Listing Writer (Real Estate)

Fair Housing compliance is enforced at the generation layer, not as a post-generation filter. The build has a constraint set derived from HUD's guidance on discriminatory advertising language: protected-class references (familial status, national origin, religion, race, sex, disability, color), neighborhood steering phrases that imply demographic composition, and coded language that courts have found to function as proxies for protected characteristics. That includes phrases like "walking distance to top-rated schools," "quiet family neighborhood," "perfect for young professionals," and similar expressions that sound neutral but carry implied demographic signals under HUD enforcement precedent. The constraint layer is updated when HUD guidance or case law shifts. The agent still reviews output before posting — Fair Housing compliance is the agent's legal responsibility, and the build is a risk-reduction tool, not a legal guarantee. But clean output by default is meaningfully safer than writing freehand under deadline pressure.

The calibration pass at onboarding is what separates this from a generic AI writing tool. We pull your last fifteen to twenty published MLS descriptions and run a style analysis: sentence length distribution, vocabulary range, how you handle property features versus neighborhood context, whether you write in second person or third, how you open a description, how you close it. That profile gets embedded in the generation layer as the baseline voice. It's not perfect on day one — the first five to ten listings after go-live usually need slightly more editing while the calibration settles. By listing fifteen, most agents are making minor tweaks rather than rewrites. If your market position shifts — you move upmarket, you start working a new neighborhood heavily, your brokerage rebrands — a voice recalibration is included in the retainer. The goal is copy a client would recognize as yours, not copy that could belong to any agent on the board.

MLS character limits vary by board — some run 1,000 characters, some allow 2,500 words, and a handful still have field-specific caps that differ from the main description field. During onboarding, we map the exact constraints for the boards the agent is active on. That mapping is baked into the generation rules, so the MLS output is always sized to fit without manual trimming. Zillow and Realtor.com allow longer copy than most MLS boards, so the build generates an extended variant automatically when syndication is in scope. Social captions are sized for Instagram (no clickable links, visual-forward copy) and Facebook (slightly longer, call-to-action included) separately — the same caption doesn't work well on both. Email subject lines come in three options: one urgency-forward, one feature-forward, one curiosity-gap, so the agent picks the register that fits the property and the list segment.

The intake form takes about five minutes to fill out per listing. Required: square footage, bedroom and bathroom count, year built, key upgrades and their approximate ages, any seller-provided context about the property or neighborhood, and the listing price. Optional but high-value: photography notes (what does the photographer plan to emphasize?), the seller's favorite feature of the property, any quirks the agent wants to address head-on rather than leave buyers to discover, and the target buyer profile the agent has in mind. The more specific the input, the stronger the output — "updated kitchen" produces generic copy, "2022 kitchen renovation with Thermador range, quartz counters, and custom cabinetry" produces a description buyers actually respond to. For lot-only or off-market listings, there's a separate intake form that prompts for zoning classification, utility status, and access details instead of room-level specs.

Yes, and this is one of the scenarios where the constraint layer earns its keep. For as-is listings, the build applies a framing ruleset that avoids implied habitability claims, avoids characterizing the condition in ways that could create disclosure liability, and keeps language that speaks to the buyer's opportunity rather than the property's problems. For estate sale properties, there's a sensitivity layer that avoids language the seller's family might find inappropriate and keeps the tone transactional rather than clinical. For distressed properties — REO, short sale, auction — the build generates copy calibrated for investor buyers rather than end-user buyers, with different feature emphasis and a different call to action. The agent still needs to apply their own judgment and their brokerage's review process before any copy goes public. The build reduces the blank-page friction and the common-mistake risk; it doesn't replace the agent's professional responsibility.