Remove personal data before pasting into an AI chatbot
Summarising a contract, rewriting a complaint or analysing a spreadsheet with an AI chatbot means handing it the whole document. Depending on the service and your settings, what you paste may be kept in your history, reviewed or used for training. Your employer's policy may also forbid sharing customer data with outside tools. Redacting first lets you get the help without passing the personal details on.
Labels keep the meaning
A chatbot answers better when it can tell people and things apart. "Replace with label" turns "Anna Novak emailed anna@example.com" into "[NAME] emailed [EMAIL]", so the sentence still makes sense and the model can still reason about it. Black bars make the text harder for the model to follow.
Steps
- Paste the text into the text tool, or open the file in the tool for its format (PDF, Word, Excel and the rest).
- Allow the scan and check the Found list.
- Add company terms under "Always redact": client names, project code names, internal host names.
- Choose "Replace with label" and download the result.
- Paste the redacted copy into the chatbot.
If the answer comes back with labels in it, put the real names back yourself in your own copy. Keep the original file: the redacted one cannot be turned back.
Detection uses AI too
To find what to remove, AnyRedact sends the text it reads from your file to its own server, where an AI model running on Cloudflare marks the personal details. Nothing is sent until you agree, the text is not stored or used for training, and you can decline and mark everything by hand. The privacy policy has the details.
Worth removing even when it seems harmless
- Signatures at the bottom of emails, with phone numbers and job titles.
- Customer and order IDs, which can be looked up in your systems.
- File paths and URLs that contain user names.
- Health, salary and disciplinary details about named staff.