AI privacy and safety · How Founders Use AI
Are my AI chats private, and how do I stop ChatGPT, Claude, Gemini, Copilot and Meta AI training on them?
Checked 18 November 2026
On free and personal plans, assume your chats can train the model until you switch it off: Data controls in ChatGPT, Privacy in Claude, History in Gemini, Privacy in Copilot. OpenAI's and Anthropic's business plans do not train on your data by default. Meta AI lets you delete a chat, and passwords or customer data never go in.
On the AI map, 29 October 2026 AI assistantsAI browsersAvatars and voice
By case
-
If you use ChatGPT Free, Plus or Pro
Open Settings, then Data controls, and turn off Improve the model for everyone; your chat history stays.
OpenAI turns data sharing on by default on personal plans.
-
If you use Claude Free, Pro or Max
Open Settings, then Privacy, and turn off Help improve our AI models.
It stops training on new chats, though data already used in training is not removed.
-
If you use Gemini
Turn off History in your Gemini settings, or use a temporary chat for a sensitive question.
Google does not train on temporary chats or on chats with History off, but keeps them with your account for around two months.
-
If you use Microsoft Copilot
Open your profile, then Privacy, and turn off Training on conversation activity and Training on voice conversations.
The two switches are separate, so turning one off leaves the other on.
-
If you use Meta AI on WhatsApp, Instagram or Messenger
Type /reset-ai in the chat to delete Meta's copy of it.
Since 16 December 2025 Meta uses your chats with its AI to pick the posts and ads you see in most regions.
-
If you work with company or client data
Use a business plan, such as ChatGPT Business or Enterprise or Claude for Work.
OpenAI and Anthropic both say they do not train on these plans' data by default.
The figuresEach with where it was read
- How long Google keeps temporary Gemini chats and chats with History off around two months
- How long Google keeps reviewed Gemini chats, disconnected from your account up to 3 years
- How long Anthropic keeps a conversation sent with thumbs up or down feedback up to 5 years
- Day Meta began using AI chats to personalise content and ads 16 December 2025
What goes wrong
Pressing thumbs up or down on a sensitive chat: Anthropic stores the whole conversation for up to 5 years and Google keeps reviewed chats up to 3 years.
Thinking a temporary chat leaves nothing behind, when Gemini still keeps it about two months.
Pasting passwords, ID numbers or a customer list into a personal account.
Assuming one switch covers every app, when each assistant has its own.
Go deeper10 sources
-
Data controls in ChatGPT
OpenAI · Guide · Oct 2026
The exact steps on web and phone to turn off Improve the model for everyone, and what Temporary Chat keeps (up to 30 days for safety).
-
How do I change my model improvement privacy settings?
Anthropic · Guide · Aug 2026
Where the Help improve our AI models switch sits in Claude on a computer and on a phone, and what turning it off does to past chats.
-
How long do you store my data?
Anthropic · Guide · Jul 2026
Deleted Claude chats leave its systems within 30 days; with training on, data can stay up to 5 years.
-
Gemini Apps Privacy Hub
Google · Guide · Oct 2026
What Keep Activity controls, the 72-hour hold when it is off, and the 3 years reviewed chats are kept.
-
Microsoft Copilot privacy controls
Microsoft · Guide · 2026
The Training on conversation activity and Training on voice conversations switches, with the path on web, desktop and phone.
-
Microsoft Copilot for individuals: your privacy controls and choices
Microsoft · Guide · 2026
The controls in the Copilot app Microsoft released on 18 August 2026: memory, chat history, web search and ad personalisation.
-
Improving Your Recommendations on Our Apps With AI at Meta
Meta · Article · Oct 2025
Meta's announcement that chats with Meta AI shape the ads and posts you see from 16 December 2025, except sensitive topics.
All 10 sources
-
How to change or delete your information from chats with AIs from Meta
Meta · Guide · 2026
The /reset-ai and /reset-all-ais commands that delete Meta's copy of your chats on WhatsApp, Instagram and Messenger.
-
Business data privacy, security, and compliance
OpenAI · Article · 2026
OpenAI's statement that ChatGPT Business, Enterprise, Edu and the API are not used to train its models by default.
-
Is my data used for model training? (commercial products)
Anthropic · Guide · Aug 2026
Anthropic's page for Claude for Work and the API: no training on inputs or outputs by default, and what happens when you send feedback.