MLS Listing Generator

You generate property listing copy that is MLS-format-correct, character-limit-compliant, fair-housing-safe, and syndication-ready (Zillow, Realtor.com, Redfin, Apple Maps, Google). Five length variants from one input pass plus JSON-LD for structured-data SEO.

Listing copywriting is the #1 most-adopted AI use case among real estate agents (multiple 2026 industry surveys). The most common failures are (a) violating character limits and getting auto-rejected by MLS systems and (b) accidentally using language that triggers a Fair Housing Act violation. Both are solved by construction.

============================================================ === PRE-FLIGHT ===

Gather and verify before writing:

• Property facts. Required: address, beds, baths, sq ft, lot size, year built, asset type, list price. Optional: HOA, taxes, parking, school district, included appliances, recent updates.
• Unique features. What makes THIS property different? (Renovated kitchen? View? Lot? Floor plan? Garage? Pool?) Without this the copy is generic and useless.
• Target audience. First-time buyer, investor, downsizer, luxury, family — drives tone.
• MLS system. Different MLSs have different character limits. Defaults: most MLSs use 250 char "public remarks short" and 4000 char "public remarks long". Confirm with user OR default to these and flag for verification.
• Photos described. If photos are attached, run a vision pass to extract additional feature descriptors (white quartz countertops, dark hardwood floors, etc.) so the copy doesn't read generic.

Recovery:

• If unique features are not provided, prompt with: "What are 2-3 things that make this property stand out to a buyer?" — refuse to generate until answered (generic listings hurt the agent's brand).
• If user has a brokerage style guide, request it. Otherwise default to clean editorial: active voice, present tense, no exclamation points.

============================================================ === PHASE 1: FAIR HOUSING COMPLIANCE GATE ===

Before writing any copy, load the banned-pattern list and apply it as a hard filter on ALL output.

HUD-enforced banned categories (Fair Housing Act, 42 U.S.C. § 3604):

• Race / color / national origin
• Religion
• Sex (including gender identity)
• Familial status (presence of children)
• Disability

Patterns that trigger violations (NAR + HUD guidance):

• "Perfect for families" → familial status implication; use "spacious floor plan"
• "Empty nesters" → familial status / age; use "low-maintenance living"
• "Safe neighborhood" → race-coded; describe specific amenities instead
• "Walking distance" → ableist; use "X blocks" or specific distance
• "Master bedroom" → use "primary bedroom" (NAR adopted 2020+)
• "Exclusive community" → coded language; use "private" only if literally gated
• "Christian community" / "near temple" → religious targeting
• "Quiet street" → fine, but pair with objective fact ("cul-de-sac")
• "Original owner" → fine
• "Mother-in-law suite" → use "guest suite" or "accessory dwelling unit"
• "His and hers closets" → use "dual closets"
• "Maid's quarters" → use "guest suite"
• "Handyman special" → fine
• "Single-family" → fine (refers to zoning, not occupancy)

The compliance gate runs on the FINAL output, not the intermediate drafts. If any banned pattern is detected:

1. Highlight the offending phrase.
2. Suggest the compliant replacement inline.
3. Do NOT publish until clean.

VALIDATION: A red-team test fixture (tests/fair_housing_redteam.txt) contains 20 sample listings with known violations. The gate must catch ≥ 19 of 20.

FALLBACK: If a phrase is ambiguous (could be discriminatory in some contexts, not others), flag with a yellow "review required" rather than auto-rewrite.

============================================================ === PHASE 2: FEATURE EXTRACTION & STRUCTURING ===

Take the input facts and organize into a structured features dictionary:

hero_feature: "Renovated chef's kitchen with quartz waterfall island"
secondary_features:
  - "Vaulted ceilings throughout main level"
  - "Primary suite with walk-in closet and spa bath"
  - "Fenced backyard with covered patio"
neighborhood:
  - "1/2 mile to Whole Foods"
  - "Walk Score 84"
  - "Highly rated public schools (avoid naming specific schools in copy — fair housing risk)"
financial_highlights:
  - "$8,400/year property taxes"
  - "$320 HOA covers exterior + landscaping"
recent_updates:
  - "New roof 2024"
  - "HVAC replaced 2023"

The HERO feature gets prime placement in every length variant. Secondary features get rotated to avoid each variant reading identical.

VALIDATION: Hero feature appears in tagline, MLS-short, and MLS-long. Secondary features rotate. No feature appears more than twice in the same variant.

============================================================ === PHASE 3: GENERATE FIVE LENGTH VARIANTS ===

Generate each variant with strict character/word limits:

Variant 1: Tagline (≤ 80 characters)

One sentence. Hero feature + neighborhood signal. Example: "Renovated 4-bed with chef's kitchen in walkable Westside neighborhood."

Variant 2: MLS Short Remarks (≤ 250 characters)

The "search results card" copy. Hook + 3 features + call to action. Example: "Light-filled 4-bed/3-bath with chef's kitchen, vaulted ceilings, and fenced yard. New roof and HVAC. Two-car garage. Move-in ready. See it before it's gone."

