How I Use Claude (and ChatGPT) to Write 12 Listing Descriptions in 30 Minutes

Last Saturday, Priya — a solo agent in Sacramento I’ve been trading workflow tips with for a year — texted me a screenshot of her open Word doc at 11:47pm. She had nine new listings going live Monday and was 200 words into the first description. “Tell me you have a better way,” she wrote.

I do. I sent her the prompt template I’ll show you below, plus a 4-minute Loom of me running it on three of her actual MLS data sheets. She finished all nine descriptions in 42 minutes. They were better than the ones she’d been writing for 18 months. She called me Monday morning, somewhere between excited and annoyed, and said “I have wasted so much time.”

This workflow is not magic. It’s two tools (Claude and ChatGPT, both available on free or $20/mo plans), one specific prompt template, and a 90-second editing pass that catches the things AI consistently gets wrong. If you can copy-paste, you can run this.

Why this works (and why most “AI listing description” tutorials fail)

If you’ve read other guides on using ChatGPT for listing descriptions, you’ve seen prompts like “Write a listing description for a 3 bed, 2 bath home in Charlotte.” That works if your goal is to produce mediocre-sounding copy that all your competitors are also producing. It doesn’t work if your goal is a description that actually moves a listing.

Three things separate this workflow from the generic version:

1. You feed AI the bullet points, not the address. AI doesn’t know your neighborhood. You do. The prompt structure forces you to extract what’s actually special before you write a word.
2. You use two models, not one. ChatGPT for the fast draft, Claude for the editing pass. They catch different things.
3. You have a 6-item compliance checklist baked into the prompt itself. Fair Housing, MLS character limits, sensory specificity, no clichés.

That’s the entire system. Now the actual prompts.

Step 1: Extract the “interesting” bullets from your MLS sheet (3 minutes)

Before you write anything, look at your MLS data sheet and ask: “what would make a buyer pause on the listing photo grid?” Write down 5-8 bullet points. Not features. Interesting features.

Bad bullets (what most agents do):

• 3 bedrooms, 2 bathrooms
• 2-car garage
• Hardwood floors
• Updated kitchen
• Fenced backyard

These tell me nothing. Every house in a $400K bracket has these.

Good bullets (what to actually extract):

• North-facing kitchen window over the sink — best natural light in the house
• 1947 craftsman with original interior trim, restored not replaced
• Backyard backs to a creek-bed greenway — no neighbors behind
• Detached studio (heated, with sink) — Etsy seller’s dream
• Quartz counters with waterfall edge installed by current owner Aug 2024
• Walking distance to the Tuesday farmers market (3 blocks)

See the difference? The first list is data. The second list tells a story. AI cannot extract this — it’s the part of the job you keep.

For a 12-listing batch, you’ll do this for each one in a single Google Sheet or notes doc. Three columns: Property, 5-8 Interesting Bullets, Target Buyer Type (first-time, investor, downsizer, family, etc.). This sheet is the single source of input for the AI.

If you find this part painful — and most agents do at first — it’s because you’ve never been forced to articulate why your listings are interesting. The discipline of doing it makes you a better listing agent, AI or not.

Step 2: The core ChatGPT prompt (paste this exactly)

Open ChatGPT (Plus tier recommended for GPT-4 access, but the free tier works fine for this). Paste:

You are a senior real estate copywriter writing for a solo agent who closes 25 listings per year in a competitive U.S. residential market. Your tone is confident, specific, and warm — not pushy, not flowery, and never sales-cliché.

I will give you bullet points about a specific listing. Write a listing description that:

1. Opens with a single concrete sensory detail or scene — not "Welcome home" or "This stunning..."
2. Highlights 3-4 of the most interesting bullets I provided (you choose which based on what would resonate with the target buyer I name)
3. Is 180-240 words for MLS-friendly length (Bright MLS, CRMLS) OR 100-130 words if I specify "short version"
4. Avoids these protected-class phrases: "family-friendly," "great for kids," "walking distance to good schools," "safe neighborhood," "quiet community" (any community/neighborhood judgment), "cozy" (often a euphemism that draws fair housing scrutiny when paired with size descriptors)
5. Uses sensory and concrete details over adjectives ("morning light hits the kitchen at 7am" beats "bright kitchen")
6. Closes with a forward-looking line that invites action without being salesy

Wait for my bullets, then ask me one clarifying question before writing. After I answer, write the description. Then below it, list any phrases you almost used but caught yourself on — so I can spot patterns over time.

Hit enter. Wait for ChatGPT’s clarifying question (usually about tone or buyer type). Answer. Then paste your bullet points + target buyer type.

You’ll get a draft. It will be 80-90% there.

Step 3: The Claude edit pass (paste this exactly)

Open Claude (claude.ai — free or Pro). Paste the description ChatGPT generated, then this prompt:

You are a Fair Housing compliance reviewer and senior real estate editor. Below is a listing description draft. Do three things:

1. Flag any phrase that could be interpreted as steering toward or away from a protected class (race, religion, national origin, sex, familial status, disability, age). Be conservative — when in doubt, flag.

2. Identify any cliché phrases that signal AI-generated copy ("welcome home," "stunning," "perfect for entertaining," "must see," "won't last long," "your dream home awaits," "boasts," "nestled").

3. Suggest specific replacements that preserve meaning but read like a human wrote them. Show before/after for each.

Then output a clean, final version with all changes applied. Do not change the structure or length — only the flagged phrases.

Claude will catch 2-4 things per description. Sometimes ChatGPT was already clean. Sometimes Claude flags a “boasts a chef’s kitchen” or a “perfect for growing families” that slipped through.

