Guide · Security & privacy
How to use AI on confidential documents
Reviewed
Short answer
To use AI on confidential documents, first check you're allowed to send the text to a third-party processor, then share as little as possible — redact personal data and extract only the pages you need — and choose processing that matches the risk. In PDFBetter, browser tools never upload the file, and Private AI mode (Settings → Data & privacy) keeps no text index, embeddings or chat history after each request and can send model calls to an endpoint your operator runs.
On this page (6)
1. Check what you're allowed to do
Confidentiality agreements, client terms and data-protection rules can forbid sending a document's text to an outside service. If they do, no setting changes that; keep to tools that run in your browser, or use private mode only where the operator has configured their own model endpoint.
1Your PDF
2Ask or summarize
3Answers that cite pages
2. Share less
Redact names, ID numbers, card numbers and IBANs the AI doesn't need, with AI redaction review or the manual Redact tool. Pull out only the relevant pages with Extract pages before you ask about them. A smaller, cleaner document is also cheaper and faster to process.
3. Turn on Private AI
- 1
Open Settings → Data & privacy
Find “AI processing mode”.
- 2
Choose Private AI
Switching deletes the stored AI index for every document you own right away. Your documents themselves aren't touched.
- 3
Read the endpoint line
It says whether model calls in private mode go to an operator-run endpoint or to the hosted AI provider with storage turned off.
- 4
Decide on chat history
Turn off “Save AI chat history” and use “Delete stored AI index now” whenever you want derived data gone.
What private mode does and doesn't change
Private mode controls what is kept after a request. Without an operator-run endpoint, document text still goes to the hosted AI provider (with storage turned off). Your uploaded files stay in your workspace until you delete them. Chat doesn't remember earlier questions, and search by meaning falls back to keyword search unless the private endpoint provides embeddings.
PDFBetter makes no compliance or certification claims for private mode and doesn't sign BAAs.
Tools used in this guide
- Chat with PDFUpload a PDF and ask questions in plain language; every answer points to the pages it is based on.
- AI redaction reviewUpload a PDF, review suggested sensitive text by category, apply real redaction to what you select, and see a check that it is gone from the text.
- Redact PDFMark what must go, apply redaction, and download a PDF where the marked content no longer exists in the file.
- Extract PDF pagesSelect specific pages and save them as a new PDF, without changing the original.
- AI PDF summarizerUpload a PDF, pick a length and optional focus, and get key points you can check against the cited pages.
- Protect PDFSet a password to open the file, optionally restrict printing, editing and copying, and download an encrypted PDF.
Questions people ask
Which PDFBetter tools never upload my file?
Browser tools such as merge, split, rotate, extract pages and images to PDF run on your device. Each tool page says where processing happens.
Are my documents used to train AI models?
PDFBetter doesn't train models on your documents, and the hosted provider's API terms by default exclude API data from training.
Does private mode cost more?
No. AI actions cost the same credits; private mode re-reads the document on each request instead of reusing a stored index.
Can my company run its own model?
The operator of a PDFBetter deployment can configure a private, OpenAI-compatible endpoint such as a self-hosted vLLM or Ollama server. Features that send page images need that endpoint to accept images.