Hidden Tokens vs manual redaction before AI
Manual redaction is free, familiar and fully under your control. It is also slow, easy to get wrong, and it gives the AI a document it can no longer reason about.
- Manual redaction is fine for an occasional short document with a handful of names.
- It routinely misses hidden content: tracked changes, comments, hidden sheets, pivot caches, metadata. Hidden Tokens strips comments, revisions and metadata from Word copies and checks the result before writing it.
- Black boxes destroy the structure the AI needs; consistent codes keep it.
- Manual redaction is one-way: someone has to put every name back by hand. Hidden Tokens restores them automatically, on your computer.
Side by side
| Manual redaction | Hidden Tokens | |
|---|---|---|
| Time for a 40-document case file | Hours | Seconds per document, in the background |
| Hidden content (tracked changes, comments, hidden sheets, metadata) | Only if you remember to inspect each file | In Word, the coded copy drops comments, tracked changes, metadata and embedded objects, and is not written at all if a known value remains anywhere in the file; file names are coded too |
| Name variants ("Ms Whitfield", "S. Whitfield", email addresses) | Easy to miss | Detected and given the same code |
| Can the AI tell people apart? | No: everyone becomes "[REDACTED]" | Yes: NOM-001, NOM-002… consistent across the case |
| Amounts in spreadsheets | Deleting them breaks formulas | Readable, placeholder values or secret factor; restored afterwards |
| Getting the names back into the AI's output | By hand | Automatic, in Word or Excel, on your computer |
| Proof of what was sent | None | Local log and signed proof pages |
| Cost | Free, apart from time | €190 to €390 per seat per year; 14-day free trial |
When manual redaction is enough
If you use AI occasionally, on short documents with a few names and no spreadsheets, a careful manual pass is a reasonable choice. Follow a checklist like the one in how to redact a Word or Excel file before an AI reads it.
When it stops being enough
- Case files with many documents and the same people throughout.
- Spreadsheets where figures matter.
- Several people working on the same clients, who need the same codes.
- A client, a DPO or a regulator who may ask what the AI received.
Frequently asked questions
Is manual redaction safer because a human checks everything?
A careful human is very good on one page and much less reliable on the fortieth. Typical misses are name variants, email addresses that contain a name, and content that isn't visible on screen. Hidden Tokens shows a plan you review, then codes every document the same way.
Can I combine both?
Yes. Hidden Tokens shows its plan before coding, and you can add or remove values. For very sensitive matters, review the coded copy before sending it: it is quicker than redacting from scratch.