Anonymize data before Gemini: preparing a document for Google's assistant
To anonymize data before Gemini is to replace, on your computer and offline, the names, email addresses and numbers in a file with aliases before dropping it into Gemini, then to restore the real values in its answer. The approach follows the same rule as with other online assistants: the stage you keep full control of comes before sending, and that is where everything is decided. Gemini, developed by Google, is often used by teams already working in the Google environment, which makes uploading a file particularly fast: preparation deserves to be just as fast, and that is what Biffer makes possible.
Here we describe how Biffer prepares a batch of job applications, a business letter or a CSV export on your computer, offline, and how you then restore the real values in Gemini's answer. Three concrete cases, a five-step workflow, and what stays in your hands.
What Gemini does with your documents, and where to check your settings
Key points
- Google documents the Gemini settings that concern activity and data, by account type; this documentation changes over time and is the reference.
- Account settings and anonymization complement each other: the former frame how data is used, the latter acts on the content before it leaves your computer.
- Biffer removes what identifies and what is confidential before sending, offline, and keeps the document readable thanks to stable aliases.
Google provides documentation on Gemini, its settings and how activity is handled depending on whether you use a personal or a work account. These terms differ from one plan to another and are updated regularly; Google's documentation remains the up-to-date reference. To know the exact state of your account, see Google's Gemini Help Center.
What anonymization adds: a file dropped into Gemini, sent to a remote service and processed by it, leaves already stripped of whatever points to a person. A candidate's name, an employee's ID or a client's French company number (SIRET) together with their IBAN trigger the obligations the GDPR attaches to a transfer; a document prepared with Biffer lets you control them at the source. We summarize them on the page personal data and AI.
Preparing a document for Gemini therefore means sending a file that has become anonymous for the assistant, while staying structured enough for it to work properly. Biffer, desktop software published by Romain Rissoan, a trainer and consultant based in Lyon, France, does this work entirely offline: your files and every fragment of text stay on your machine, during preparation as during restoration.
Three use cases we handle with Gemini
The most frequent requests we come across concern everyday management documents: human resources, customer relations, data exports. For each one, the file goes through Biffer, then Gemini, then Biffer again for restoration.
Sorting job applications
A set of resumes and cover letters in PDF or Word, with names, addresses, phone numbers and dates of birth. Biffer processes the batch in a queue: each candidate becomes PERSONNE_001, PERSONNE_002, contact details are masked, and career paths and skills are kept. Gemini builds a reading grid on anonymous profiles, which also reduces bias linked to name or age. The mapping table then gives you back the identities for the rest of the hiring process.
Translating a business letter
A sales proposal or a follow-up to translate, with the client's name, address, prices and terms. Biffer pseudonymizes the recipient and the company, masks contact details and keeps the amounts if the translation needs to include them. Gemini translates a letter addressed to PERSONNE_001 at SIRET_001; you restore the values in the translated version before sending it.
Working with a CSV export
An export from your sales management tool, one row per contact, with email addresses, phone numbers, EU VAT numbers and amounts. Biffer replaces each identifier with a stable alias, row after row, and keeps the amounts and dates the analysis needs. Gemini calculates, groups and comments on rows that have become anonymous. You read the result alongside your original file or restore the aliases in the commentary produced.
In all three cases, the document stays in its original format, with formatting kept. Gemini receives a clean file, and the original stays on your computer.
The Biffer workflow in five steps
The workflow is identical whatever the assistant; we describe it as it appears on screen. Each option is detailed on the page how Biffer works.
1. Drop in one or more files
PDF, Word, Excel, CSV or text. Analysis happens on your computer: people, email addresses, phone numbers, addresses, French company numbers (SIREN and SIRET) verified by the Luhn check digit, French social security numbers, IBANs, license plates, dates, and your profession's fields if you selected one (employee ID, contract number, EU VAT number, French training provider number NDA).
2. Review the highlighted preview
You untick false positives and add whatever is missing by selecting the text; all occurrences are handled together. For each data type, you choose to pseudonymize, mask or keep, depending on the question you will ask Gemini.
3. Save the mapping table
Encrypted with a password (Argon2id and AES-256-GCM) or in plain text. Without a table, processing is permanent: the right choice when the document must stay anonymous for good, for example an applicant grid shared with a selection panel.
4. Send the prepared document to Gemini
Drop the pseudonymized file into the conversation and state your request. Since the aliases are stable, Gemini follows each actor from one end of the document to the other, while their identity stays on your computer.
5. Restore the answer
Give Gemini's answer to Biffer with the mapping table: each alias gets its original value back, on your computer, offline. Review before using it.
What stays in your hands
Anonymizing data before Gemini protects people and identifiers. Four points remain in your hands, and they are what makes the result good.
- You choose what goes out. A price list, a selection method or a commercial term remain readable without the names; you decide whether that content can be entrusted to an online assistant or stays processed on your computer, as we explain on the offline software page.
- You approve the result. A translation, a shortlist of applications: the result is reviewed like any delegated work. Biffer restores the names, you have the final say on the content.
- You complete the detection. Biffer verifies check digits and covers profession-specific fields; the preview lets you add an internal reference or a rare name in one action, and your own keyword fields recognize them from the next analysis on.
- Pseudonymized, therefore restorable. As long as a mapping table exists, the GDPR applies and you keep the ability to restore. Biffer gives you the means to minimize; you remain responsible for the processing.
This page applies to Gemini our general method to prepare a document for AI. The same steps work for ChatGPT and Claude.
Frequently asked questions
Does Gemini read a document correctly when names are replaced by aliases?
Yes. The file prepared by Biffer remains an ordinary PDF, Word, Excel, CSV or text file in which certain values have been replaced by stable aliases such as PERSONNE_001 or SIRET_002. Gemini reads it like any other document. The advantage of aliases over deletion is consistency: the same person carries the same alias at every occurrence, and the assistant can follow roles and exchanges. Keep what your question needs, dates or amounts for example, and leave the professional vocabulary intact: it carries the document's meaning.
Should I anonymize even with a Google work account?
We recommend it. The terms attached to a work account cover how the service uses the data, whereas anonymization acts on what the document contains at the moment it leaves: you remain responsible for that transfer under the GDPR, and you control it at the source. These terms change with the plans; anonymization, for its part, stays in place. Anonymizing before sending acts on the content itself, regardless of the account used and its settings. The two precautions complement each other. For the exact state of your account settings, Google's documentation remains the reference source.
How do I process a batch of resumes before handing them to Gemini?
Drop the files into Biffer together: they form a queue. In reviewed mode, you check the preview of each document; in automatic mode, your policies apply to the whole batch without intervention. The human resources profession profile activates the fields specific to recruitment, such as the employee ID. Each resume is processed on your computer, offline. You then upload the prepared files to Gemini to build a reading grid, and restore the identities with the mapping table from each processing run to continue the hiring process.
Your business documents anonymized, ready to be published or handed to an AI tool.
Biffer runs on your computer without sending a single file. Free for five files a month, unlimited for €10 excl. VAT per computer.
100% offline · Free: 5 files a month · No file sent