing. Include this line, or a version of it, in every listing prompt:

Do not use any of the following words or phrases: stunning, nestled, boasts, offers, features, spacious, rare find, sought-after, perfect for, don't miss, impressive, meticulously, thoughtfully designed, or any phrase that reads like MLS boilerplate.

Claude responds well to explicit constraints. The exclusion list is not paranoia; it's precision. These words aren't forbidden because they're wrong. They're forbidden because every other listing uses them, and your copy loses personality every time one of them appears.

If you want to go further, add: "Every sentence should contain a specific detail. No sentence should make a claim that couldn't be verified by someone standing in the unit."

Local market context — what you can and can't ask Claude to do

You can ask Claude to incorporate neighbourhood context you provide. You cannot ask Claude to generate neighbourhood context on its own and expect it to be accurate.

The safe version: "The unit is a 4-minute walk to Ossington station. Reference this in the description naturally."

The risky version: "Describe the neighbourhood around Ossington and Queen." Claude will produce something that sounds authoritative and may contain details that are out of date, inaccurate, or simply invented.

Local market context comes from you. Transit access, walk scores, named parks, named streets, proximity to specific amenities you've personally verified. That's the material Claude works with. It's not a research tool for local facts; it's a writing tool that shapes facts you already have.

What to do with the first draft

Run the prompt. Read the output. You're looking for three things:

1. Voice match. Does it sound like you, or like a generic agent?
2. Factual accuracy. Every claim should be verifiable. "Hardwood floors throughout" needs to be true throughout.
3. MLS compliance. Your brokerage likely has rules about what can and can't appear in a listing description. AI doesn't know those rules. You do.

Edit from there. The goal is a draft that's 70-80% of the way to publish-ready in under two minutes, not a draft you can blindly post. Listing descriptions carry legal and marketing weight. They need a human pass before they go live.

The pattern I see in realtors who get the most value from Claude on listing copy: they've built a standard prompt template, they paste in a fresh voice sample once a month to refresh it, and they spend two minutes editing rather than 20 minutes writing from scratch. That's the workflow. It's not dramatic, but it compounds.

The full prompt template, assembled

Here's what the complete prompt looks like, assembled:

Here is a sample of how I write. Match this voice, tone, and sentence structure in everything you produce.

[2-3 paragraphs of your own writing]

Now write a listing description for the following property. Use only the facts I provide. Do not add neighbourhood context I haven't given you.

Property details: [Your property brief — unit size, layout, finishes, address, notable features, one or two specific neighbourhood facts you've verified]

Do not use any of the following words or phrases: stunning, nestled, boasts, offers, features, spacious, rare find, sought-after, perfect for, don't miss, impressive, meticulously, thoughtfully designed.

Keep the description under 200 words. Every sentence should contain a specific detail.

That's the whole structure. Voice sample, specific facts, explicit exclusions, word count, specificity rule. Run it once, see what comes back, adjust the constraints if needed. Most realtors need one or two rounds of iteration to get a prompt template that works reliably for their voice and their market.

---

FAQ

Can Claude write real estate listing descriptions? Yes. Claude produces listing copy with less MLS-cliche language out of the box compared to many other AI tools. The quality depends heavily on what you feed it. A prompt with a voice sample, specific property details, and explicit word exclusions produces copy that sounds like a human agent wrote it.

What words should I tell Claude to avoid in listing descriptions? Tell Claude to avoid: stunning, nestled, boasts, offers, features, spacious, rare find, sought-after, perfect for, and don't miss. These are the words that make MLS copy feel generic. Exclude them explicitly in your prompt.

What is a voice sample in a Claude prompt and why does it matter? A voice sample is a paragraph or two of copy you've already written and approved, pasted into the prompt before your instructions. It tells Claude how you phrase things, how long your sentences run, and what tone you use. Without it, Claude defaults to a generic real estate register that sounds like no one in particular.

Should I use Claude or ChatGPT for listing descriptions? Either works. Claude tends to avoid MLS-cliche language without needing as many guardrails. ChatGPT can match that quality with a tighter prompt. The model matters less than the structure of what you give it. A voice sample plus specificity constraints plus explicit exclusions will produce better output than either tool's defaults.

How do I add local market context to AI listing copy without fabricating facts? Include only verifiable, public information you already know to be true. Walk scores, transit lines, named streets, named parks or landmarks. Don't ask Claude to generate market context on its own. It will produce plausible-sounding details that may be wrong. You supply the facts; Claude shapes the prose.

Do I need to edit AI-generated listing descriptions before publishing? Always. AI produces a strong first draft. You still need to verify every factual claim, confirm the tone fits your brokerage's compliance requirements, and make sure the copy reflects what a buyer would actually experience at the property. Listing descriptions carry legal and marketing weight. They need a human review before they go live.

---

Emma Pace — strategic marketing consultant, AI coach for realtors, keynote speaker. Realtor at Monstera Real Estate. Builds AI-operated marketing systems at emmapace.ca.