The problem
Listing sites like Yad2 (Israel) or Zillow (US) turn every property into a uniform row in a database. Same template, same structure, same visual noise. A family home with a story behind it — the garden someone spent ten years building, the kitchen that's actually the heart of the house — becomes "4BR / 2BA / 120sqm."
Sellers who aren't real estate agents have no good tools for presenting their property in a way that makes buyers stop and feel something. They either rely on the listing site's generic format, or they try to build something themselves and spend days on it.
The insight: a standalone landing page does what a listing row can't. It sets the tone, tells the story, and makes a buyer imagine themselves in the space rather than comparing it to a spreadsheet.
What I built
Property Landing Builder is a zero-friction web tool. You input the property details — rooms, location, features, photos, and anything you want to highlight — and Claude AI writes a compelling narrative and generates a beautiful, standalone landing page.
No account. No design skills. No subscription. You get a page in under five minutes. Share the URL directly with buyers, post it in WhatsApp groups, or link to it from a listing site.
The AI writing is the core value. Claude doesn't just summarize the inputs — it writes the property's story. It finds the angle: maybe the location is everything, or the renovation history makes it special, or the layout is perfect for a specific kind of buyer. The output reads like it was written by someone who actually cares about the property, because the prompt engineering makes it behave that way.
The key product decisions
No account required. Every friction point before "see the output" is a reason to leave. I removed them all. You fill in the form, hit generate, and get a URL. If you want to edit it, the editor is right there. No sign-up gate.
Claude over GPT. I tested both. Claude writes more nuanced, less generic real estate copy. The tendency to vary sentence structure and avoid filler phrases made the output significantly better in blind tests I ran on early users.
Mobile-first generated pages. Buyers share listing links on WhatsApp. WhatsApp is mobile. The generated pages needed to look great on a phone first, desktop second.
Opinionated design with no customization. Early versions had style options. They slowed everything down and introduced bad choices. I removed them. One beautiful template, done well, beats twenty mediocre options.
What I learned about AI product design
The hardest part wasn't writing the prompt — it was handling the cases where the AI output was wrong in a way users couldn't immediately see. A hallucinated detail about a property (e.g., implying there's a pool when there isn't) is a legal and trust problem, not just a UX problem.
My solution: the prompt instructs Claude to only use information explicitly provided in the form inputs, never to invent or assume, and to flag any detail it's uncertain about with a placeholder instead of a guess. This produces more conservative copy in some cases, but it means the user can trust what they're publishing.
The second big learning: streaming the response dramatically improves perceived quality. When users see text appearing in real time, they feel like something smart is happening. When they wait 15 seconds for a full response to appear, they get anxious about whether it's working. The underlying quality was identical — the experience was night and day.
What's next
Currently working on a sharing/tracking layer: a simple analytics dashboard on each generated page so sellers can see who's viewed it and when. This turns a static page into a lead-tracking tool, which changes the value proposition from "nice presentation" to "actual business intelligence about your buyers."