From one listing sheet to four assets: AI content for Vértice Properties
A production flow where AI drafts from verified data and a person checks every claim before publishing.
- Agency
- Vértice Properties (fictitious)
- Input
- One verified data sheet
- Rule
- AI proposes, a person approves
The story
Concept · local demo with prepared content
- Built with
- TypeScriptReact
01
The context
An agency with about twenty new properties a month and a sales team with no time for social media. Each agent wrote in a different tone.
02
The problem
Listings with copied text, unedited photos and, worst of all, data that didn’t match across channels: one extra square metre on a portal is a problem, not a style mistake.
03
The reasoning
Quality depends on the input sheet more than on the model. Sensitive data —price, area, legal status— is never generated: it is copied from the sheet. AI drafts, adapts and proposes; the person checks and approves.
04
Content studio
One fictitious sheet, four deliverables and a human review. Turn on review to see which data comes from the sheet and which draft claim was stopped before publishing. Prepared content, no calls to any AI.
Fictitious property and brand · prepared content
Property portals
Apartment of 86 m² with 3 bedrooms, 2 bathrooms and a balcony with mountain views. The open kitchen renovated in 2023 joins the living and dining room in one space with morning light. It is on the 7th of 12 floors, with lift, with 1 covered parking space and a two-minute walk from the metro. Price: COP 520,000,000. Service charge: COP 380,000 a month. Property free of liens (checked by the legal team).
The draft invented the distance. The sheet says “6 blocks”: corrected with the verified data.
- Every figure comes from the sheet: pending
- Brand tone: warm, no superlatives: passed
- No return or appreciation promises: passed
- Length fits the channel: passed
Can’t approve: there is a claim that isn’t in the sheet. Switch to “After review”.
05
How it would be measured
Preparation time per property, errors caught in review and enquiries per post. It has not been used with real properties.
06
Learnings
Without verified data, AI fills gaps with plausible sentences. The review checklist isn’t red tape: it is what allows fast production without publishing something false.
Next case
Market research for Rin, software for repair shops