What not to paste into AI chatbots

Checked 26 October 2026 · by A.Frick · 5 min read

Passwords, client data, IDs and secrets do not belong in a chatbot: a traffic-light checklist, provider data rules, Swiss and EU basics, and safer options.

A chatbot is not a private notebook. What you paste leaves your device, lands on a provider's servers, and may be stored, reviewed or used to train models, depending on the plan and the settings. A simple test: would you send this text to a stranger at a company you have no contract with? If not, do not paste it.

This is general information, not legal advice. Rules differ by country, job and contract.

The checklist

What providers do with your chats

Rules differ by plan. Consumer apps, business plans and the API are three different cases. Facts below come from each provider's own pages, checked on 26 October 2026.

OpenAI. - API: abuse-monitoring logs are kept up to 30 days. OpenAI states that data sent to the API is not used to train its models unless you opt in. Approved customers can ask for zero data retention. - ChatGPT Free, Plus, Pro: data sharing for model improvement is on by default. Turn off Improve the model for everyone under Data controls to stop it. Temporary chats are not used for training and are kept up to 30 days for safety. - ChatGPT Business: OpenAI's help pages say business data is excluded from training by default.

Anthropic. - Free, Pro, Max: chats can be used to improve models if the Model Improvement setting is on, if a chat is flagged for safety review, or if you opt in. Incognito chats are not used. Feedback ratings are kept up to 5 years. Deleted chats are purged from back-end storage within 30 days. If you allow training, chats may be kept up to 5 years in de-identified form. Flagged chats can be kept up to 2 years. - API and business products: Anthropic deletes inputs and outputs within 30 days. Retained data is not used for training without your permission. Zero data retention exists by agreement.

Google. - Gemini app: activity auto-deletes after 18 months by default (you can pick 3 or 36 months, or never). With Keep Activity off, and in temporary chats, chats are kept 72 hours and not used to train models unless you send feedback. A subset of chats is reviewed by humans, and reviewed chats can be kept up to three years even if you delete your activity. - Google Workspace: Google says it does not train on customer data without prior permission, and that Gemini app chats are not reviewed by human reviewers.

Rules can change. Read the current page before you rely on a setting. The Gemini API for developers has its own terms; read them if you build on it.

Switzerland: the revDSG

The revised Federal Act on Data Protection (revDSG, also nDSG; in force since 1 September 2023) applies directly to AI use, says the EDÖB. A few basics:

• Personal data is any information about an identified or identifiable person (Art. 5). Sensitive data includes health, religion, politics, intimate sphere and biometric data.
• Principles (Art. 6): lawful, in good faith, proportionate, and used only for a stated purpose.
• Security (Art. 8): you must protect data with suitable technical and organizational measures.
• Processors (Art. 9): you may hand data to a processor only if it handles it as you may, and no secrecy duty forbids it.
• Abroad (Art. 16): sending data abroad needs adequate protection or safeguards such as standard clauses.
• Professional secrecy (Art. 62): revealing secret personal data you learned through your job is punishable, with a fine up to CHF 250,000.
• The EDÖB adds that users of conversational AI may ask whether their inputs are used to improve the system.

EU: the GDPR

The GDPR rests on principles in Art. 5: lawfulness, fairness and transparency; purpose limitation; data minimisation; accuracy; storage limitation; integrity and confidentiality. Art. 9 sets stricter rules for special categories such as health, biometric and genetic data, and political or religious views. If a provider handles personal data for you, Art. 28 asks for a contract and for processors that give sufficient guarantees. The EU AI Act comes on top for AI systems.

Safer options

1. Anonymize. Swap names, numbers and places for placeholders. Check the result: odd details can still point to a person.
2. Share less. Paste the paragraph that matters, not the whole file.
3. Use the right tier. Business or enterprise plans and the API give data terms, admin controls and no training by default.
4. Go local. A local model, often an open-weight model, keeps text on your machine. See how to choose a model.
5. Check settings. Turn off training where offered, use temporary or incognito chats for one-off questions, and know how long deletes take.
6. Ask first. Follow your employer's AI rule. If none exists, ask.

Warning: If you pasted a secret by mistake, treat it as exposed. Revoke the key, change the password, and tell your security contact. Deleting the chat is not enough.

Tip: Keep a short list of placeholders you use often. It makes safe prompts quick to write.