Guide · AI

How to redact a PDF before pasting it into ChatGPT or Claude

AI assistants are great at reading documents. They are also a place your client's data can end up. Here is how to strip the sensitive parts first, on your device, before anything reaches the model.

AI tools are genuinely useful for summarizing a contract, pulling figures out of a statement, or drafting a reply to a letter. The catch is that the moment you paste or upload a document, it leaves your control.

Why a raw document is a problem for AI tools

When a file reaches an AI assistant, what happens to it depends on the product and your settings. The content may be retained, reviewed by people for quality, or used to help improve future models. Policies vary and they change, which is exactly the point: once it is out, the terms are no longer yours to set.

For a public article, that is no issue. For a contract, a medical note, a tax form, or anything under confidentiality or an NDA, it can be a real one. "I pasted the client file into a chatbot" is not a sentence you want to be explaining afterward.

The fix is not to avoid AI. It is to remove the identifying details first, so the model gets the substance without the people.

What to strip before a document reaches a model

  • Names of people, both direct and the indirect hints that point to one person.
  • Account numbers, Social Security numbers, card numbers, policy and case numbers.
  • Addresses, phone numbers, email addresses, and dates tied to a specific person.
  • Anything that, alone or combined with the rest, points back to an individual.

Two ways to do it

Option A: redact

Permanently remove the sensitive parts and work from the clean version. Best when the model does not need the real details to help you. See how to redact so the text is actually gone.

Option B: anonymize for AI

Replace names and identifiers with consistent placeholders, for example [PERSON 1] and [ACCOUNT 1], so the document still reads coherently and the model can reason about it. You map the answer back to the real names yourself afterward. KeptPDF has an anonymize mode built for exactly this.

How to do it in KeptPDF, step by step

  1. Open your PDF in the redact tool. It loads in your browser, and nothing is uploaded.
  2. Auto-detect the personal data. Confirm the names, numbers, and dates it finds, add anything specific to your document, and choose whether to black it out or replace it with placeholders.
  3. Export the clean copy and use that with ChatGPT, Claude, or any assistant. The original never left your device.

The point most people miss

Cleaning the file in a cloud tool, just to then feed a cloud AI, defeats the purpose: you have uploaded the sensitive document to one more place on the way. The cleanup has to happen on your device. KeptPDF runs in your browser, so the document is sanitized before anything leaves, and you can prove it by running the tool in Airplane Mode.

FAQ

Is it safe to upload documents to ChatGPT or Claude?
For non-sensitive content, generally yes. For confidential, regulated, or privileged material, remove the identifying details first. Depending on the product and your settings, uploaded content may be retained or used to improve models, so treat anything you send as no longer fully under your control.
How do I remove names from a PDF before using AI?
Use a redaction or anonymize tool that runs on your device. KeptPDF auto-detects names, numbers, and dates, and can either black them out or replace them with consistent placeholders so the document still makes sense to the model.
What is the difference between redacting and anonymizing for AI?
Redacting permanently removes the sensitive content. Anonymizing replaces it with placeholders like [PERSON 1], so the text stays coherent and you can map the AI's answer back to the real names afterward. Both keep the identifying data out of the model.
Does KeptPDF send my document to its servers?
No. It runs in your browser, so the file is cleaned locally before anything reaches an AI tool. You can confirm it by going offline and watching the tool still work.

Clean a document before it reaches a model, on your device.

Redact for AI