AI redaction without uploading your document

People searching for AI redaction usually want one of two things: a tool that finds the personal information for them, or a way to take it out before a document goes into ChatGPT, Claude or Gemini. Both come down to the same question: where does the unredacted file go while it is being redacted?

Open the tool Free below twenty pages, in your browser. Nothing is uploaded.

Four headshots arranged in a block, each with a name under it, then the company that person works for and their job title. — A presenters slide before and after: the names under the headshots are replaced with consistent labels, and one headshot is covered.

Redaction that finds things for you

Blinded proposes what to remove without being told. The detectors find email addresses, phone numbers, web addresses, street addresses, and the names of people where their position gives them away: under a sign-off, beside a job title, in a contact block. Type the names you know as well, and each is found on every page, in the text and in the lettering of pictures and scans.

It does this without an AI model. The reading is text recognition and the finding is rules and pattern matching, all running in your browser. That is a choice, not a gap: an AI redaction service almost always means sending the document to a server where a model reads it, which is the thing you were trying to avoid.

What rules catch, and what they miss

Rules are predictable: an email address written out in the text is found, and the same document gives the same result every time. What they do not do is understand the text. A sentence such as “our largest client in Singapore” can identify somebody without containing a name, and no detector will know. So every proposal is shown before anything is covered, with a count per page, and you add what only a reader would notice.

Redacting before you use an AI chatbot

The safe order is to redact first, on your own machine, and give the chatbot only the redacted copy. Uploading the original and asking the model to ignore the names does not help: by then the document has been sent.

Turn on labels and each removed thing becomes a consistent code, [P1] for a person and [E1] for an email address, so the model can still follow who did what without knowing who anybody is. Export with searchable text on, and the chatbot receives text it can read rather than a stack of pictures.

It is not only PDFs. Plain text, Markdown, CSV and JSON files open too, and come back as text with the personal data replaced, ready to paste.

Nothing is uploaded, to a model or anywhere else

Blinded has no server. The document is opened, read and rebuilt in your browser tab, and the page is not permitted by the browser to send anything anywhere, so it cannot pass your file to a model even by mistake. You can check it: watch the network panel, or turn off the internet once the page has loaded, and it keeps working.

How to do it, in four steps

Open the document here. Nothing is uploaded, to a model or anywhere else.
Turn on the detectors, and type any names you know.
Check every proposal, and add anything only a reader would spot.
Turn on labels, export, and give the chatbot the redacted copy.

Who it is for

Anyone who wants personal data found for them without handing the document to an AI service, and anyone who uses ChatGPT, Claude or Gemini on work documents and has to keep names and contact details out of them.

Open the tool

Questions

Does Blinded use AI?

No AI model reads your document. Personal data is found by text recognition, rules and pattern matching, all running in your browser, and nothing is sent anywhere.

What does it find automatically?

Email addresses, phone numbers, web addresses, street addresses, and names of people shown by their position, such as under a sign-off or beside a job title. Anything else you type once and it is found on every page.

Will a chatbot still understand the redacted document?

With labels on, yes. Each removed thing keeps a consistent code, so references between people and clauses survive even though the identities do not.

Can I redact text before pasting it into ChatGPT?

Yes. Save it as a .txt file and open it here; the result is the same text with the personal data replaced, ready to paste.

Also

Back home