Straight answers to the questions that decide whether this tool is right for you. If something is missing, the contact page explains how to ask.
Questions and answers
Is PromptScrubber really free?
Yes. Detection, anonymization, restore, copy and TXT export are free and require no account. There is no paywall in front of a feature you can see on the site, and no trial countdown. If a paid tier is ever introduced, the free text features stay free.
Are my documents uploaded to a server?
No. Files are opened with the browser's own FileReader and processed in the page. There is no backend endpoint that could receive them, and the page ships a Content Security Policy with connect-src 'none', which makes outbound requests fail even if a bug tried to make one.
Is PromptScrubber GDPR compliant?
It is built to minimise data: no account, no upload, no analytics, no advertising, no server-side document storage. That is privacy by design, but it is not a certification and we do not claim to be 'GDPR certified'. Whether your specific use of the tool is lawful depends on your role, your legal basis and the data you process — the Privacy Policy describes exactly what happens to the data so you can assess it.
Which personal data can it detect?
Email addresses, phone numbers, IBANs from the ISO 13616 registry (with mod-97 checksum validation), Italian tax codes and VAT numbers, payment cards that pass the Luhn test, IPv4 and IPv6 addresses, URLs including sensitive query parameters, API keys and access tokens when their format or context is recognisable, dates of birth when the context is explicit, and person or company names via local dictionaries and heuristics. Anything else can be marked manually by selecting it in the editor.
Does it work with ChatGPT, Claude and Gemini?
Yes. The output is ordinary text with placeholders such as [PERSON_1], so it can be pasted into any assistant, any chat interface or any document. PromptScrubber has no integration with those services and never connects to them.
Can I get my original data back after anonymizing?
Yes, as long as the tab is open: the mapping is kept in the memory of the page. Paste the AI answer into the restore panel and the placeholders are swapped back. You can also export the mapping as a JSON file to restore later — that file contains the original values, so treat it like the original document.
Do you keep logs of what I paste?
No. The application has no logging endpoint and no storage for document content. The only thing kept locally is a theme preference, a detection preference and your consent choice. Your hosting provider may keep standard technical logs (IP address, requested URL, timestamp) like any web server; the Privacy Policy describes that honestly.
Do I need an account or an installation?
No account, no extension, no installation. It is a normal web page that runs in the browser. Once loaded, the interface keeps working without a network connection because there is nothing to fetch.
Which file formats are supported?
TXT and other plain-text formats, PDF with a selectable text layer, and DOCX. Scanned PDFs are refused because OCR is not implemented. Legacy .doc and .rtf files must be saved as .docx or .txt first. PDF and DOCX are read as text, so the anonymized output is a .txt file, not a redacted copy of the original layout.
How accurate is the detection?
Structured identifiers — email, IBAN, card, tax code, IP — are reliable because they are validated by pattern and checksum. Names, companies and dates of birth are heuristic and can be missed. The tool therefore asks you to review the highlighted preview and lets you add anything it missed manually. No automated redaction should be trusted without a human check.
What happens if a placeholder is missing from the mapping?
It is left unchanged in the restored text and listed in the notice under the output. The tool never invents a replacement. This usually means the AI introduced a new placeholder, or that the mapping was cleared by reloading the tab.
Can I use it for work documents?
You may use it for professional purposes, but you remain responsible for what you share and for your organisation's rules. Special categories of data (health, biometrics, judicial data) deserve extra care, and residual indirect identifiers can survive anonymization. Check the Terms of Service and the Disclaimer.
Does the site use cookies or analytics?
No cookies are set by the application, and no analytics, advertising or third-party embed is loaded. The consent manager exists and is reachable from the footer, but it lists the optional categories as 'Not in use' because there is nothing to switch on. If an optional tool is added later, the banner appears and nothing loads before consent.
Is the tool open to inspection?
The application is plain JavaScript and CSS with no build step, so the behaviour described on this site can be verified by reading the files. That is deliberate: 'trust us' is not a privacy guarantee.
Task-specific guides
For a closer look at a particular situation, these guides apply the same local workflow to text, email, documents, client material and AI prompts. They also spell out where review and responsibility remain with you.
- Remove personal data from text
- Anonymize text before ChatGPT
- Redact personal information
- Use a private text anonymizer
- Anonymize client data
- Anonymize emails
- Anonymize documents
- Protect sensitive data in AI prompts
Still unsure?
Read the technical walkthrough, the disclaimer about the limits of automatic redaction, or the privacy policy for the data flow in detail.