User Guide · Redact
A step-by-step walkthrough of every control — from choosing your engine and region to reading the redacted output, downloading as Word, and saving to My Docs.
In this guide
The engine controls the AI model used for the LLM redaction sweep. Both engines use Azure OpenAI (West Europe). gpt-4o-mini costs 1 credit and handles most documents well; gpt-4o costs 2 credits and is better for dense, complex, or multi-person documents.
| Engine | Cost | Best for |
|---|---|---|
| Azure gpt-4o-mini ★ | 1 credit | Default. Everyday redaction, single-person documents, standard case notes. |
| Azure gpt-4o | 2 credits | Complex documents with many named parties, overlapping roles, or poor formatting. |
Mode controls how aggressively the tool redacts. Found in the Advanced settings panel.
| Mode | What it does |
|---|---|
| Standard ★ | Pattern matching + LLM sweep for known entity types (names, orgs, places, ID numbers). Balanced accuracy and precision. |
| Strict | Also replaces any capitalised two-word phrase as a potential name. More aggressive — use when you need certainty over precision. May produce false positives in formal text. |
Region configures which identifier patterns the regex first-pass targets. The LLM pass always runs regardless of region.
| Region | Targets |
|---|---|
| Nordic ★ | Norwegian fødselsnummer, D-number, +47 phone, email, Norwegian addresses |
| European | Adds IBAN, Swedish personnummer, Danish CPR, Finnish HETU, UK NI |
| ECHR | Adds ECHR application numbers, DOB phrases, ECtHR case references |
| Global | Adds US SSN, driver's licence formats, generic document numbers |
Four checkboxes toggle which entity types the LLM looks for. All are checked by default.
| Entity type | What is replaced |
|---|---|
| Names ★ | Personal names — first name, surname, or full name of any individual |
| Organisations ★ | Organisations — companies, agencies, authorities, institutions |
| Places ★ | Places — cities, streets, addresses, countries, regions |
| Dates ★ | Dates — dates of birth, age references, specific personal dates |
Default: OFF (unchecked). When unchecked, all names — including judges, experts, social workers, and lawyers — are replaced with generic role tags like [PERSON] or contextual tags like [JUDGE].
When to check: tick "Keep official names (judges, experts)" to preserve the names of named officials in a labelled tag: [JUDGE: Andersen], [EXPERT WITNESS: Dr. Hansen]. Use this when the official's identity is legally relevant and must remain visible in the redacted version — for example, when submitting to a court that knows the judge.
Three options control how redacted content is represented in the output.
| Format | Result |
|---|---|
| Contextual tags ★ | Each person gets a descriptive role tag. [FATHER], [MOTHER], [JUDGE: Name if officials kept]. Narrative remains readable. Best for most legal filings. |
| Generic tags | All persons become [PERSON], organisations [ORG], places [PLACE]. Maximum anonymisation with no role context. |
| Pseudonyms | Names replaced with plausible Norwegian substitutes (Ola Nordmann, Kari Hansen). Phone numbers rewritten. Addresses replaced with fictional Norwegian addresses. Best when a human-readable document is required. |
Names listed here are never redacted — even if the AI would otherwise remove them. Use this for names that must remain visible in the output, such as a public figure, a quoted source, or an institution name that is part of the case reference.
Add each name on its own row. The match is case-insensitive. Partial matches are not supported — enter the name exactly as it appears in the document.
Replace a specific name with a custom bracketed label. For example: "David Jr" → [Junior], or "Kari Pettersen" → [SISTER]. Useful when the document uses a name that the LLM might not recognise as a person, or when you want a specific label instead of the AI's default choice.
Aliases are applied as a final substitution step after both redaction passes complete. They override whatever the LLM assigned to that name.
Drag a file onto the upload zone or click browse. Supported formats: PDF, DOCX, TXT. One file per run.
You can also paste text directly into the main text area — or combine a file and pasted text. All sources are processed together as a single input.
Files are extracted to text in memory. The redacted result is not retained on the server unless you choose to save it to Min Sak, your corpus, or download it.
After the tool runs, the redacted text appears in the results section. What you can do from here:
Free for Do Better Norge members.