Data Privacy Checklist

Brand and Use Checklist

 

What to anonymise before pasting into an LLM — and how to use AI safely with sensitive property data.

The Core Rule

Assume anything you paste into an LLM could be seen by a third party. This is not always the case — but it is the right default posture until you have verified your tool’s data settings.

Before you paste anything
Check that your LLM has "improve model" or "train on conversations" turned off.
Claude: Settings → Privacy  ·  ChatGPT: Settings → Data Controls

 

What is Safe to Share

Data type Safe? Notes
Generic property descriptions ✓ Safe Property type, size, location suburb — no identifying details
Anonymised lease terms ✓ Safe Rent amounts, terms, review mechanisms — without tenant name or address
Process questions ✓ Safe "How do I handle a lease expiry?" — no personal data involved
Draft communications ✓ Safe With placeholder names — replace real names before pasting
Market research questions ✓ Safe General market analysis with no client-specific data

 

What to Anonymise First

Data type Risk Replace with
Tenant full name ⚠ Replace [Tenant Name] or [Tenant A]
Property street address ⚠ Replace [Property Address] or suburb only
Tenant contact details ⚠ Replace [Phone] / [Email]
Financial account numbers ⚠ Replace [Account Number]
Owner personal details ⚠ Replace [Owner Name]
NZ/AU IRD or tax numbers ⚠ Replace [Tax Number]

 

Never Share

Full tenancy application documents
Contain identity documents, financial history, references.
Full lease agreements with all parties named
Summarise or extract specific clauses with names removed.
Bank statements or financial records
Even for arrears analysis — use anonymised figures only.
Maintenance records with tenant vulnerability information
Medical, family, or personal circumstances.

 

Pre-Send Review Checklist

Check Done?
🔍 Scanned for full names (tenant, owner, guarantor) [ ]
🔍 Removed or replaced street addresses [ ]
🔍 Removed phone numbers and email addresses [ ]
🔍 Removed account numbers and financial identifiers [ ]
🔍 Verified LLM privacy settings are set to "do not train" [ ]
🔍 Not uploading a full document when only a clause is needed [ ]

 

Privacy Settings by Tool

Tool Where to find it Setting to enable
Claude Settings → Privacy Turn off "Improve Claude for everyone"
ChatGPT Settings → Data Controls Turn off "Improve the model for everyone"
Microsoft Copilot Privacy settings in M365 admin Configure tenant-level data boundaries

 

Best practice
Create a firm-wide rule that all AI use happens with these settings verified. Add a line to your AI Usage Policy making this a requirement.