Anonymize data before ChatGPT: what we do before every upload
To anonymize data before ChatGPT is to replace, on your computer, the names, contact details and identifiers in a document with aliases before dropping it into ChatGPT, then to restore the real values in the answer. It is the stage of the process that depends entirely on you. Before sending, you decide everything: what goes out, what stays, what is replaced. That is when Biffer puts you in control, on your computer, before the file reaches OpenAI's servers and the terms of the service.
We use ChatGPT every day, in training as in consulting, and we only ever give it prepared documents. This page describes how Biffer prepares a contract, a letter or an Excel spreadsheet on your computer, offline, and how you then restore the real values in the answer you get.
What ChatGPT does with your documents, and what you can configure
Key points
- OpenAI offers settings for conversation retention and for the use of data for training; they change over time, and the official documentation 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 personal and confidential data before sending, offline, and gives you back a coherent document thanks to stable aliases.
In the account settings, OpenAI provides options covering history retention and the use of content to improve its models. These options depend on your plan and on the version of the service, and they change regularly: the official documentation remains the up-to-date reference. To know the exact state of your settings, see the OpenAI Help Center.
What anonymization adds: the file you send, stored for the time of processing and readable by the service, leaves already stripped of whatever points to a person. A client's name, their social security number or a supplier's IBAN trigger GDPR obligations toward a third-party provider; a document prepared with Biffer lets you control them at the source. We summarize them on the page personal data and AI.
Using confidential documents with ChatGPT becomes straightforward once the document you send is cleared of whatever identifies a person or falls under trade secrets, while staying readable. That is exactly what Biffer does: desktop software published by Romain Rissoan, a trainer and consultant based in Lyon, France, that works offline, so your files stay on your computer throughout processing.
Three use cases, from raw file to restored answer
Here are the three requests we come across most often, and how we handle them. In each case, the document goes through Biffer, then ChatGPT, then Biffer again for restoration.
Summarizing a contract
A twenty-page service agreement, with the parties, their French company numbers (SIRET), the signatories, amounts and deadlines. Biffer pseudonymizes people and companies (PERSONNE_001, SIRET_001), keeps dates and amounts because the summary has to cover them, and clears the Word file's metadata. ChatGPT summarizes a contract between SIRET_001 and SIRET_002, signed by PERSONNE_001 and PERSONNE_003. The mapping table puts the names back in place in the summary.
Rephrasing a letter
A payment reminder or a reply to a complaint: the recipient, their address, their case number, sometimes an amount due. Here the question is about tone and wording: Biffer pseudonymizes people and masks amounts and contact details. ChatGPT rephrases a letter addressed to PERSONNE_001 about case DOSSIER_001. After restoration, the letter is ready to go, with the tone corrected and the right information.
Analyzing an Excel spreadsheet
A billing export or a payroll tracker, with one row per person. Biffer replaces each name, employee ID and IBAN with a stable alias, row after row, and keeps amounts and dates, which give the analysis its meaning. ChatGPT spots gaps, duplicates and trends in rows that have become anonymous. You then restore the identities in the commentary produced, or read the analysis alongside your original file.
In all three cases, formatting is kept: the contract remains a Word file, the spreadsheet remains an Excel workbook. ChatGPT receives a clean document, in its original format.
The Biffer workflow in five steps
The workflow is the same whatever the document. We describe it here as it unfolds on screen; each option is detailed on the page how Biffer works.
1. Drop the file into Biffer
PDF, Word, Excel, CSV or text, one or several at a time. Analysis starts 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, plus your profession's fields if you chose one.
2. Review the highlighted preview
Untick false positives and add whatever is missing by selecting the text: all occurrences are handled together. For each type, choose the policy that suits your question: pseudonymize, mask or keep.
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.
4. Send the prepared document to ChatGPT
Drop the pseudonymized file into the conversation and ask your question. ChatGPT works on consistent aliases: it follows PERSONNE_001 from one paragraph to the next, while the name stays on your computer.
5. Restore the answer
Give ChatGPT's answer to Biffer along with the table: each alias gets its original value back. The summary, letter or analysis is ready to be reviewed, then used.
What stays in your hands
Anonymizing data before ChatGPT protects people and identifiers. Four points remain in your hands, and they are what makes the result good.
- You choose what goes out. A sales strategy or an unusual clause remain readable without the names; you decide whether that content can be entrusted to an online AI or stays processed on your computer, as we explain on the offline software page.
- You approve the answer. A summary or analysis from ChatGPT is reviewed like any delegated work: Biffer restores the names, you have the final say on the content.
- You review the preview. Biffer detects a lot and verifies check digits; the highlighted preview lets you add a rare name or an internal reference in one action, and your own keyword fields recognize them from the next analysis on.
- The GDPR still applies as long as a mapping table exists: the data is pseudonymized, and you keep the ability to restore it. Biffer gives you the means to minimize; you remain responsible for the processing.
This page is part of our general method to prepare a document for AI, which applies in the same way to Claude and Gemini.
Frequently asked questions
Can I send a confidential document to ChatGPT if I have turned off training?
The setting frames how your data is used; anonymization acts on what the document contains at the moment it leaves. Turning off training, if your plan allows it, is a good practice to keep; anonymizing before sending is another, independent one, which removes from the file the names, identifiers and amounts that make it sensitive. We do both. For the exact state of your account settings, OpenAI's documentation is the reference source, updated continuously.
Does ChatGPT understand a document where names are replaced by PERSONNE_001?
Yes, and that is the main advantage of stable aliases over plain deletion. Each person, each company, each identifier gets a unique alias, repeated identically at every occurrence. The model can therefore follow who signs, who pays, who disputes, and produce a coherent summary or analysis. We see in training sessions that answer quality stays at the same level, provided you keep what the question needs: dates for a schedule, amounts for a numerical analysis. Professional vocabulary and the document's structure are always kept.
How do I restore the names in ChatGPT's answer?
During processing, Biffer offers to save a mapping table, encrypted with a password or in plain text. Once you have the answer, you give it to Biffer with that table: each alias is replaced by its original value, and you get a complete text back. Restoration happens on your computer, offline, like the rest of the processing. If you chose to work without a table, processing is permanent by design: the choice to make when the document must stay anonymous for good.
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