Introducing Anonymizator: protect personal data before AI

The interview is over. You have a page of notes and need a clear summary for the hiring manager. You want to ask AI to organise them.

The notes include the candidate's name and contact details, their current employer, salary expectations and an internal reference. They also contain what you actually want help with: making the conversation easier to follow without losing or inventing anything.

Before those notes become a prompt, you have a decision to make about each private detail they contain.

Anonymizator helps with that preparation. You use it to replace supported private IDs with placeholders, review what remains and then share the protected version with the service you have chosen, which can be any AI, including ChatGPT.

The web app takes pasted text, Word documents, Excel workbooks and text files, one file or the whole folder. The Chrome and Edge extension handles text you copy or select, in a side panel beside the page you are working on.

AI is becoming part of ordinary work

PwC's Global Workforce Hopes and Fears Survey 2026, released on 29 September, found that 64% of respondents had used AI at work in the previous 12 months; 22% used generative AI daily. The survey collected responses from 49,364 workers across 48 countries and regions in May and June 2026. These are self-reported usage figures, not a measure of data exposure. PwC's 2026 survey and methodology

When AI becomes part of drafting correspondence, preparing reports or organising notes, the decision about sharing data comes up during the work itself. A useful prompt often needs context. Copying the source material is an easy way to provide it, and the private details come along with the useful parts.

That is why the moment before sharing matters. Once you submit the text, the service has received it. Asking that same service to remove personal details afterwards means giving it the details first.

Everyone has the right to privacy.

The people described in your work may never use the AI service themselves. They may be candidates, clients, colleagues or people who spoke to you in confidence. Their information can become part of a prompt because somebody else wants help with a task.

France's data protection authority, CNIL, advises users to submit only information they are authorised to share and warns against providing confidential information to consumer AI services. Choosing an appropriate service and following your organisation's rules comes before deciding how to prepare the text. CNIL's guidance on generative AI

The European Commission also explains data minimisation as limiting personal data to what the purpose requires. In practice, this means asking which details contribute to the task and leaving out those that do not. European Commission guidance on data minimisation

What Anonymizator helps you protect

Anonymizator replaces private details, such as names or bank account numbers, with placeholders before you share text with any third party.

In a working document, those details rarely arrive on their own. Names appear beside email addresses and phone numbers. A client record may contain an address, an account number, a payment amount and a reference to a case. Interview notes can tie somebody to an employer, a salary expectation or a personal circumstance.

Some details are covered by the enabled Predefined IDs. Others need your input. You can add an internal reference or another private detail to Current IDs yourself, so Anonymizator uses a placeholder for it too. The Additionally protect options let you include all numbers, all currency values, or company names with a legal suffix.

This is why review is part of the process. You know what a project name means inside your organisation and whether a combination of details would make someone recognisable. Anonymizator does not find every private detail or decide whether the remaining text is appropriate to share.

From interview notes to Protected text

This fictional prompt asks AI to organise interview notes for a hiring manager to check.

You prepare this in the Original text, in the app or the extension, before pasting anything into AI:

Turn these interview notes into a factual summary for the hiring manager.
Use these sections: experience, availability, salary expectations and
follow-up questions. Do not rank the candidate, infer personal traits or
add information that is not in the notes.

Candidate: Jane Doe
Email: jane.doe@example.com
Current employer: Example Company d.o.o.
Candidate reference: KAND-042

Notes:
Jane Doe has four years of customer-support experience and now leads a
small team. She can start one month after accepting an offer. Her salary
expectation is EUR 2,800 gross per month. She would like to know how often
the role involves travel. Send follow-up questions to jane.doe@example.com
and include KAND-042 in the subject line.

For this illustration, you select All currency values and Company names (with legal suffixes only) under Additionally protect. You start Protect and review the result, adding the candidate reference to Current IDs so it is protected too. The Protected text could then read:

Turn these interview notes into a factual summary for the hiring manager.
Use these sections: experience, availability, salary expectations and
follow-up questions. Do not rank the candidate, infer personal traits or
add information that is not in the notes.

Candidate: [PERSON_a7k2q]
Email: [EMAIL_v4n8r]
Current employer: [COMPANY_c6w3s]
Candidate reference: [CUSTOM_m3v9d]

Notes:
[PERSON_a7k2q] has four years of customer-support experience and now leads a
small team. She can start one month after accepting an offer. Her salary
expectation is [AMOUNT_p2z6h] gross per month. She would like to know how often
the role involves travel. Send follow-up questions to [EMAIL_v4n8r]
and include [CUSTOM_m3v9d] in the subject line.

The same person and email address use the same placeholders throughout. The instructions and the relationships between the details remain readable. The example also shows the difference between finding supported IDs and adding something specific to your work during review.

You still need to check the remaining notes before copying them. If the answer uses the placeholders, Reveal can bring back the corresponding details using Current IDs. Export IDs lets you keep an ID file for later use with Load IDs. Keep that file private and separate from the Protected text.

When the notes are a folder of files

Interview notes are one passage. Often the work is a folder: contracts, a spreadsheet of invoices, meeting notes. The web app reads that folder without changing it and saves protected copies into a second folder you choose. You can protect one file or all the supported ones, then review each copy and upload the ones you decide to share.

It works with DOCX, XLSX and text files, up to 100 MB per file. PDFs, pictures and scans are not supported. The folder guide goes through it step by step.

Reversible protection still needs judgement

By default, this process is reversible pseudonymization. Current IDs or an exported ID file holds the information needed to bring the private details back.

It does not make the text anonymous. Even the recruitment example leaves experience, availability and a description of the person's current responsibilities. In a small professional circle, that combination could be revealing. Depending on the task, you may need to generalise or remove more of the context, or decide against using an external service.

The European Commission explains that information which can identify a person when combined is personal data, and that pseudonymised information remains personal data when it can be used to re-identify someone. Replacing direct IDs does not settle the question of identification or establish GDPR compliance. European Commission explanation of personal data

In the hosted service, Protect sends text encrypted to Anonymizator, where it is decrypted to find IDs. The service does not retain the text or the detection results. Placeholders, Current IDs and Reveal stay in your browser.

For work that involves other people's information

Recruitment is one example. Insurance and property advisers work from client conversations. Lawyers, social workers and counsellors handle accounts of people's private lives. Researchers and journalists may need to protect the people behind their notes. Supervisory-board work also brings personal data together with confidential business material, which can require controls beyond replacing private IDs.

The appropriate use depends on the task and the organisation's rules. Anonymizator gives you a way to prepare files and text within that workflow: start Protect manually, review what remains and choose what to share. It does not protect a message when you press Send, or a file you attach directly to an AI service. Prepare the copy first and upload that.

To try Protect, review and Reveal on the documents and text you work with, start a 7-day free trial.