Short answer
Never paste Social Security or passport numbers, card or bank details, passwords, confidential company files, or other people's health or customer data into an AI chatbot, whatever your settings are. Turning off training doesn't stop chats from being stored for a while. You can switch off training use of your chats in each service's settings, and temporary or incognito chats reduce what is kept. For work, use an account your employer approves, not a personal one.
Three different things: storage, training, human review
“The AI sees my chats” mixes up three things. Storage means your conversation sits on the provider's servers for some time. means saved conversations may be used to build future models. Human review means that, for safety or quality, some conversations may be read by staff or contractors.
The “don't train on my data” settings mainly stop training. They don't remove storage or safety-related review. That is why sensitive information is safer left out than protected by a setting.
Where each setting lives (checked against official help pages)
Menu names and locations change with app updates. The table reflects official help pages checked on November 5, 2026. If you can't find a setting, open Settings and look for a Privacy or Data menu.
ChatGPT
- Setting name
- Improve the model for everyone
- Where to find it
- Profile menu > Settings > Data controls
- Notes
- When off, new conversations are not used to train OpenAI models. They still appear in your chat history.
Claude
- Setting name
- Model improvement (model training setting)
- Where to find it
- Settings > Privacy (claude.ai/settings/data-privacy-controls)
- Notes
- When on, chats can be kept up to five years. Incognito chats are not used to improve Claude even if it's on.
Gemini
- Setting name
- Keep Activity (Gemini Apps Activity)
- Where to find it
- gemini.google.com > Settings & help > Activity
- Notes
- Even when off, chats may be saved in your account for up to 72 hours for service and feedback.
ChatGPT: Data controls and Temporary Chat
- On the web, open your profile menu, choose Settings, then Data controls, and turn off “Improve the model for everyone.” In the mobile app, tap your profile to find the same menu.
- The setting applies to new conversations. It doesn't delete the chat history you already have.
- Temporary Chat doesn't appear in your history and isn't used to improve OpenAI models. OpenAI says it may keep a temporary chat for up to 30 days for safety purposes.
- If you save a temporary chat, it becomes a regular chat and follows your account's settings.
- OpenAI says content in business workspaces (Business, Enterprise, Edu) is not used to train its models by default.
Claude: the 2025 policy change
In August 2025, Anthropic changed its consumer terms for Free, Pro and Max accounts, including Claude Code used from those accounts. If you allow it, your chats and coding sessions can be used to improve Claude, and retention extends to five years. If you don't allow it, the existing 30-day retention continues. The change doesn't apply to commercial products such as Team, Enterprise, education, government, or the API.
Anthropic's privacy help page says consumer chats may be used for model improvement in three cases: you opted in to the model improvement setting, a conversation was flagged for safety review, or you explicitly joined a program such as Trusted Tester. Turning the setting off keeps your chats out of future training. It doesn't remove data from training that already happened.
Incognito chats are not used to improve Claude, even if model improvement is on. You can change the setting any time in Settings > Privacy.
Gemini: turn off Keep Activity
Google's help page didn't cover temporary chats or human review in the sections checked, so they aren't described here. Read Google's Gemini Apps privacy hub for those details.
- Go to gemini.google.com, open Settings & help, then Activity. Near the top, click On, then choose Turn off or Turn off and delete activity.
- Google says that even with Keep Activity off, your conversations are saved to your account for up to 72 hours to provide the service and process feedback. That activity doesn't show up in your Gemini Apps Activity list.
- Activity older than 18 months is deleted automatically by default. You can change it to 3 or 36 months, or turn auto-delete off.
What not to paste into an AI
Even with training off, keep the items below out of your prompts. If you need help with a task that involves them, swap in fake values.
| Do not paste | Do this instead |
|---|---|
| Social Security, passport or driver license numbers | Mask them or use made-up values, like “123-45-XXXX” |
| Bank account or card numbers, one-time codes, passwords | Never enter them. Ask a general question, such as “what makes a strong password?” |
| Confidential company files, unreleased numbers, contract terms | Replace names and amounts with placeholders and rough figures |
| Patient records, student grades, customer lists | Remove names and contact details before asking |
| Other people's photos, addresses or phone numbers | Don't use them without their consent |
A customer, Alex Kim (555-0100), emailed asking for a refund. Leave out the name and phone number. Summarize the situation and write a short, polite reply. (Paste the email body here after removing the name, phone number and order number.)
The name and number above are made up. Remove real customer details before you paste.
Personal accounts vs. work accounts
On personal accounts, free or paid, you choose the training setting yourself. On workplace plans (such as ChatGPT Business and Enterprise, or Claude Team and Enterprise), an admin sets the policy, and the providers say conversations are not used for model training by default. That can make a work account a better fit for work material than a personal one.
Even so, each employer sets its own rules. Many ban work content in personal accounts, so check your security team or company policy first. Legally protected data, such as health or financial records, may have extra requirements even on a work account.
Related terms
- Training (model improvement)
- Using user conversations to build future AI models. Often you can switch it off in settings.
- Temporary chat / incognito chat
- A chat mode designed to leave no history and stay out of training. Names and retention differ by service.
- De-identification
- Removing or replacing details that identify a person, such as names.
FAQ
Is it safe once I turn off training?
It helps, but it isn't enough. Chats can still be stored and reviewed for safety, so leave sensitive information out from the start.
What about chats I already sent?
Delete the conversations and turn the setting off. Anthropic says deleted chats are not used for future training, but it doesn't remove data from models already trained.
Do the settings apply to both the app and the web?
They are mostly account-level, but menu locations can differ by app version. Check on the web first.
Which account should I use for work?
An employer-approved team or enterprise account comes first. If you only have a personal account, read your company policy before using it.
Does deleting a chat erase it completely?
It disappears from your list right away, but the servers delete it later. Anthropic says it removes deleted chats from its back end within 30 days. Check other services' help pages for their timing.
Keep going
Checked against sources on 2026-11-05. Sources: Claude Privacy Center: Is my data used for model training?, Claude Privacy Center: How long do you store my data?, Anthropic: Updates to our consumer terms, Gemini Apps Help: Gemini Apps Activity, OpenAI Help: Data controls in ChatGPT, OpenAI Help: Temporary Chat. How we write