Every time you type into an AI chatbot, you share something: a work email, a health worry or a business plan. Many services can use those chats to train their next AI models unless you switch a setting off. This guide shows you the AI training opt out for each major service, with menu paths taken from each company’s own help pages where we could open them, checked in October 2026.
Most big chatbots now give you a switch, but each one has a different name, menu and set of limits. We checked the official help pages for ChatGPT, Google Gemini, Claude, Microsoft Copilot and Grok. We also explain what we could not confirm for Meta AI and LinkedIn.
Quick answer: Turn off Improve the model for everyone in ChatGPT, Help improve our AI models in Claude, Keep Activity in Gemini, both training switches in Copilot and the training option for Grok. These switches tell each company not to use your future chats for training. Most companies say they do not pull back data that was already used, and Microsoft says its Copilot setting also covers past chats.
What an AI training opt out really does
The official pages share a few common limits:
- It mostly works going forward. New chats stop being used. Anthropic says data already used in training is not pulled back out, but Microsoft says its Copilot setting also covers past chats.
- Chats may still be stored. Gemini keeps chats for up to 72 hours even with Keep Activity off.
- Feedback can still count. OpenAI says a thumbs up or thumbs down can send the whole conversation into training. The Grok FAQ says the same.
How to stop ChatGPT from training on your chats
OpenAI says that for personal services such as ChatGPT and Codex, “we may use your content to train our models.” Here is how to turn it off:
- On the web, open your account menu, then select Settings > Data controls.
- On iPhone or Android, open the sidebar, tap your profile icon to open Settings, then tap Data controls.
- Turn off Improve the model for everyone.
When you are signed in, the choice follows your account on all devices. Choosing Do not train on my content in the OpenAI Privacy Portal also works.
According to OpenAI’s data controls page, new conversations will not be used for training, but they stay in your history. The switch does not delete chats, and archiving a chat does not change whether it can be used.
For one private question, open a new chat and select Temporary. Temporary chats stay out of your history, do not create memories and are not used for training. OpenAI may keep them for up to 30 days for safety.
How to turn off AI training in Gemini and Claude
Google Gemini: Keep Activity
Google began renaming Gemini Apps Activity to Keep Activity after an update it announced in August 2025. If you are 18 or over, it is on by default. On a computer:
- Go to gemini.google.com.
- Click Settings & help, then Activity.
- Near the top, click On, then choose Turn off or Turn off and delete activity.
On Android, tap Menu, then your profile picture or initial, then Gemini Apps Activity. Google’s Android page still uses this older name for Keep Activity. Then finish the same way.
The Gemini Apps Privacy Hub says that with Keep Activity off, chats stay with your account for 72 hours and are not used for training unless you send feedback. Human reviewers read a subset of chats, and reviewed chats are kept for up to three years. With it off, only a few connected apps, such as Messages and WhatsApp, keep working.
For one private conversation, use a Temporary Chat, which Google says is not used for training. We did not find a separate EU only switch.
Claude (Anthropic): Help improve our AI models
Since an August 2025 update, Anthropic asks Claude users to choose whether their chats can train its models. New users pick at sign up. To change it:
- Click your name, then select Settings.
- Select Privacy.
- Switch off Help improve our AI models.
The steps are the same on mobile. The setting covers Claude Free, Pro and Max, plus Claude Code on those plans.
According to Anthropic’s Privacy Center, new chats and coding sessions will then not be used for future training. Chats flagged by safety systems may still be used to improve safety models. Anthropic’s retention page says data may be kept for up to 5 years with the setting on, and the usual 30 day period applies with it off.
For extra privacy, use the ghost icon at the top right of a new chat to start an incognito chat, which is not used for training.
Microsoft Copilot and Grok settings
Microsoft Copilot
Copilot has two switches, Training on conversation activity and Training on voice conversations. Turn off both:
- copilot.com: select your profile icon, then your profile name, then Privacy.
- Windows or macOS app: select your profile icon, then Settings > Privacy.
- Mobile app: open the menu, select your profile icon, then Account > Privacy.
The Copilot privacy FAQ says the change takes effect within 30 days. Microsoft does not train on chats from signed out users, users under 18, work or school accounts, or people in Brazil, China (excluding Hong Kong), Israel, Nigeria, South Korea and Vietnam.
One limit matters: Microsoft says the setting does not stop your chats from being used for other product improvements, advertising, safety, security and compliance.
X and Grok
If you use Grok inside X, the X Help Center gives this path:
- Open your X settings and select Privacy & Safety.
- Scroll to Data sharing and personalization.
- Select Grok and Third party Collaborators (X spells it with a hyphen), then Data Sharing.
- Turn off the option that lets your public data and Grok chats be used for training and fine tuning.
X notes a catch: X features powered by Grok can still learn from your everyday use. Making your account private also keeps your posts out of Grok training, and you can delete your Grok history in the same area.
