Maria Delgado used to spend Sunday nights writing MLS descriptions until midnight. Now she finishes 20 of them before her second espresso — and her listings get an average of 38% more saves on Zillow, according to her brokerage's internal dashboard.

Delgado, a mid-tier agent with Coldwell Banker in Phoenix, isn't a prompt engineer. She's a 41-year-old former teacher who started using ChatGPT in late 2023 because her teenage son showed her how. Three years later, her AI stack is the quiet engine behind a business that closed $42 million in volume last year.

This is exactly how she does it.

The Stack: Three Tools, $66 a Month

Delgado's setup is deliberately boring. No custom GPTs, no Zapier spaghetti, no Make.com diagrams taped to her monitor. Just three subscriptions and a Google Drive folder.

ToolMonthly CostJob in the Workflow
ChatGPT Plus$20First-draft listing copy and headline variations
Claude Pro$20Polish, tone matching, compliance check
Listing AI by Restb.ai~$26Auto-tags property photos and feeds visual data into prompts

That's it. No Jasper, no Copy.ai, no $300/month real estate "platform." She tried them. They wrote like a 2019 SEO blog.

Monday Morning: The 40-Minute Sprint

Every Monday between 7:00 and 7:45 a.m., Delgado batches her week's listings. By batching, she keeps context loaded and avoids the dead time of switching properties one by one.

  1. Photo intake (5 min). She drops 20–40 photos per property into Listing AI, which returns a structured JSON file: "stainless appliances, quartz counters, south-facing yard, mature mesquite, no pool." Cheap, accurate, and faster than typing notes from a walkthrough.
  2. Master prompt (2 min/listing). She pastes the JSON, the MLS data sheet, and the neighborhood into a saved ChatGPT prompt that specifies tone, word count (180–220), Fair Housing compliance, and three required elements: a sensory hook, a lifestyle line, and a logistical close.
  3. Draft generation (30 sec). GPT-4o returns three versions. She picks the strongest opener from one and the strongest closer from another.
  4. Claude pass (1 min). She drops the merged draft into Claude Sonnet 4.5 with one instruction: "Tighten. Remove anything a buyer would skim. Flag any Fair Housing risks." Claude is noticeably better at this than GPT in her experience.
  5. Human pass (1 min). She reads it out loud. If it sounds like a robot trying to sell a house, she rewrites one sentence by hand. Usually it doesn't.

Twenty listings × roughly 2 minutes each = 40 minutes. Add the photo intake and she's done before 8 a.m.

Pro tip: Save your "house style" as a Claude Project or a ChatGPT Custom Instruction. Delgado's includes three of her best past listings and the line: "Never use the words luxurious, stunning, or must-see." That single constraint kills 80% of AI tells.

The Prompt That Does the Heavy Lifting

Delgado shared the skeleton she's iterated on for two years. It's deceptively simple:

"You are writing an MLS description for a [bedrooms]/[baths] home in [neighborhood], [city]. Buyer persona: [first-time / move-up / downsizer / investor]. Use these verified features only: [paste JSON]. Write 180–220 words in three short paragraphs. Paragraph 1: a sensory hook about arrival or first impression. Paragraph 2: lifestyle — how someone actually lives here on a Tuesday night. Paragraph 3: logistics — schools, commute, HOA. Fair Housing compliance required: no references to family makeup, religion, or demographics. Avoid clichés: stunning, luxurious, must-see, dream home, oasis."

The "Tuesday night" line is the secret. It forces specificity. AI defaults to weekend-brochure language; weekday framing produces copy that sounds observed rather than marketed.

What Actually Improved — and What Didn't

Time savings are real: roughly 5 hours a week reclaimed. But Delgado is clear that AI didn't make her a better writer. It made her a faster editor with a reliable first draft.

Conversion lifts are harder to attribute cleanly. Her listings get more saves and showing requests than the office average, but she also shoots better photos and prices sharper than most of her colleagues. AI is one variable in a stack.

What didn't work: fully automated posting. She tested a workflow that pushed AI copy straight to the MLS via her broker's API. Within two weeks she had a compliance flag (a phrase Claude let slip about a "quiet street perfect for families"). Now every listing gets human eyes before it goes live. Non-negotiable.

Pro tip: Run every AI-generated listing through a Fair Housing keyword check. The NAR maintains a free list of flagged phrases. A 10-second scan can prevent a five-figure fine.

FAQ

Does the MLS allow AI-written descriptions?

Yes, in every U.S. market as of mid-2026. The agent remains legally responsible for accuracy

Written by

Founder & AI Automation Researcher

Mahendra Bugaliya is the founder of AI Profit Automation. He tests AI tools and automation workflows hands-on and writes practical, no-hype guides on using them to build and grow online income.

Tags
ai for real estate listing description ai chatgpt real estate ai copywriting realtor automation mls descriptions claude for agents real estate ai tools listing photos ai real estate workflow