How to Redact a PDF Before Uploading It to ChatGPT
How to redact a PDF before uploading it to ChatGPT or any AI: what to remove, step by step in Yantra Privacy, Acrobat or Preview, and the checks to run first.
Yantra Team6 min read
On this page
To redact a PDF before uploading it to ChatGPT, open it in a real redaction tool, mark the names, addresses, account numbers and other details that should not leave your hands, apply the redaction so the text is removed rather than covered, and upload the redacted copy instead of the original. Then search the copy for a name you removed to make sure it is really gone.
This guide shows how to redact a PDF with three tools: Yantra Privacy, which redacts on your device and can put the real values back into the AI's answer, Adobe Acrobat Pro, and Preview on a Mac.
Why a black box is not redaction
Drawing a black rectangle or highlighting text in black changes how a page looks, not what the file contains. The words are still in the PDF's text layer, so anyone (or any AI) that reads the file can still see them. Adobe makes the same point: a black marker or a changed font color is not foolproof. Real redaction deletes the text from the file.
What to remove before you upload
Go through the document with this list:
- People and places: names, street addresses.
- Contact details: email addresses, phone numbers.
- Identifiers: social security numbers, dates of birth, account numbers, policy or patient IDs, and medical identifiers such as NPI or DEA numbers.
- Your own terms: a client's name, your employer, a project codename, a deal value.
The AI usually does not need any of these to answer your question. "Is this termination clause standard?" reads the same with [PERSON_1] in place of a name.
How to redact a PDF, step by step
Step 1: Pick a tool that removes text
Use a tool with a redaction feature, not an annotation pen. Three options:
- Yantra Privacy: pick Privacy in the apps launcher. A Runs on this device badge sits at the top: the document is read and redacted in the page, and nothing about it is uploaded.
- Adobe Acrobat Pro: open the PDF and choose Tools > Redact, or Edit > Redact Text & Images.
- Preview on a Mac: use the Redaction Selection tool in the Markup toolbar (Apple's guide). Work on a duplicate, because Apple notes the redaction becomes permanent once you close the document.
Step 2: Add the document
In Privacy, Drop a document here or Choose a file. PDF, Word, CSV and text files are opened by the page itself, and redaction starts as soon as the document is in. If you want to see how it works first, choose Try a sample.
In Acrobat or Preview, the PDF is already open from Step 1.
Step 3: Choose what to redact
In Privacy, What to redact is grouped into People and places, Contact details and Identifiers, and each toggle shows how many it found. Under Your terms, add anything else that should never leave, such as a client name. Your terms are redacted wherever they appear, in every document, and Privacy suggests terms it found in the file (like the parties to a contract) that you can add in one click.
In Acrobat, double-click a word or drag across text to mark it. To catch every instance of a name, use Find Text & Redact and search for it. In Preview, select the text with Redaction Selection.
Step 4: Check what was missed
Automatic detection is a strong first pass, not a guarantee. In Privacy, a box lists anything risky to check before you share, such as a value that may have been cut rather than removed. Always redact removes it everywhere and runs again. The Copy tab shows the result, with Show originals to compare, and Replacements lists every value and what replaced it.
In Acrobat, click Apply, and in the Apply Redactions dialog leave the option to sanitize and remove hidden information on. That also clears hidden data such as metadata.
Step 5: Save the AI-safe copy
In Privacy, choose how replaced values read: Labels (bracketed tags like [PERSON_1]) or Stand-ins (realistic fake values that read naturally). Then Download AI-safe copy or Copy text.
In Acrobat, save under a new name so the original is not overwritten; Acrobat adds "_Redacted" to the filename by default. In Preview, save the duplicate.
Step 6: Test the copy, then upload it
Open the redacted copy, search for a name you removed, and try copying text from a redacted area. If anything turns up, go back to Step 3. When it is clean, upload the copy to ChatGPT, Claude, Gemini or any other assistant.
For extra caution in ChatGPT, you can turn off Improve the model for everyone under Settings > Data controls so your chats are not used to train OpenAI's models. OpenAI's data controls FAQ covers the details, including Temporary Chats and business plans.
Step 7: Put the real values back in the answer
The AI's reply will refer to [PERSON_1] or your stand-ins. In Privacy, open Restore reply and paste what the AI wrote. Every label and stand-in is swapped back for the real value on your device, and Copy restored text gives you the finished answer. With other tools, you replace the placeholders by hand.
Skip the steps for everyday files
If you attach documents in the Yantra chat often, the composer has a Privacy switch. With it on, documents you attach are redacted on your device and only the AI-safe copy is sent, and the setting is kept on your account. See attachments. For a contract, Contract Review runs the same kind of redaction before anything is sent; the contract workflow guide shows where it fits.
Common mistakes
- Covering instead of removing. Black boxes, highlights and white text leave the words in the file.
- Forgetting scans and images. Text-based redaction only sees the text layer. A scanned page or photo has no text for the browser to read; the Yantra desktop app can read it, or retype the parts you need.
- Trusting detection blindly. In a browser, Privacy's engine is pattern-based: it catches named parties, mailing names and addresses, titled names, your terms and personal columns in tables, but not every stray mention. On the desktop app, a model running on your machine finds names and addresses. Always read the warnings and the replacements list.
- Leaving hidden data. Metadata, comments and attachments can carry names. Use Acrobat's sanitize option, or send redacted text rather than the original file.
- Overwriting the original. Redaction cannot be undone on the saved file, so keep the unredacted version somewhere safe.
- Missing the table of contents. In Preview, table of contents entries that link to redacted text are not removed.
FAQ
Is it safe to upload a PDF to ChatGPT?
It is as safe as what is inside it. Anything in the file reaches OpenAI's servers, so remove personal and confidential details first, and review your data controls. A redacted copy lets you get the same help without sharing the details.
Does ChatGPT train on files I upload?
On personal plans, conversations can be used to improve OpenAI's models unless you turn off Improve the model for everyone in Settings > Data controls. OpenAI says it does not train on content from ChatGPT Business, Enterprise or Edu workspaces by default. Redacting first means the details never reach the model either way.
How do I redact a PDF for free?
On a Mac, Preview's Redaction Selection tool is built in. Yantra Privacy is included in Yantra plans and runs in the browser on your device. Avoid free websites that ask you to upload the document, since the point is not to send it anywhere first.
Can I redact a scanned PDF?
Not with tools that read only the text layer, because a scan is an image. Use the Yantra desktop app, which can read scans and photos, or an OCR tool that runs on your own machine before redacting.
Will ChatGPT still understand a redacted document?
Yes, for most questions. Labels like [PERSON_1] keep the structure clear, and stand-ins read like a normal document. Redact identities and identifiers, not the facts the question depends on.
For a side-by-side look at more tools, see the best document redaction tools for AI privacy, and the Privacy walkthrough for every screen in Yantra.