Claude General

Private Data Prompts: 6 for Work You Should Not Paste Into a Chatbot

Apple spent its whole event arguing that on-device processing is the point, and for personal data it has a case. But most people still paste client documents, salary data and medical details into a cloud assistant without thinking. These six prompts are about getting the help without handing over the data — redacting properly, working from structure rather than content, and knowing when to stop.

⌨️ 6 prompts 🕐 Updated Sep 10, 2026

💡 How to use these prompts: Replace everything in [BRACKETS] with your specific details before sending. Click Copy to copy any prompt to your clipboard instantly.

Work from structure, not content

Most document help is about structure. You rarely need to hand over the content to get it.

I need help with a document I cannot share. Here is its structure and purpose: [DESCRIBE THE SHAPE, NOT THE CONTENT]

Tell me:
- What a strong version of this document contains
- The order the sections should go in
- What is commonly missing
- What a reader would look for first

I will apply it myself. Do not ask me to paste the document.

Check my redaction before I paste anything

Redaction usually fails on combination rather than on names. This is the check most people skip.

Here is a document I have already redacted: [PASTE THE REDACTED VERSION]

Tell me what could still identify a person or organisation:
- Names I missed
- Details that identify by combination rather than directly
- Dates, amounts or locations that narrow it to one entity
- Anything distinctive in phrasing

Assume a reader who wants to work out who this is.

Build a reusable substitution scheme

A scheme you apply inconsistently is worse than none, because you stop checking.

I regularly need help with documents containing: [LIST THE CATEGORIES OF SENSITIVE DATA]

Design a substitution scheme I can apply consistently — what to replace, what to replace it with, and how to keep it reversible on my side.

Keep it simple enough that I will actually use it every time.

Ask about the category, not the case

The last line matters. Assistants ask for detail by default, and the detail is the thing you are protecting.

I have a situation I cannot describe in detail. In general terms it involves: [THE CATEGORY]

Tell me:
- What usually matters in situations of this kind
- What people typically get wrong
- What questions I should be asking
- When this needs a professional rather than a model

Do not ask me for specifics.

Decide what actually needs to stay local

Most people have never drawn this line explicitly, which is why it gets crossed accidentally.

Here is the work I do: [DESCRIBE]

Sort it into what could safely go to a cloud assistant, what needs redacting first, and what should never leave my machine.

For the last category, tell me what I could do locally instead.

Be strict. I would rather over-classify than explain a leak.

Review my own habits

Written now this is an audit. Written after a problem it is a disclosure.

Here is what I have pasted into AI tools over the past month: [DESCRIBE THE CATEGORIES]

Tell me:
- What I should not have shared
- What the realistic consequence is, if any
- What to change going forward
- Whether anything warrants telling someone

Be direct. I am asking because I want to know, not to be reassured.