This step takes 60 seconds per listing. It is the difference between competent and excellent.

Step 4: The 90-second human review (don’t skip this)

The output from Claude is now publish-ready 90% of the time. The remaining 10% is on you:

• Does it reference any feature that isn’t in the property? AI sometimes hallucinates. I once had ChatGPT add “imported Italian tile” to a listing that had vinyl flooring. The seller would not have been pleased.
• Does the address and city match? AI defaults to plausible locations if you didn’t specify.
• Does it pass the “would I be embarrassed if my broker saw this?” test?
• Does the closing line work for your specific brand voice? I always rewrite the last sentence in my own voice — takes 15 seconds, makes the description feel like mine.

That’s it. That’s the workflow.

Doing 12 in 30 minutes: the batch process

The headline is true, but here’s how the time actually breaks down for a batch of 12 listings:

• Pre-work (do this once, not per listing): Build the bullet-points spreadsheet for all 12 listings. This takes 8-12 minutes if you have your MLS sheets open. Time: 10 min.
• ChatGPT drafts: Run all 12 through the ChatGPT prompt sequentially. Each takes 60-90 seconds (the AI is doing the work, you’re copy-pasting). Time: 14 min.
• Claude edits: Run all 12 through Claude. Faster because you’re just pasting and reading. Time: 12 min.
• Human review: 90 seconds × 12 = 18 minutes. This is where the time floor lives — you cannot skip it.

Total: 54 minutes for 12 listings if you’re new to this. After running it 8-10 times, the spreadsheet pre-work and review get faster. I clock in at 28-32 minutes for 12 now. Priya is at 35 minutes after two weeks of practice.

For comparison: she was averaging 18 minutes per description writing them by hand, which is 216 minutes (3.6 hours) for 12 listings. The math is uncomfortable.

When to use specialized tools instead

ChatGPT + Claude is the lowest-cost workflow ($40/month combined Pro tiers, often $0 if you use free tiers). It’s not always the right call.

Use Homesage.ai instead if:

• You do 8+ listings/month consistently
• You want listing descriptions, virtual staging, social posts, and CMAs in one tool
• You want Fair Housing guardrails baked in so you can skip the Claude pass
• You sometimes hand listing prep to a VA who shouldn’t be using your personal AI accounts

Use ListAssist instead if:

• You want the AI to teach you why it edited (good for newer agents)
• You want a polished output without dual-tool workflow
• You write your own first drafts and want AI as an editor, not a generator

For everyone else: Claude + ChatGPT is plenty.

Three prompts that handle 80% of listing variations

The core prompt above works for standard listings. Here are three variations I use for specific situations.

The luxury listing prompt addition

For $1M+ properties, append this to the core prompt:

This is a luxury listing ($1M+). Adjust tone:
- More restrained, less exuberant
- Specific architectural and finish-level vocabulary (e.g., "honed marble" not "fancy stone")
- Reference period/architectural style if applicable
- No "luxury" or "luxurious" as adjectives — show, don't tell
- Reading level: educated buyer who travels and has design opinions

The flip / new construction prompt

For builders and flippers, this version:

This property has been fully renovated/newly constructed. Highlight:
- 2-3 specific finishes by name and brand (Bosch appliances, Hubbardton Forge lighting, etc.)
- Permitted work (especially structural, electrical, plumbing) if I provide it
- Warranty / builder reputation if applicable
- Move-in ready language without sounding generic

The “as-is” / investor-targeted prompt

For homes priced for the investor or contractor market:

This is an as-is / investor-targeted listing. Adjust:
- Honest about condition without being negative
- Highlight bones, structure, lot, location upside
- Use language like "ready for your vision" sparingly — better to be specific ("priced to allow for kitchen and bath updates while still leaving comp room")
- Target buyer: experienced investor, flipper, or owner-occupant willing to renovate

What I do NOT use AI for in listings

A short list, important:

• Square footage, lot size, year built, tax data — never trust AI to remember these. Always pull from MLS or county records.
• HOA fees and rules — get them in writing.
• Disclosures and material facts — known issues, repairs, defects. Legal liability lives here.
• Photo captions about virtual staging — use the exact disclosure language your MLS requires; don’t paraphrase.
• Open house dates, listing prices, contingencies — anything that could be cited in a complaint.

Use AI for the prose. Verify everything else against the source.

The compounding gain you don’t see at first

Here’s what surprised me four months in: my listing descriptions stopped sounding like anyone else’s listing descriptions. Because the bullet-extraction step forces me to think about each property like a journalist would — what’s the story here — my brand voice sharpened. Buyers and other agents now tell me they can spot one of my listings without seeing my name.

That’s not what AI did. That’s what AI freed me to do, by handling the part of the writing that’s just typing.

If you want a deeper library of prompts beyond listings — buyer follow-ups, open house scripts, FSBO conversion, expired-listing outreach — the 15 ChatGPT Prompts for Real Estate Agents piece on this site is the next thing to read.

Frequently asked questions

• Can I just paste my MLS data sheet into ChatGPT and get a usable description?

  Yes, but the output will be 70% of the way there and read like every other AI listing in your market. The two extra steps below — the seed-prompt approach and the editing pass — are what turn 70% into 95% and make the description sound like it came from you, not a template.

• Why Claude and ChatGPT for this — why not just one?
• What about specialized real-estate AI tools — aren't those better?
• Is AI-written listing copy compliant with Fair Housing?
• How long does this actually take after I've done it a few times?
• Will MLSs flag AI-generated descriptions?