Skip to content
PiiBlocker
Back to blog

How to Mask PII in ChatGPT, Claude, and Gemini

PiiBlocker Team

privacychatgptclaudegeminigdprpii

To mask PII in ChatGPT, Claude or Gemini, replace names, emails, ID numbers and other identifying details with placeholders before you send the prompt, so the provider receives [PERSON_A] and never the real value. You can do it by hand, or use a browser extension that does it at the moment you send. PiiBlocker is a free Chrome extension that does this on ChatGPT, Claude, Gemini, Perplexity, Grok, DeepSeek and Kimi, with detection running in your browser. Privacy settings help too, but they only change what happens to your data after it arrives. This guide covers both.

The risk is not theoretical. In a decision published in February 2026, a UK immigration solicitor told the Upper Tribunal he had put client emails and Home Office decision letters into ChatGPT to improve and summarise them. The tribunal said that doing so breaches client confidentiality and waives legal privilege.

What happens to your data when you send a prompt

When you type into ChatGPT, Claude or Gemini, the prompt is sent to the provider's servers, processed and stored. What happens next depends on the provider and your settings. This is the position for personal accounts as of October 2026.

ChatGPT (OpenAI): Chats stay in your history until you delete them. They may be used to train models unless you turn off "Improve the model for everyone". Temporary Chats are not used for training and are deleted after 30 days.

Claude (Anthropic): Since a change to its consumer terms in 2025, Anthropic asks Free, Pro and Max users to choose whether their chats are used to train models. With the model improvement setting on, chats can be used for training and kept for up to five years. With it off, the retention period is 30 days.

Gemini (Google): Keep Activity is on by default for adults, and activity is kept for 18 months unless you change it to 3 or 36 months. A share of chats is read by human reviewers, and reviewed chats are kept for up to three years even if you delete your activity. With Keep Activity off, chats are still held for up to 72 hours.

The core issue is the same across all three. Your data leaves your device the moment you press send.

What is at risk

Most people underestimate what they share. A single prompt often contains several kinds of personal data without the writer noticing.

Healthcare workers share patient symptoms, medications and diagnoses when asking for help with clinical notes. Lawyers paste client names, case details and confidential correspondence when drafting. Developers include API keys and credentials when debugging. HR teams share employee names, salaries and performance data when writing reports. See real-world examples across industries.

Why "just be careful" does not work

The standard advice is to review your prompts before sending them. In practice it fails for three reasons.

Speed defeats caution. People use AI chatbots because they are fast. Checking every prompt by eye removes that benefit, so the checking stops after the first few uses.

Personal data is easy to miss. A card number inside a transaction log, a key inside a code snippet or a National Insurance number in an employee record does not stand out when you are focused on the task.

Pasting is the biggest risk. The riskiest prompts are not typed. They are pasted from documents, spreadsheets and emails, and one paste can carry dozens of details you never consciously read.

Step 1: Change the settings on each chatbot

This takes two minutes and everyone should do it.

  • ChatGPT: Settings, then Data Controls, then turn off "Improve the model for everyone".
  • Claude: Settings, then Privacy, then turn off "Help Improve our AI models".
  • Gemini: Settings & help, then Activity, then turn off Keep Activity, or set auto-delete to 3 months.

ChatGPT and Gemini also offer a Temporary Chat, and Claude offers incognito chats. These stay out of your history.

None of this stops the data being sent. Settings control training and retention. Your prompt is still transmitted, processed and held on the provider's servers for some period. For more on what one provider keeps, see what ChatGPT remembers about you, and how to check.

Step 2: Mask PII before you send

Masking is the only approach where the real data never reaches the provider.

By hand. Replace each identifying value with a placeholder before you paste: [PERSON_A] for a name, [EMAIL_A] for an address. Keep the same placeholder for the same value, and swap the real values back into the reply afterwards. It works in any chatbot and costs nothing, but it is slow and easy to get wrong. Our guide to masking personal data before pasting into ChatGPT has the full steps.

Automatically. PiiBlocker is a free Chrome extension that masks at the point of sending. It detects 15 or more types of personal data, including card numbers, Social Security numbers, names, addresses, phone numbers and emails. Card numbers, Social Security numbers and API keys are masked automatically. Names, addresses and emails wait for one click from you. When the reply comes back, the real values are restored on your screen.

Detection runs in your browser, and the free version makes no network calls. It works on ChatGPT, Claude, Gemini, Perplexity, Grok, DeepSeek and Kimi.

PiiBlocker is not the only tool that does this. We compare it with five others in Best Chrome Extensions to Hide Personal Data from ChatGPT (2026).

For teams

An extension protects what one person types in one browser. Organisations that need policy enforcement across every device and app use data loss prevention tools such as Microsoft Purview, Nightfall or Strac, usually alongside a browser extension. We cover the trade-offs in DLP vs. browser extensions for AI chatbots.

GDPR and compliance

If you handle EU residents' data, sending personal data to an AI provider is processing, and it needs a legal basis. Masking before sending means the provider does not receive the identifying values, which reduces what you have to justify. It does not remove your other obligations, and whether masked data still counts as personal data depends on your situation. Our GDPR guide for DPOs goes into the detail. If a customer has already asked you to delete their data, see deleting customer data from ChatGPT, Claude and Gemini.

Frequently asked questions

How do I mask my prompt before sending it to AI? Replace names, emails, ID numbers and other identifying details with placeholders before you press send. You can do it by hand in any chatbot, or use an extension such as PiiBlocker, which does it automatically on ChatGPT, Claude, Gemini, Perplexity, Grok, DeepSeek and Kimi.

Does turning off training stop ChatGPT, Claude or Gemini storing my data? No. The setting stops your chats being used to train models. The prompt is still sent to the provider and stored for a period that depends on the provider and your settings.

Does masking work the same way in Claude and Gemini as in ChatGPT? Yes. The chatbot receives placeholders in place of the real values and answers as normal. PiiBlocker restores the real values in the reply on all seven sites it supports.

Does the AI still give a useful answer when the data is masked? Yes, for drafting, rewriting, summarising and analysis. It cannot do tasks that need the real value, such as looking up a named company.

Is PiiBlocker free? Yes. Core detection and masking is free with no account and no usage limits.

Sources

PiiBlocker is a free Chrome extension that masks personal data before it reaches AI chatbots. The free version runs in your browser and makes no network calls. Install from the Chrome Web Store.