What this automation does
Agents and marketing coordinators write the same kinds of descriptions over and over, under character limits, and under legal rules that are easy to trip over. This workflow turns one confirmed fact sheet into every version you need, and checks each one before a person approves it.
- Collect facts. The listing agent completes a structured intake form: beds, baths, finished square footage and its source, lot size, year built, updates with years (roof 2021, HVAC 2019), included appliances, HOA details and their own notes about what makes the home stand out. Photos land in a shared folder from the photographer.
- Build the fact sheet. The AI merges the form and notes into a clean fact sheet. It reviews the photos and proposes extra features, each marked “from photo, confirm”. The agent ticks what is true.
- Draft every channel. From confirmed facts only, the model writes MLS public remarks within your MLS’s character limit, a longer website description, a flyer blurb and two or three social captions.
- Check the language. Code scans each draft against a flagged-phrase list. A second model pass looks for wording about people instead of property, unverified claims (school assignment, “new” anything without a year) and banned clichés.
- Approve. The agent sees the drafts with any flags highlighted, edits, and approves. Approved web and social versions publish or queue automatically, and the agent pastes MLS remarks into the input screen.
When it is worth building
Build this when your brokerage or management company publishes enough listings that writing time adds up, when you want every agent’s listings to read in one house style, or when your broker wants the same compliance check applied to every listing instead of trusting each agent’s judgment.
Do not build it for a single agent with a few listings a year. A chat assistant and a saved prompt will do. Many MLS platforms and real estate CRMs also now include AI description writers, and if one is good enough for your listings, use it. The custom workflow earns its place through the fact sheet, the per-channel outputs and the compliance check, not through the writing itself.
Fair Housing language risk
The Fair Housing Act makes it unlawful to publish ads that indicate a preference, limitation or discrimination based on race, color, religion, sex, disability, familial status or national origin, and many states and cities protect additional classes. AI models trained on years of real estate marketing have absorbed phrases that cross that line. The rule of thumb: describe the property and its features, not the people who should live there.
| Risky | Safer |
|---|---|
| Perfect for a young couple | Two bedrooms with an open living area |
| Great for empty nesters | Single-level living with a main-floor primary suite |
| Ideal for families with kids | Fenced backyard and a finished lower level |
| Exclusive, safe neighborhood | Cul-de-sac location, 0.4 miles to the train station |
| No vouchers or no Section 8 (where source of income is protected) | Leave tenant screening criteria out of the ad and apply them consistently |
There are lawful exceptions. A community that qualifies as housing for older persons can advertise as 55-plus, for example. Your broker and counsel should own the flagged list, and the AI only applies it.
Architecture
| Layer | What we use | Why |
|---|---|---|
| Intake | Jotform or a Google Form writing to Google Sheets | Structured facts beat free-text notes, and agents can fill it from a phone |
| Photos | A Google Drive folder per listing | Photographers already deliver there |
| Drafting | Claude Sonnet 5.5 with vision, or OpenAI’s GPT models | Good at tight, varied copy and at proposing features from images |
| Compliance check | A phrase list in code plus a second model pass | Code catches known phrases every time, and the model catches paraphrases |
| Orchestration | n8n | Built-in Google Drive and Sheets nodes, with approval steps |
| Publishing | Your website CMS (for example Webflow) and social scheduler | Channels you control can publish automatically after approval |
How to build it
- Collect 20 to 30 of your best past listings. These set the house style, and the agent rewrites any that would fail the new compliance check.
- Design the intake form. Required fields for facts, a notes field for the story, and a source field for square footage.
- Write one prompt per channel. MLS remarks get the strictest rules: character limit enforced in code after generation, no contact details or URLs if your MLS prohibits them in public remarks, and no showing instructions.
- Add the photo pass. The model returns proposed features as a list, never as prose, so the agent confirms them one at a time.
- Build the compliance check. Start with the broker’s list of flagged terms, add a cliché list (“nestled,” “boasts,” “won’t last”), and route any flag to the approval screen with the reason.
- Wire approval and publishing. Approve in a simple review page or by email, then push to the website and social queue.
- Handle changes. A price change or new photos regenerate only the affected versions, and the agent re-approves.
The same confirmed-facts approach drives our ecommerce product description guide. Once listings are live, AI lead follow-up can quote the same fact sheet in its first reply.
What breaks in production
- Virtual staging. The model describes a staged sectional as if it comes with the house. Tag virtually staged photos at upload, and exclude them from the feature pass.
- Material misstatements. “Hardwood throughout” when two rooms are carpet. Only confirmed facts reach copy, and adjectives about materials need a fact behind them.
- School assignments. Boundaries change. Name the district only if the agent confirmed it, and point buyers to the district to verify.
- Character counting. Some MLS inputs count spaces and line breaks differently. Enforce the limit in code with a margin.
- Sameness. Fifty listings that all open with “Welcome home” read like a template. Rotate openings and penalize repeats from the brokerage’s recent listings.
- Square footage disputes. Tax records, appraisals and builder plans disagree. The fact sheet shows the source next to the number.
Timeline and cost
This is a small build, usually one to two weeks. Lumeor builds it in the range of $2,500 to $8,000. The low end is intake, MLS remarks and website copy with the compliance check. The high end adds the photo pass, social and flyer versions, automatic publishing to your site and per-office style rules. Model usage per listing is small, usually cents. See what workflow automation costs for how these are priced, and ChatGPT vs Claude for business if you are choosing a model.
How to measure success
- Agent minutes per listing, from signed agreement to live listing
- Edits per draft before approval
- Compliance flags per draft, which should fall as prompts improve
- Corrections requested after publication
- Time from listing agreement to MLS go-live