Practical guide · Checked 14 October 2026
Switch off training on your data.
Most AI tools can learn from what you type or upload. Each one hides the switch in a different place. Here's where, for the six you're most likely to use.
It takes about a minute per tool. Do it on every account you use, including the personal ones you use for work.
It's mostly forward-looking.
Opting out stops future use. Data already used in a trained model generally stays there.
Work accounts differ.
Business, Enterprise and workspace plans usually have their own terms. These steps are for personal accounts.
Switching off isn't deleting.
Your chats are still stored. Training is one use of your data, and safety review is another.
ChatGPT
Free, Plus, Pro and Go plans. On by default for personal plans. Business, Enterprise and Edu workspaces are off by default.
01
Open your account menu
On mobile: sidebar, then your profile icon.
02
Settings
03
Data controls
04
Turn off
Improve the model for everyone
Worth knowing
- Temporary Chat isn't used for training (it can be kept up to 30 days for safety).
- Your chats stay in your history unless you delete them.
- Using Codex? It has a separate "Include environments" setting.
Watch out
- Giving a thumbs up or down may let OpenAI use that whole conversation for training.
Source: OpenAI, Data controls in ChatGPT
Claude
Free, Pro and Max, including Claude Code on those accounts. Commercial plans and the API have separate terms.
01
Click your name
The drop-down next to your name. Same route on mobile.
02
Settings
03
Privacy
04
Turn off
Help improve Claude
Worth knowing
- Anthropic says it won't use new chats and coding sessions for future training once this is off.
- With the toggle on, data can be kept for five years. Off, it's 30 days.
Watch out
- Training that has already started still includes your earlier data.
- Conversations flagged by safety systems may still be used for safety work.
Source: Anthropic privacy centre
Gemini
Personal Google accounts. Work and school accounts follow Workspace terms.
01
Open Gemini Apps Activity
02
Turn off
Keep ActivityYou can also choose to delete past activity.
Worth knowing
- With it off, new chats aren't used to train Google's models, unless you send feedback.
- Temporary chats aren't used for training.
- Audio, Gemini Live video and screen shares are off for improvement by default, with a separate setting.
Watch out
- Chats are still kept for 72 hours.
- Any chat a human reviewer has already seen can be kept up to three years, even after you delete your activity.
Source: Google, Gemini Apps privacy hub
Microsoft Copilot
Personal Microsoft accounts only. Microsoft 365 Copilot with a work or school account isn't covered by this setting.
01
Open your profile menu
Mobile: menu, then your profile icon.
02
Settings, then Privacy
copilot.com: profile name, then Privacy. Mobile: Account, then Privacy.
03
Turn off both
Model training on textModel training on voice
Worth knowing
- Microsoft says the opt-out covers past, present and future conversations.
- You can switch training off and keep personalization on.
Watch out
- It can take up to 30 days to apply across Microsoft's systems.
- It doesn't stop other product improvement, advertising, safety, security or compliance uses.
Perplexity
Free, Pro and Max. On by default. Enterprise data is never used for training.
01
Open Account settings
02
Preferences
03
Turn off
AI data retention
Worth knowing
- The setting applies to individual users.
Watch out
- It applies only to data collected after you switch it off. Perplexity says data collected earlier can't be removed.
Source: Perplexity help centre
Grok
Set through your X account, since Grok on X shares X's data settings.
01
Open X settings
02
Privacy & Safety
03
Data sharing and personalization
04
Grok & Third-party Collaborators
Untick "Allow your public data as well as your interactions, inputs, and results with Grok and xAI to be used for training and fine-tuning"
Worth knowing
- Protecting your posts keeps them out of training.
- "Delete Conversation History" removes your data within 30 days.
- There's a separate Grok Personalization toggle.
Watch out
- Thumbs up or down feedback can be used for training.
- X says opting out doesn't stop a feature already deployed from learning in normal use.
Source: X Help, About Grok
Not covered yet: Meta AI, DeepSeek, Mistral and smaller tools. If a tool isn't here, assume the default is on until you've found its setting.
Menu names change often. This guide was checked against each provider's own help pages on 14 October 2026. If a label doesn't match what you see, search the app's settings for "training" or "improve".