Not everything belongs in an AI chat. Learn what not to share with ChatGPT or Claude, how to anonymize text, and which privacy settings to check.
ChatGPT, Claude, and other AI tools can help with writing, learning, and work. But they are not private vaults. When you upload text, a document, or an image, you are sending it to an external online service. Anything whose exposure could harm you, another person, or a business should stay out of the chat.
A simple rule helps: if you would not send the information to an unfamiliar online service, do not put it into an AI chat. Anonymize sensitive text first, and use only tools approved by your employer or client for work-related material.
Quick answer: what not to share with an AI chatbot
Do not share the following with ChatGPT, Claude, or another public AI tool:
- national identification numbers, passport details, or driver's license numbers,
- passwords, recovery codes, API keys, tokens, or private SSH keys,
- medical records, test results, or other sensitive health information,
- legal documents, contracts, bank statements, or invoices containing personal data,
- information about clients, customers, patients, students, or employees,
- internal company files, pricing, proposals, strategies, or trade secrets,
- private photos, videos, recordings, or information about children,
- confidential or copyrighted material that you are not allowed to process this way.
If you are unsure, do not upload the full document. Remove names, contact details, identification numbers, internal labels, exact amounts, and any detail that could identify a person or company.
Why information in an AI chat can be risky
A prompt is not a private note. Submitted content may remain in chat history, be subject to retention rules, or be processed for security purposes. How data is handled depends on the service, account type, and your settings.
Personal and business accounts can also follow different rules. An organization may control retention, access, and approved integrations. Using a tool personally does not give you permission to upload a client or company document.
Privacy settings in ChatGPT and Claude
In ChatGPT, review Data Controls, Memory, and temporary chat options. In Claude, check the privacy settings, whether conversations may be used to improve the service, and incognito chat options. Names and availability can vary by plan and may change over time.
Turning off training, memory, or chat history does not make the service a safe place for passwords or personal data. Before handling sensitive material, check the current terms and your organization's rules.
If you are new to the tool, start with ChatGPT in English: how to get started.
How to anonymize text before uploading it
Anonymization means more than deleting a name. Remove anything that could reveal the person, company, or case when combined with other details.
For example:
| Original information | Safer replacement |
|---|---|
| Jane Smith | [NAME] |
| ID number 123456789 | [ID_NUMBER] |
| Example Company Ltd. | [COMPANY] |
| [email protected] | [EMAIL] |
| +1 202 555 0142 | [PHONE] |
| $12,450 | [AMOUNT] |
| internal project name | [PROJECT] |
For work documents, also replace client names, order numbers, internal codes, office locations, and exact prices. Several innocent-looking details can identify someone even when the name is gone.
A safer prompt might look like this:
Help me simplify this email. Keep anonymized labels such as [NAME], [COMPANY], [AMOUNT], and [DATE]. Do not invent facts or add personal information.
Text:
[paste anonymized text]
When it is better not to use AI
Avoid AI when sending the information to an external service would itself create a risk. Without an approved and properly secured environment, do not use it for:
- processing medical records or information about children,
- legal or financial decisions with real consequences,
- processing client data without permission and a suitable legal basis,
- analyzing internal company documents outside an approved tool,
- uploading private photos or recordings of other people without consent,
- generating or storing passwords, tokens, or recovery codes.
For health, legal, or financial matters, AI may help you prepare general questions or an outline. Final decisions should be made by a qualified professional or through an approved internal process.
Passwords, tokens, and access credentials do not belong in AI chats
Never paste a password or API key into an AI chat, even to explain an error. If you already did, treat it as exposed:
- Change the password or revoke the token.
- Review recent sign-ins and active sessions.
- Enable two-factor authentication when available.
- Save the new password in a password manager.
- Report a work account incident according to company policy.
If you need a new secure password, use the password generator. You can also read How to create a strong password and Two-factor authentication: what it is and how to enable it safely.
Practical tip
Generate a strong password in seconds
Create a secure password in seconds.
A short checklist before sending a prompt
Before you press Enter, confirm that:
- The text contains no passwords, identity documents, or sensitive personal data.
- I am allowed to process this document in the chosen AI tool.
- I removed names, contact details, amounts, and internal labels.
- I am using a tool approved by my organization for work material.
- I reviewed privacy, history, and memory settings.
- I will treat the AI output as a draft, not a final professional decision.
If the answer to any item is “I don't know,” do not send the prompt yet.
Summary
ChatGPT and Claude are useful for explaining concepts, preparing structures, and editing anonymized text. Full contracts, client material, personal data, and access credentials do not belong in a public AI chat.
Remember one practical rule: anonymize first, ask second. For work, health, legal, and financial topics, use only approved tools and processes.