The chatbot is useful and the contract is confidential: prepare the document first
Before pasting a contract or statement into an AI chatbot: redact what the model doesn't need, on your device — then share the cleaned copy instead.
UnboundPDF is a free suite of 43 PDF and image tools that run entirely in your browser — merge, split, compress, edit text, OCR in 126 languages, redact, sign, convert and archive to PDF/A. Your document is read and written by the page on your own device; there is no document-upload endpoint in the core tools, no account, no watermark and no daily cap. Every result can be checked — with the Network tab, or with the Document Passport the Workspace writes for a chain of steps.
The habit. Work on a copy: redact what the AI doesn't need for the task, verify, then paste from the cleaned copy. The preparation runs in your browser, so nothing leaves your device until you decide what does.
The mismatch nobody prices in
AI assistants are genuinely good at documents — summarise this lease, explain this clause, check this invoice — which is exactly why whole confidential files get pasted into them daily. The mismatch: the model needs the structure and substance of the document to help you, but it rarely needs the identifiers — the account numbers, the addresses, the third parties. Those travel along only because separating them used to be work.
Make the separation cheap
It is a two-minute pass now. "Explain what this contract commits me to" works identically with the names redacted. "Why did my balance drop in March" needs the transactions, not the account number. The statement walkthrough maps what to remove for which audience — an AI service is just one more audience, with the longest memory and the least-knowable retention.
The rule that outlives every policy change
Service terms differ between providers, tiers and years; some offer training opt-outs, some retain for review, some change quietly. You can track all that — or apply the rule that doesn't require tracking: what the model never receives, no terms have to protect. Redaction before pasting is that rule, made mechanical.
When you need the text itself
For long documents, pasting page after page mangles tables and loses structure. Extract to Markdown first — clean reading order, headings intact — then redact the extracted text before it goes into the prompt. Same discipline, better input.
Troubleshooting
I already pasted the raw document. Check the service's data controls for deletion options, then adopt the habit going forward — it only protects the next document. The document is a scan. OCR first so redaction's text search can find every instance. Too many identifiers to mark by hand? Statements repeat them in headers — page through once; repetition makes marking faster, not slower.
Frequently asked questions
Why prepare at all — isn't the AI private?
What you paste leaves your machine and is handled under that service's terms, which vary and change. The reliable rule is simpler: what the model never receives, no terms have to protect.
What should come out before pasting?
What the task doesn't need: names and account numbers for a summarising task, the counterparty's details for a clause explanation, anything about third parties.
Does the preparation itself leak the document?
No — redaction and extraction here run in your browser; the document stays on your device until you choose what to paste.
Try it yourself
Free, private, no account. Runs entirely in your browser.
