The Fair Housing words AI always gets wrong
AI-generated listing descriptions are fast, convenient, and — if you're not careful — a Fair Housing lawsuit waiting to happen.
Large language models are trained on decades of real estate copy, much of it written before modern Fair Housing enforcement. They happily reproduce phrases that reference protected classes, imply preferences, or steer buyers. The Fair Housing Act prohibits discrimination based on race, color, religion, sex, disability, familial status, and national origin — and the language that triggers a violation is often subtle.
The most common offenders
"Perfect for a family" — References familial status. Replace with "spacious layout with room to grow."
"Safe neighborhood" — Implies a preference based on crime statistics, which courts treat as a proxy for race. Replace with factual descriptions: "walking distance to parks and schools."
"Exclusive community" — Suggests exclusion. Replace with "established neighborhood."
"Not suitable for children" — Direct familial status violation. Remove entirely; let buyers decide suitability.
"Walking distance to church" — References religion. Replace with "close to community amenities."
"Ideal for young professionals" — Implies age and familial status. Replace with "great for commuters."
Why AI gets this wrong
AI models optimize for appealing marketing copy. They've learned that emotional, targeted language sells — but that same targeting is what Fair Housing law restricts. The model doesn't know the legal line; it just knows what sounds good.
What to do
- Never publish an AI-generated listing without review. Every description must be checked.
- Scan before you publish. Use ComplianceGuard to catch violations automatically.
- Focus on the property, not the buyer. Describe features, not who should live there.
- Remove protected-class references entirely. Don't try to soften them — delete them.
The good news: once you know the patterns, they're easy to catch. The bad news: if you're publishing dozens of AI listings a week, manual review doesn't scale. That's exactly the gap ComplianceGuard was built to close.