How to redact a PDF before sending it to an AI
Pasting a PDF into a chatbot is a disclosure: client names, identifiers and addresses reach a third-party server the moment you hit send. Before any AI sees the document, the sensitive data has to be redacted or anonymized on your own computer. This guide explains how to redact a PDF for AI safely, why blurring often fails, and the fastest local workflow.
Redaction, anonymization, pseudonymization — the difference
True redaction destroys information forever — the black bars you see on court filings and published records. Anonymization removes identity from the document. Pseudonymization replaces names and identifiers with consistent placeholders (Person_1, Client_1) so the text stays fully readable, and the change is reversible on your own computer. For AI work you almost always want pseudonymization, because the model still needs to reason about who did what.
Why blurring or black boxes often fail
Drawing a black rectangle or a blur over PDF text usually leaves the text layer intact underneath — anyone can select and copy the words you thought you hid. Scanned PDFs often carry an invisible OCR text layer that behaves the same way. PDF metadata (author, title, revision history) can leak names even when the visible page looks clean. True redaction has to remove the underlying text, not just cover it.
Prepare a PDF for AI with Redact for AI, step by step
- 01
Download Redact for AI
Free, runs entirely offline on macOS 13+ (Apple Silicon and Intel) and Windows 10+. No account, no cloud.
- 02
Open the PDF
Drop the file into the app. Scanned PDFs are read with local OCR — Apple Vision on macOS, Tesseract on Windows — in English, Italian, Portuguese and Spanish.
- 03
Automatic local pseudonymization
Every name, identifier and address becomes a consistent placeholder on your computer. Nothing is uploaded.
- 04
Review the highlighted preview
A side-by-side preview shows every replacement. Add anything the detector missed with one selection.
- 05
Copy the anonymized text into your AI
Paste into ChatGPT, Claude, Gemini or any other assistant. Only placeholders reach the model.
- 06
Restore the real names
Paste the AI's draft back into the Restore tab. The mapping stays on your machine and the real names return with one click.
When you need true visual redaction instead
If the goal is filing or publishing a censored PDF — a court exhibit, a FOIA release, a document you will hand to the other side — use a dedicated PDF redaction tool and verify the text layer is really removed, not just covered. Redact for AI's job is different: it keeps documents useful for AI without exposing anyone. The two workflows solve different problems.
Frequently asked questions
How do I redact a PDF for free?
For AI workflows, Redact for AI is free and runs locally — it replaces names and identifiers with placeholders you can safely paste into any AI. For permanent visual redaction (black bars for filing or publishing), use a dedicated PDF redaction tool and confirm it removes the underlying text layer.
Is blurring text in a PDF safe?
Usually not. A blur or black rectangle is drawn on top of the page, but the text layer often stays intact underneath and can be selected and copied. True redaction has to remove the underlying text, not just hide it visually.
How do I redact a scanned PDF?
A scanned PDF is an image, so it must be OCR'd first. Redact for AI does the OCR locally, then pseudonymizes the recognized text on your computer. Nothing is uploaded.
Can redaction be undone?
True redaction, no — that is the point. Pseudonymization is different: a local mapping file lets you restore the real names in the AI's draft, and only you hold that file.
What should I do before pasting a document into ChatGPT or Claude?
Replace names and identifiers with placeholders on your own computer, review the preview to catch anything the detector missed, and paste only the anonymized text into the AI.