Skip to content
PrivacyShare

Free text anonymizer · no upload · no account

Anonymize text before it reaches ChatGPT, Claude, Gemini and other AI models

Paste a text – names, email addresses, phone numbers, IBANs, addresses and IDs become consistent placeholders. Copy it into the AI of your choice, paste the answer back, and the original values return. Everything happens in your browser.

Nothing leaves your browser Restores the AI answer Open source (MIT)

Step 1 · Paste your text

0 / 100.000
Detect

Honest limits

This tool finds the obvious identifiers. It does not understand context: a person can still be recognizable from a job title, a place or a story. Names are detected by salutations, signatures and a first-name list – unusual names or names without such anchors can be missed. Read the result before you send it, and add anything missing manually.

3 texts anonymized so far – counted without content or identifier

Nothing leaves your browser

Detection, replacement and restoration run as JavaScript on your device. Your text is not sent to any server – the only request on use is an anonymous counter ping without content, which you can verify in the network tab of your browser.

Reverse mapping

Placeholders like [PERSON_1] stay consistent across the whole text. Paste the AI answer back and the real names, numbers and addresses are restored – locally.

No model, no download

Patterns, check digits and word lists instead of a machine-learning model. The page loads instantly, works offline once opened, and behaves predictably.

How it works

Three steps – from raw text to a safe prompt and back.

1. Paste & anonymize

Paste your text and click "Anonymize". Personal data is detected and replaced with consistent placeholders such as [PERSON_1] or [IBAN_1].

2. Review & copy

Check the highlighted matches, switch individual ones off, add missing terms – then copy the anonymized text into the AI model of your choice.

3. Restore the answer

Paste the AI answer back. The placeholders are replaced with the original values from the mapping table that never left your browser.

What is detected

Reliable where a checksum or a strict format exists – heuristic where language is involved.

Verified by checksum

IBAN (MOD-97 check), credit card numbers (Luhn), German tax identification numbers (check digit), German social insurance numbers (format).

Strict patterns

Email addresses, URLs, IPv4/IPv6 addresses, phone numbers (German and international formats), dates (12.03.2026, 2026-03-12, 3 March 2026), German street addresses and postal code + city.

Heuristic – please check

Names via salutations ("Dear Ms. …", "Herr Dr. …"), role and kinship words ("our colleague …", "CEO …", "his niece …"), signatures, field labels, a German/English first-name list and first-name/surname pairs mid-sentence (with filters against ordinary noun pairs such as "Deutsche Bahn"); companies via legal forms (GmbH, AG, Ltd., Inc.); German license plates. Once a name is found, its surname and genitive form ("Hitlers") are masked everywhere.

Not detected

Names without any anchor and outside the list, nicknames, indirect identifiers ("the CFO of our Munich office"), non-Latin scripts, content of images or attachments. Add such terms manually.

Typical use cases

Whenever a text is useful for an AI but must not carry real people into it.

Customer emails & support tickets

Let an AI draft a reply or summarize a thread – without handing over the customer's name, address, contract or account numbers.

Contracts & legal texts

Have clauses explained or compared while the parties, IBANs and dates stay on your side.

Applications & HR documents

Structure a CV, prepare interview questions or check a reference letter without exposing the applicant.

Meeting notes & medical letters

Summarize notes or explain a doctor's letter – names and dates of birth become placeholders. Health data deserves extra care (Art. 9 GDPR).

Why anonymize before using AI?

Every text you paste into ChatGPT, Claude, Gemini or another AI model is transmitted to that provider. If it contains personal data – a customer's name, an applicant's address, a patient's date of birth – that is a processing operation under the GDPR, and depending on the provider and settings the data may be stored or used for training. Replacing the identifiers with placeholders before pasting removes the most obvious personal data from the prompt while keeping the text useful. The AI works with [PERSON_1] just as well as with a real name – and you get the real name back at the end.

Anonymization or pseudonymization?

Strictly speaking this is pseudonymization (Art. 4 No. 5 GDPR) from your side: you keep the mapping table and can restore the values. The AI provider never receives that table, so the text it sees contains no direct identifiers. Remaining risk: context can still make a person recognizable – read the result before sending.

