How to anonymize data before pasting it into ChatGPT

For ChatGPT, Copilot, Claude, Gemini and other AI assistants

AI assistants are good at spreadsheets: summarising, finding patterns, writing formulas. But everything you paste or upload is sent to the provider. If the rows describe real customers or employees, replace the personal data first. Done well, the AI can still do the analysis.

Why it matters

• The data leaves your organisation. A chat is processed and stored on the provider's servers, whatever its training settings.
• Personal accounts may be used for training. OpenAI, for example, can use conversations from personal ChatGPT accounts to improve its models unless you turn off "Improve the model for everyone" (OpenAI data controls). Business plans and the API are not used for training by default (OpenAI business data), but the data is still sent to and held by OpenAI.
• Your own rules still apply. Under the GDPR you need a reason for every use of personal data, and many company policies forbid putting customer data into AI tools. Anonymized data avoids the question.

How to do it without losing the analysis

1. Drop what the question does not need. Asking about sales by region? Remove names, emails and phone numbers entirely.
2. Replace identifiers with consistent stand-ins. If the AI needs to tell customers apart, replace each name with a fake one, and the same name with the same fake every time. Totals per customer, repeat purchases and joins between sheets keep working.
3. Generalize the details that point to one person. A birth year instead of a date, the first digits of a postal code, an age band instead of an exact age.
4. Check free-text columns. Notes and comments often contain names and phone numbers. Anonymize the personal data inside the text, or leave the column out.
5. Keep the key yourself. Create a mapping file (original next to replacement) and keep it on your computer. When the AI answers "customer Eino Kinnunen is your biggest account", you can look up who that really is. Never paste the mapping into the chat.

Doing it with Scrubsheet

Open your file in the Scrubsheet browser tool. It suggests what each column contains, and you pick Fake, Mask, Hash, Generalize, Redact or Clear per column. Turn on consistent output with a project secret to keep the same replacement for the same value across sheets and files, and tick Also create a mapping file to keep the key. The file never leaves your browser, so you are not sharing the real data with yet another service to prepare it for the AI.

Before you paste

• Scroll through the result: automatic detection cannot guarantee it found everything.
• Remember that rare combinations (a small town and a job title) can still identify someone.
• Send the anonymized file, never the mapping file.

This guide explains the rules as we understand them; it is not legal advice. Whether a particular dataset is anonymous depends on its content and on who receives it.

More guides