How to clean documents before using ChatGPT or Claude

Banning AI at work rarely works; people use it anyway. The realistic goal is to let your team use AI tools without the client's identity going along for the ride.

The real risk isn't AI, it's the details

Pasting a document into a chatbot to summarise it or draft a reply is genuinely useful. The problem is what's in the document: names, addresses, account numbers, dates of birth. Once that text is sent to an external service, you no longer control where it's stored or how it's used. For regulated professions, that's the part that creates exposure.

Remove the identifiers, keep the usefulness

Most of the value an AI tool provides doesn't depend on knowing who the document is about. A summary, a rewrite or an analysis works just as well when "Marie-Claude Bélanger, SIN 046 454 286" has been removed. The trick is to strip the identifying details before the text ever reaches the AI, not after.

A simple workflow

1. Redact locally first. Open the document in an on-device tool and remove names, numbers, addresses and other personal details.
2. Review the result so nothing sensitive slips through.
3. Then use the AI tool on the clean version.
4. Follow your own policy. Redaction reduces risk, but what you're allowed to share is still governed by your firm's rules and your professional obligations.

Where Anonyma fits

Anonyma is built for exactly this step: it finds and removes the personal information in a document on your own computer, before anything is shared. You review every page, export a clean copy, and keep the original intact. Because the cleaning happens locally, the sensitive version is never uploaded anywhere, including to Anonyma.

Anonyma helps you share less, but you stay responsible for what you send and how it's used downstream.