How to anonymise a document for ChatGPT
The short answer. Don't upload a confidential document to ChatGPT as it is. First replace the details the task doesn't need: people's names, company names, account numbers, addresses and sometimes amounts become placeholders such as [NAME_01], with the same placeholder each time for the same value. You can do this by hand, but it takes time and it's easy to miss something. With ShareSafe.ai you upload the file, check what was found and download the protected version. Choose reversible replacement if you want to put the real values back into the AI's answer later: you get a recovery key (Identity Key) for that, and ShareSafe Free keeps no copy.
When is this risky?
ChatGPT is genuinely good at document work, which is exactly why real client material keeps ending up in it. Four situations that happen every day:
- An advisor uploads a client's annual figures for a summary. The client's name and figures are now outside his control.
- A recruiter pastes a CV to generate interview questions. Address, date of birth and ID-like numbers travel along.
- A consultant has a draft contract rewritten. Parties, amounts and penalty clauses sit in the prompt.
- An HR consultant summarises interview notes. Employee names and health details go with them.
The problem is not that OpenAI does something malicious. The problem is that you can no longer show what happens to that data, and you have to be able to explain it to your client. "No training on business data" is not the same as "nothing sensitive was sent".
What to replace
It varies per document type, but this is the core list:
| Category | Examples |
|---|---|
| People | names, job titles in small teams, signatures |
| Organisations | client names, suppliers, subsidiaries |
| Financial | IBANs, VAT and registration numbers; amounts and rates only if the task doesn't need them |
| Contact | addresses, email addresses, phone numbers |
| Identifiers | national ID numbers, employee numbers, case numbers, licence plates |
| Context | dates and places that identify someone in combination |
Replace data the task doesn't need. Leave amounts the model has to compare or calculate with.
That last category in the table is missed most often. "The CFO of a family business in a small Dutch town" is not a name, but it identifies a person.
Placeholders replace personal data. Contract terms, trade secrets or medical content stay readable. In the review step you can select extra text and have it replaced too. So read the result before you share it.
Three steps
- Drop your file into ShareSafe.ai. PDF, Word, Excel or text, up to 25 MB. No account. Processing happens in the EU and your document is not kept as a file.
- Check the before and after view. You see exactly what was found: Jan de Vries becomes [NAME_01], Acme Corp becomes [ORG_01], Kerkstraat 12 becomes [ADDRESS_01]. Missing something? Add it before you continue. That review is part of working safely, not a formality.
- Download the protected file and your recovery key. The protected file goes to ChatGPT. When the answer comes back with placeholders, your recovery key puts the real names back. The key is given to you; ShareSafe Free does not keep a copy.
If you want no way back, choose permanent replacement instead of reversible replacement. Then no key is created and nobody can restore the replaced values, including us. That does not by itself make the document anonymous: other details in the text can still point to someone.
FAQ
- Does this make my document GDPR-proof?
- No tool can guarantee that. Replacing personal data lowers the risk, but no automatic detection is infallible. That is why ShareSafe.ai shows what was found and you review before using the file. With reversible replacement a key exists, so legally you are still processing personal data. You hold that key.
- Why not just black bars in the PDF?
- Black bars are often a layer on top of the text: the content is still underneath and can be copied out. ChatGPT also loses the document's coherence. Consistent placeholders ([NAME_01] stays [NAME_01] everywhere) keep the document usable for analysis.
- Does ShareSafe.ai store my document?
- ShareSafe Free processes your document in the EU and does not keep it as a file after the run. With reversible replacement you get the recovery key and Free keeps no copy. Every run comes with a processing receipt. See the security page for the exact route.
- Does this work for Claude, Gemini or Copilot too?
- Yes. The protected file is an ordinary file. Which AI you give it to is up to you.
- What if I lose my recovery key?
- Then the placeholders in AI answers stay placeholders; ShareSafe Free keeps no copy, so we can't put them back either. Your original on your own machine is untouched.
ShareSafe.ai keeps raw files in the EU. Minimal retention. You hold the key. Try it with your own document →