1. Recognizing Sensitive Data

Before you send a prompt to an AI model, you should know which types of data are particularly worthy of protection:

2. The Basic Rule: What Should Not Be Shared

Note: AI usage must be discussed/approved within the company.

Before every prompt, ask yourself: „Would I share this information publicly?“

If the answer is „No,“ you should either omit this information or anonymize it. The more confidential the information, the more cautious you should be.

Particularly Risky Data:

• Passwords and access credentials
• Internal financial figures
• Personal data of customers or employees
• Information covered by NDAs
• Unpublished strategies or product information

3. Practical Anonymization Techniques

With these simple methods, you can design prompts so they don’t contain sensitive data but still deliver good results:

Replace Names

Instead of real names, use generic terms or placeholders:

• „Customer A“ instead of „John Doe“
• „Employee B“ instead of „Jane Smith“

Use Placeholders

Replace specific information with placeholders:

• „COMPANY_NAME“ instead of the actual company name
• „PRODUCT_NAME“ instead of the actual product

Generalize Data

Describe information more generally:

• „A medium-sized company“ instead of a specific company
• „A five-figure amount“ instead of the exact sum

Use Sample Examples

Replace real examples with similar but fictitious ones:

• Replace a real email with a similar but made-up version
• Replace real data with representative sample data

4. Concrete Examples: Problematic vs. Harmless