AI Personal Data Redaction Pass

Purpose

Help the user prepare real-world content for use with an AI tool by removing or replacing personal data before sharing. The deliverable is a redaction pass: a flagged identifier table, sensitivity notes, placeholder map, and a safe-to-paste version of the content.

This is a prompt-only privacy cleanup workflow. It does not store secrets, create a private-data database, provide legal advice, provide medical advice, provide financial advice, or guarantee that every identifier has been removed. Extra caution is required for legal, medical, financial, employment, immigration, education, child-related, and identity documents.

Use This Skill When

Use this skill when the user wants to:

• Paste notes, screenshots, forms, emails, chats, transcripts, resumes, records, or support tickets into an AI tool.
• Remove names, addresses, phone numbers, emails, account numbers, IDs, dates of birth, signatures, faces, locations, employer names, school names, case numbers, or other identifiers.
• Replace sensitive details with consistent placeholders while keeping the useful structure of the content.
• Create a before-and-after privacy artifact they can review before sharing externally.
• Separate what is safe to paste from what should stay private.

Do not use this skill to preserve, store, reconstruct, infer, deanonymize, expose, or trade personal data. Do not ask the user to provide secrets or unnecessary sensitive details.

Best Inputs

Ask for the minimum content needed to create the redaction pass. If the user has already pasted sensitive material, work from it cautiously and do not repeat secrets unnecessarily.

• The text or a user-provided summary of the material to clean.
• Intended destination, such as a general AI assistant, vendor support form, public forum, coworker, contractor, or classroom.
• Purpose of the AI task, such as summarizing, drafting, translating, extracting action items, or checking tone.
• Required facts that must remain useful after redaction.
• Content type, such as medical, legal, financial, employment, school, family, travel, customer support, or personal notes.
• Any identifiers the user specifically wants preserved, generalized, or removed.

Workflow

1. Set privacy scope. Confirm that the output is a cleanup artifact and safe-to-paste copy, not secret storage or professional legal, medical, or financial guidance.
2. Identify direct identifiers. Flag obvious names, contact details, government IDs, account numbers, case numbers, addresses, exact locations, faces in image descriptions, usernames, links, and signatures.
3. Identify indirect identifiers. Flag combinations that could reveal a person, such as rare job titles, small towns, exact dates, unique events, school names, employer names, medical details, family roles, or timestamps.
4. Classify sensitivity. Mark items as low, moderate, high, or critical sensitivity based on possible harm if shared.
5. Choose placeholders. Replace identifiers with consistent labels such as [PERSON_A], [COMPANY_B], [CITY], [DATE_RANGE], [ACCOUNT_ID], or [MEDICAL_DETAIL]. Keep placeholder names generic.
6. Protect context. Keep enough non-sensitive context for the AI task while removing details that are not needed.
7. Produce the redaction table. Show each flagged item category, risk level, replacement, and reason. Avoid repeating the original value when it is highly sensitive; describe the category instead.
8. Create safe-to-paste copy. Provide the cleaned version using placeholders. Do not include original secrets, full IDs, access tokens, passwords, or private keys.
9. Add a review checklist. Tell the user what to manually inspect before sharing, especially for images, attachments, hidden metadata, file names, comments, tracked changes, and screenshots.

Output Format

Return the artifact in this order.

1. Scope and Caution Note

State that this is a privacy cleanup pass for AI use. If legal, medical, financial, employment, immigration, or child-related content is present, recommend extra human review before external sharing.

2. Intended Use Snapshot

3. Identifier Review Table

For critical items, do not repeat the original value. Use category descriptions such as "bank account number" or "minor child's school name."

4. Placeholder Map

List every placeholder and what general role it represents. Keep the map generic enough that it can be shared with the safe copy.

5. Safe-to-Paste Copy

Provide the cleaned content with placeholders. Preserve structure, order, tone, and necessary facts when possible.

6. Removed or Generalized Details

Summarize what was removed, generalized, or converted to ranges. Do not include secret values.

7. Manual Review Checklist

Include checks for:

• Names, contact details, addresses, account numbers, IDs, dates of birth, links, usernames, and signatures.
• Rare combinations that could identify someone.
• Image backgrounds, faces, forms, screenshots, file names, hidden metadata, comments, and tracked changes.
• Legal, medical, financial, employment, immigration, education, or child-related details that deserve extra care.

8. Paste Decision

End with one of these labels:

• Ready after review: only low-risk placeholders remain.
• Needs more redaction: more identifiers are visible.
• Do not paste externally: secrets, regulated data, or high-risk personal details remain.

Example Prompts

• "I want to paste a support ticket into an AI tool. Help me remove names, account numbers, and addresses first."
• "Redact personal details from this meeting transcript so I can ask an AI to extract action items safely."
• "I have a form with medical details to summarize. Create a safe-to-paste copy with placeholders instead of real data."

Safety Boundaries

• Do not store secrets or create a reusable private-data record.
• Do not reveal, reconstruct, or infer redacted identities.
• Do not ask for passwords, access tokens, private keys, full government IDs, bank credentials, or unnecessary sensitive data.
• Do not claim the copy is perfectly anonymous.
• For legal, medical, financial, employment, immigration, or child-related content, advise extra review and qualified support where appropriate.