Verifiable, not just promised

The detection engine is open source (MIT). Anyone can read how the patterns work, run it locally or improve it. Combined with the network tab – where a click shows only one content-free counter request – "nothing leaves your browser" is something you can check rather than trust.

View the engine source code (MIT)
100% client-side
No upload, no logging
Open source (MIT)
Works offline
Free, no account

This page describes a technical safeguard and is not legal advice. Whether and how personal data may be processed with AI tools depends on your specific situation – ask your data protection officer or a lawyer.

Made by difdele – web design with privacy built in.

Frequently asked questions

About AI tools, personal data and what this anonymizer can do.

Can I paste customer data into ChatGPT, Claude or Gemini?
Transmitting personal data to an AI provider is a processing operation under the GDPR. Whether it is permitted depends on the legal basis, the contract with the provider (data processing agreement), the provider's settings (for example training opt-out) and the purpose. Many companies allow AI tools only without personal data. Replacing names, contact details and identifiers with placeholders before pasting removes the most obvious identifiers – it is a technical safeguard, not a legal clearance. This is not legal advice; ask your data protection officer if in doubt.
Is this anonymization or pseudonymization?
From your perspective it is pseudonymization (Art. 4 No. 5 GDPR): you keep the mapping table and can restore the values. The AI provider never sees that table, so the text it receives contains no direct identifiers. Context can still identify a person ("the mayor of [CITY_1]"), so review the text before sending.
Does my text really never leave my browser?
Yes. Detection, replacement and restoration run as JavaScript in your browser. When you click "Anonymize", the page sends exactly one request to our server: an anonymous counter for the transparency page, without text and without any identifier – you can verify this in the network tab of your browser's developer tools. The engine is open source under the MIT license.
How reliable is the detection?
Structured data is reliable: IBANs are validated with the MOD-97 check, German tax IDs with their check digit, card numbers with the Luhn algorithm; emails, URLs, IP addresses, dates and phone numbers follow strict patterns. Names, companies and addresses are heuristic – salutations, signatures, a first-name list, street patterns – and are marked "please check". There is deliberately no AI model in the browser: no download, works offline, predictable behaviour.
What does "restore the AI answer" do?
The tool keeps a small mapping table in memory: [PERSON_1] = Anna Schmidt, [IBAN_1] = DE89 … When you paste the AI answer back, every placeholder is replaced with its original value. If the same name appeared in several forms (full name, surname only, genitive), the restored text uses the complete form. The table never leaves your browser and is deleted when you close or reload the tab.
Why placeholders like [PERSON_1] instead of fake names?
AI models keep bracketed placeholders stable and do not draw wrong conclusions from invented names (a fake "Mr. Miller" might get a wrong gender or nationality). Numbered placeholders also stay consistent across the text, which is what makes restoring possible.
What if the AI changes the placeholders?
Small variations – PERSON_1 without brackets, [Person 1], {PERSON_1} – are restored anyway. If a model rewrote a placeholder beyond that, the tool tells you how many could not be matched and you fix them by hand.
Does it work offline and on my phone?
Yes. Once the page has loaded, no connection is needed. The tool works on phones and tablets; very long texts are simply easier to review on a larger screen.
What does it not detect?
Names without a salutation, signature or field label that are not in the first-name list; nicknames; indirect identifiers such as a job title plus a place; free-text descriptions; non-Latin scripts; anything inside images or attachments. Add such terms with "Add a term manually" and read the result before sending it.
Is it free, and do I need an account?
Free, no account, no ads. The only limit is 100,000 characters per run. Clicking "Anonymize" only increments an anonymous counter (the request contains no text) that is shown on our transparency page; our cookieless visitor statistics do not run on this page. As on any website, opening the page creates a server log entry with a truncated IP address that is deleted after 7 days.
Can browser extensions, translators or the clipboard see my text?
Yes – no web tool can prevent that: anything installed in your browser – spelling and grammar extensions, translators, AI assistants – can read page content and send it to its own servers. We do what a page can do: the text areas are marked "do not translate" for browser and Google translators, the browser spell-checker (which may send text to cloud services) is switched off, common grammar extensions are told not to attach, and everything is wiped when you leave the page. Also note that copied text lands in the clipboard, which some systems sync between devices. For highly sensitive texts, use a browser profile without extensions.