Anonymize data before Claude: preparing a document for Anthropic's assistant
To anonymize data before Claude is to replace, on your computer and offline, the personal data in a document with stable aliases before handing it to Claude, then to put the original values back into its answer. The approach follows the same logic as with any online assistant: the stage you fully control comes before sending, and that is where everything is decided. Claude, developed by Anthropic, is often chosen to work on long, structured documents: contracts, case files, meeting minutes. These are precisely the documents that contain the most names, identifiers and amounts.
Here we describe how Biffer prepares these files on your computer, offline, and how you then restore the original values in Claude's answer. Three concrete cases, a five-step workflow, and what stays in your hands.
What Claude does with your documents, and where to check your settings
Key points
- Anthropic documents how conversations are handled and which settings are available depending on the plan; this documentation changes over time, and it 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.
Anthropic publishes its terms of use and privacy policy on anthropic.com, and describes the settings available in Claude, including those that concern data, in its Help Center. These settings depend on the plan (individual, team, enterprise, programmatic API access) and change over time; Anthropic's documentation remains the up-to-date reference.
What anonymization adds: the file you upload, sent to a remote service and processed by it, leaves already stripped of whatever points to a person. An employee's name, a doctor's RPPS number (the French health professional identifier) or a client's IBAN trigger the information, legal basis and security 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 Claude therefore means sending a file that has become anonymous for the assistant, while staying coherent 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.
Three use cases we handle with Claude
The requests that come up most often in our training sessions are comparison, synthesis and drafting tasks on long documents. For each one, the document goes through Biffer, then Claude, then Biffer again for restoration.
Reviewing a contract clause by clause
A service agreement in Word, with the parties, their French company numbers (SIRET), the signatories, liability and termination clauses. Biffer pseudonymizes people and companies, keeps dates and amounts, and clears the metadata, including comment authors. Claude reviews the contract, flags unbalanced clauses and asks the questions a lawyer would ask, while SIRET_001 and PERSONNE_002 keep their identities on your computer. You then restore the names in its analysis.
Synthesizing a long case file
A file of several dozen pages: an expert report, correspondence, official minutes. Biffer pseudonymizes people and case and claim numbers, masks contact details and keeps the chronology. Claude produces a chronological synthesis in which PERSONNE_001 and SINISTRE_001 play their roles under aliases. The mapping table makes the synthesis readable for your team.
Drafting a reply to a complaint
A customer complaint, with the name, address, contract number and disputed amount. Here, the reply must be accurate in substance and careful in form; the name and address can stay on your computer. Biffer pseudonymizes people, masks contact details, and keeps the amount and the contract number under an alias. Claude drafts a reply addressed to PERSONNE_001, which you restore before sending.
In all three cases, formatting is kept and the document stays in its original format. Claude receives a clean file, and you keep the original on your computer.
The Biffer workflow in five steps
The workflow is the same from one assistant to the next; we describe it here 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 (French bar locker number (toque), court case number (RG), RPPS, claim number, employee ID).
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 Claude.
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 choice to make when the document must stay anonymous for good.
4. Send the prepared document to Claude
Drop the pseudonymized file into the conversation and state your request. Since the aliases are stable, Claude follows each actor from one end of the document to the other, while their identity stays on your computer.
5. Restore the answer
Give Claude'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 Claude protects people and identifiers. Four points remain in your hands, and they are what makes the result good.
- You choose what goes out. An unusual clause, an internal method or a negotiated price 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 synthesis. A comparison, a timeline: the answer 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 Claude our general method to prepare a document for AI. The same steps work for ChatGPT and Gemini.
Frequently asked questions
Does Claude accept a document where names are replaced by aliases?
Yes. A document prepared with Biffer remains an ordinary Word, PDF, Excel, CSV or text file in which certain values have been replaced by stable aliases such as PERSONNE_001 or SIRET_002. Claude reads it like any other document. The advantage of aliases over deletion is that consistency is preserved: the same person carries the same alias at every occurrence, and the assistant can follow roles, commitments and exchanges. Keep what your question needs, dates or amounts for example, and leave the professional vocabulary intact: it carries the meaning.
Should I anonymize even if my Claude plan includes data protection commitments?
We recommend it, for two reasons. First, contractual commitments cover how data is used, 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. Second, plans and their terms change with migrations and plan switches; anonymization, for its part, stays in place. Anonymizing before sending acts on the content itself, regardless of the plan. The two precautions complement each other.
Can I prepare several files for the same conversation with Claude?
Yes, that is what Biffer's queue is for. You drop the files in together, in reviewed mode to check each preview, or in automatic mode to apply your policies without intervention on a consistent batch. Each file is processed on your computer, offline, then you upload them to the conversation with Claude. For restoration, you use the mapping table from each processing run. It is the safest way to hand a complete case file to an assistant while protecting the people in it.
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