Friday evening. The agency is still busy. A negotiator is finalizing a sale agreement, a property manager is rereading a lease signed that afternoon, and someone asks an AI to "reword the rent-review clause." The instinct is right. The copy-paste is not. Because that lease carries the tenant's name, the guarantor's name, the exact address of the property and the deposit amount, and once they leave the agency every one of those details points to a real person. The rule fits in one line. Mask the parties and the address. Keep the figures that drive the analysis.
The parties and the address are masked, the price stays: the AI drafts or checks without knowing who or which property.
The mandate and the sale agreement: who signs, for which property
On a sales mandate or an agreement, the real identifiers aren't only the seller's and buyer's names. The address gives it away. "12 Lilac Street" points to a household as surely as a surname does, and that single line is often enough to recognize the file at a glance. So you pseudonymize the parties and the address. The price, the floor area, the type of lot? Leave them. They identify no one and stay useful to the AI.
The lease: tenant, guarantor and the file's paperwork
A lease gathers far more people. Beyond the tenant come the guarantor and, sometimes, a spouse or a flatmate, each with their own income and supporting documents to provide. Tax notices. Payslips. Bank statements. These are the most sensitive data in a rental file, the kind you never want drifting toward a third-party tool without precaution. Before sending a lease to an AI, replace those names and documents with pseudonyms. The rent and the deposit can stay.
Diagnostics, building management and the rest of the file
A property file doesn't stop at the lease. Energy and safety reports, inventories of fixtures, condo association minutes: all carry names, addresses and, sometimes, direct contact details. And yet the charges, the energy rating or the board's decisions summarize perfectly well without ever revealing who lives where.
How it works in practice
The principle is simple. You pseudonymize the document, the AI works on the masked version (drafting, clause checking, file summary), then you re-identify locally. Pseudonymization is reversible. You get the full document back, intact, without any individual ever reaching a third-party tool.
Safe-Doc masks mandates, sale agreements and leases before the AI. The layout stays, processing happens in the European Union, then everything is purged. See the ChatGPT and GDPR at work guide.
Sellers, buyers, tenants, guarantors: everyone stays protected. Pseudonymize the file before the AI.