How to Redact a PDF Before Uploading to ChatGPT or Any AI Tool

People are uploading sensitive documents to AI tools every day. Medical test results to understand a diagnosis. Bank statements to ask for budgeting advice. Legal contracts to check for unfavourable clauses. Tax returns to find deductions.

This is genuinely useful. But it also means your most sensitive personal information — Social Security numbers, account numbers, diagnoses, names, addresses — is being sent to a third-party server and processed by a model whose data retention policy you probably have not read.

What actually happens when you upload a PDF to an AI tool

When you attach a document to ChatGPT, Claude, Gemini, or any other AI assistant, the entire file is sent to that company's servers. The text is extracted, processed, and used to generate a response. Depending on the platform and your account settings, that content may be retained, used for model training, or accessible to support staff.

OpenAI's default settings, for example, allow conversation content to be used to improve models unless you opt out. Most people never change the defaults.

What you should redact before uploading

The goal is not to make the document useless to the AI — it is to remove the specific identifiers that create risk if they end up somewhere they should not.

• Social Security or National Insurance numbers
• Bank account and credit card numbers
• Full names (replace with initials or a placeholder if the AI does not need them)
• Dates of birth
• Home addresses
• Medical record numbers or patient IDs
• Passport or driving licence numbers
• Employer identification numbers

In most cases, the AI does not need these identifiers to help you. It can summarise a medical report without knowing your patient ID. It can review a contract without knowing the precise home addresses of the parties.

How to redact a PDF locally before uploading

True redaction means the text is permanently removed from the file — not just covered by a black box. A black annotation box in Preview or Adobe Reader leaves the underlying text intact and selectable. Anyone (or any AI) parsing the raw file can still read it.

Locdone's Redact PDF tool burns the redacted areas directly into the page as rendered pixels. The text is gone from the file structure, not just visually hidden.

Critically, the redaction happens entirely in your browser. The original file — with the sensitive content still present — is never sent anywhere. You select the areas to redact, the tool processes the file locally, and you download a clean version.

1. Open locdone.com/redact-pdf
2. Drop your PDF in
3. Draw boxes over every piece of information you want to remove
4. Click Redact and download the clean file
5. Upload the redacted version to ChatGPT, Claude, or wherever you need

Should you also strip metadata?

Yes. PDF files carry hidden metadata that most people never see — the author name, the software used to create the document, creation and modification timestamps, and sometimes GPS coordinates or organisational information embedded by enterprise software.

When you upload a document to an AI tool, that metadata is included. Use Locdone's Strip PDF Metadata tool to remove it before uploading.

The bottom line

AI tools are genuinely useful for understanding complex documents. Redacting the sensitive identifiers first takes about two minutes and meaningfully reduces the risk. The AI gets enough context to help you. The account number, SSN, and patient ID stay on your device.