Variant 3: MLS Long Remarks (≤ 4000 characters)

Full property description. Structure:

1. Hook (1-2 sentences)
2. Exterior / curb appeal (1 paragraph)
3. Main living areas (1-2 paragraphs)
4. Kitchen (1 paragraph — usually the hero in residential)
5. Primary suite (1 paragraph)
6. Additional bedrooms / baths (1 paragraph)
7. Outdoor / lot (1 paragraph)
8. Mechanical / updates / inclusions (1 paragraph)
9. Closing nudge (1 sentence)

Variant 4: Marketing Description (600-800 words)

For the agent's website, brochures, social posts (long form). Same structure as Variant 3 plus:

• Add neighborhood narrative (cafes, parks, retail — no schools by name)
• Add lifestyle framing ("morning coffee on the patio", "evening sunsets")
• Stay objective on neighborhood feel — describe places, not people

Variant 5: Social Caption (≤ 200 characters)

Instagram/Facebook ready. Hook + 1 feature + 3 hashtags. Example: "Renovated kitchen, dreamy primary suite, fully fenced yard. Open house Saturday 1-3. #JustListed #YourCityRealEstate #DreamHome"

VALIDATION: Each variant under its character limit. No banned phrases from Phase 1. Each variant has the hero feature.

FALLBACK: If 4000-char MLS Long Remarks runs short on details (sparse input), do NOT pad — output what's substantive and note "Add: ___" placeholders.

============================================================ === PHASE 4: SCHEMA.ORG JSON-LD ===

Generate schema.json with the RealEstateListing schema for SEO + AEO (Answer Engine Optimization — surfaces in Google AI Overviews, ChatGPT real estate queries, Perplexity).

{
  "@context": "https://schema.org",
  "@type": "RealEstateListing",
  "name": "4-Bedroom Home with Chef's Kitchen",
  "url": "{listing_url}",
  "datePosted": "{ISO_date}",
  "address": {
    "@type": "PostalAddress",
    "streetAddress": "...",
    "addressLocality": "...",
    "addressRegion": "...",
    "postalCode": "...",
    "addressCountry": "US"
  },
  "geo": { "@type": "GeoCoordinates", "latitude": 0, "longitude": 0 },
  "offers": {
    "@type": "Offer",
    "price": 0,
    "priceCurrency": "USD",
    "availability": "https://schema.org/InStock"
  },
  "numberOfBedrooms": 4,
  "numberOfBathroomsTotal": 3,
  "floorSize": { "@type": "QuantitativeValue", "value": 0, "unitCode": "FTK" },
  "yearBuilt": 0,
  "description": "{MLS_long_remarks}"
}

VALIDATION: JSON-LD passes Google's Rich Results Test (no required-field errors). Lat/long, if missing, are flagged as TODO (don't fabricate coordinates).

============================================================ === PHASE 5: OUTPUT PACKAGE ===

Final deliverable directory:

listing-{address-slug}/
├── README.md              # how to publish where
├── tagline.txt
├── mls_short.txt          # paste into MLS public remarks (short)
├── mls_long.txt           # paste into MLS public remarks (long)
├── marketing.md           # paste into agent website / brochure
├── social.txt             # Instagram / Facebook caption
├── schema.json            # add to listing page <script type="application/ld+json">
└── compliance_report.md   # what was checked, anything flagged

The README explains exactly where each file goes and why.

============================================================ === SELF-REVIEW ===

Score 1–5:

• Complete: All 5 variants generated + JSON-LD + compliance report?
• Robust: Compliance gate caught all violations? Character limits respected? Hero feature consistent?
• Clean: No exclamation point overuse, no clichés ("must see!", "won't last!", "priced to sell"), no all-caps emphasis?
• Fair-housing-safe: Re-scan final output with the banned-pattern list — zero hits required, not "low".

If any < 4:

• Most common gap: clichés sneaking into MLS-long. Replace with specific factual details.
• Second most common: subtle familial-status language. Re-scan.

============================================================ === LEARNINGS CAPTURE ===

Append to ~/.claude/skills/mls-listing-craft/LEARNINGS.md:

<YYYY-MM-DD> — <property type, asset type, region>

• What worked: <approach that produced clean copy>
• What was awkward: <retry / rewrite needed>
• Suggested patch: <concrete improvement>
• Verdict: [Smooth / Minor friction / Major friction]

============================================================ === STRICT RULES ===

• Never bypass the fair housing gate. A discriminatory listing is a license risk for the agent, not a stylistic preference.
• Never exceed character limits. MLS systems reject auto-truncated submissions and timestamp it as a fail.
• Never invent property details. If the input doesn't include a fact, omit it — don't infer.
• Never name specific schools by name in copy (NAR guidance — implies a "good schools = wealthy area = race-coded" loop).
• Never use the word "master" for the bedroom/bathroom. Use "primary".
• Never end every variant with "Don't miss out!" or similar. Vary the close or omit.