On the Grok app or grok.com, the SpaceXAI consumer FAQ (SpaceXAI was called xAI until 2026) gives two paths. In the Grok app, open Settings > Data Controls and turn off Improve the model. On grok.com, open Settings > Data and turn off Improve the Model. Private Chat, where available, is not used for training. In some regions outside the EU and UK, logged out users cannot opt out. New to this assistant? Read what Grok AI is and how it works.
Meta AI and LinkedIn: what we could not fully confirm
Meta AI
We could not find an official Meta help page showing a simple switch to keep your Meta AI chats out of training. Meta’s own newsroom confirms two things:
- In April 2025, Meta said it would train AI in the EU on adults’ public posts and on interactions with AI at Meta, but not on private messages with friends and family. EU users received a link to an objection form.
- Since 16 December 2025, Meta uses your chats with its AI to personalize content and ads in most regions. This is separate from model training. Meta points to Ads Preferences and other feed controls, not an off switch.
If you live in the EU, Meta’s notice links to an objection form. UK users can find a right to object form in Meta’s Privacy Centre, which Which? describes for public posts. We could not confirm it covers AI chats. Elsewhere, keep personal details out of Meta AI chats.
We could not open LinkedIn’s help center to confirm the wording. Tech outlets reported this path in autumn 2025 (see PCWorld): Settings > Data Privacy > Data for Generative AI Improvement, then turn off Use my data for training content creation AI models. Reports say it is on by default. The Register reported that from 3 November 2025 it also covers members in the EU, EEA, Switzerland, the UK, Canada and Hong Kong, and that private messages are not included. LinkedIn may have changed the label since, so check your app.
AI training opt out settings at a glance
| Service | Where the setting is | Default |
|---|---|---|
| ChatGPT | Settings > Data controls > Improve the model for everyone | On for personal plans |
| Google Gemini | Settings & help > Activity > Keep Activity | On for users 18 and over |
| Claude | Settings > Privacy > Help improve our AI models | You choose at sign up |
| Microsoft Copilot | Profile > Privacy > Training on conversation activity and Training on voice conversations | Not stated on the help page |
| Grok on X | Settings > Privacy & Safety > Data sharing and personalization > Grok and Third party Collaborators > Data Sharing | Not stated on the help page |
| Grok app and grok.com | App: Settings > Data Controls > Improve the model. Web: Settings > Data > Improve the Model | Used unless you turn it off |
| Settings > Data Privacy > Data for Generative AI Improvement | On, as reported by tech press | |
| Meta AI | No general switch found; EU users get an objection form | Not applicable |
Privacy tips that work with any AI chatbot
What never to paste into an AI chat
Google’s advice is plain: “Please don’t enter confidential information that you wouldn’t want a reviewer to see or Google to use to improve our services.” Keep these out of every chatbot:
- Passwords, PINs and one time login codes
- Bank card, account, tax or ID numbers
- Medical records or test results with your name
- Private details about other people
- Company secrets, client data or code you may not share
- Photos of your passport, driving licence or other ID
If you need help with a document, remove names, numbers and addresses first.
Use temporary or incognito chats
ChatGPT Temporary Chat, Gemini Temporary Chat, Claude incognito chats and Grok Private Chat are all kept out of training, according to each company. They are still stored for a short time: 72 hours for Gemini, and up to 30 days for ChatGPT, Claude and Grok.
Delete your history and check again later
Anthropic says a deleted Claude conversation will not be used for future training. Gemini can auto delete activity after 3, 18 or 36 months, and Copilot, which keeps history for 18 months, lets you delete it at any time. Settings get renamed over time, so review these menus every few months.
Frequently asked questions
Does turning off AI training delete my old chats?
No. The switch does not delete chats. For most services it only changes how your chats are used from now on, and Microsoft says its Copilot setting also covers past conversations. To remove chats, delete them separately in each app.
Are work accounts used for AI training?
Usually not by default. OpenAI does not train on ChatGPT Business, Enterprise, Edu or API data by default. Anthropic’s consumer setting does not cover Claude for Work or the API, and Microsoft does not train on Copilot work or school accounts.
Can a person still read my chats?
Sometimes. Google says human reviewers read a subset of Gemini chats, and Anthropic may use chats flagged by safety systems for safety work. Assume a person could read anything you type.
Are the rules different in the EU and UK?
Partly. Meta gave EU users a form to object to AI training, and the Grok FAQ says logged out users outside the EU and UK may not be able to opt out. Microsoft skips training in some countries. The EEA and UK are not named in the list it gives.
Do I need to change the setting on every device?
OpenAI says the ChatGPT setting follows your account when you are signed in. For the other services, we did not find a statement either way, so check the setting in each app once.
Final thoughts
You do not need to stop using AI to protect your privacy. Spend ten minutes switching off training in the tools you use, then make temporary chats a habit for anything personal.
If you are still choosing a chatbot, compare our list of free AI tools like ChatGPT. You can also learn what an AI agent is or browse the best free AI tools.