Privacy guide

Does your AI assistant learn from your chats? How to turn it off

Most AI assistants use your conversations to improve their models unless you say no. Here is what each one does by default, how long it keeps your chats, and exactly where the off switch is.

By Chris · Updated Oct 2, 2026 · Facts checked Oct 2, 2026

Short answer

ChatGPT, Claude, Gemini, Copilot, Grok and DeepSeek all learn from your chats by default on personal plans. Meta AI uses your chats for ads and you can’t turn that off. Everywhere else, a single setting stops future chats being used.

Every assistant at a glance

Assistant	Learns from your chats by default?	How long it keeps them	Where to switch it off
ChatGPT	Yes, opt-out	Temporary Chats: up to 30 days	Settings → Data controls → “Improve the model for everyone”
Claude	Yes, opt-out	Up to 5 years if on; about 30 days if off	Settings → Privacy → Model Improvement
Gemini	Yes	18 months by default; reviewed chats up to 3 years	Gemini Apps Activity → turn off Keep Activity
Microsoft Copilot	Yes	18 months	Copilot privacy settings (never trains on under-18s)
Meta AI	Yes, and for ads	Not stated	No opt-out for ads reported
Grok	Yes	Not confirmed	Settings → Data Controls (future chats only) unconfirmed
DeepSeek	Yes; stored in China	“As long as necessary”	No named setting; a right to opt out by request

No label = confirmed on the company’s own help or privacy page. Checked Oct 2, 2026.

How do I stop ChatGPT using my chats for training?

OpenAI’s pricing page states it plainly for Free, Go, Plus and Pro: content is used to train its models, with an opt-out. OpenAI’s help center says that when “Improve the model for everyone” is off, your new conversations aren’t used to train its models. On the web, open your account menu, choose Settings → Data controls, and turn it off. On the phone app it’s in the same place, under your profile.

For a one-off private question, start a Temporary Chat. It doesn’t appear in your history, doesn’t update ChatGPT’s memory, isn’t used for training, and is deleted within about 30 days. If you save it, it becomes a normal chat again.

Does Claude train on your data?

Anthropic’s privacy center says it uses consumer chats for training in three cases: you allow it through the Model Improvement setting, a chat is flagged for a safety review, or you join a testing program. Its pricing page now lists model training as “Opt-out” on Free, Pro and Max, so assume it’s on until you switch it off. (In the 2025 terms update Anthropic asked every user to choose, and reporting at the time found the switch pre-set to on.)

The choice changes how long chats are kept: up to 5 years if Model Improvement is on, about 30 days if it’s off. Thumbs-up and thumbs-down feedback is kept up to 5 years either way. Incognito chats are never used for training, whatever the setting says.

How do I stop Gemini from using my data for training?

Google’s Gemini privacy hub is the most explicit of the big three. Keep Activity is on by default, chats are deleted automatically after 18 months unless you change it, and a sample is read by trained human reviewers. Those reviewed chats are kept for up to 3 years, separately from your account, even if you delete your activity. Google’s own advice is not to type anything you wouldn’t want a reviewer to see.

Turn off Keep Activity in Gemini’s activity settings, or use a temporary chat, which Google keeps for 72 hours.

Copilot, Meta AI, Grok and DeepSeek

Microsoft Copilot trains on consumer chats unless you opt out, keeps them for 18 months, and never trains on users under 18 or in a handful of countries, including Brazil, China and South Korea.

Meta AI lives inside WhatsApp, Instagram and Facebook. Since December 16, 2025 Meta has used those conversations to personalize the ads you see, and there’s no setting to stop it. Its new Muse agent also trains by default, though that part can be switched off.

Grok reportedly trains by default, with an opt-out under Settings → Data Controls that applies to future chats only. We couldn’t confirm this on xAI’s own pages.

DeepSeek’s privacy policy says it stores personal data in the People’s Republic of China and keeps it as long as necessary. There’s a right to opt out but no named setting. Italy blocked the app, South Korea paused downloads, and Australia, Canada, the US Navy and Texas ban it on official devices.

Five habits that matter more than any setting

Never paste passwords, bank details or ID numbers into any assistant. No setting makes that safe.
Swap real names for placeholders (“my colleague A”) when you ask for advice about people.
Use the private mode — Temporary Chat, Incognito, or a temporary Gemini chat — for health, money or legal questions.
Check your work’s rules first. Business and education plans usually don’t train on your data, but personal accounts do.
Re-check after big updates. Terms changed for Claude in 2025 and for Meta AI in December 2025; settings can move.

Still choosing an assistant? Our comparison of the seven most-used AI tools weighs privacy alongside price and features.

Sources: ChatGPT pricing (training row) · Claude pricing (training row) · OpenAI: Data controls · OpenAI: Temporary Chat · Anthropic privacy center · Anthropic: consumer terms update · Gemini privacy hub · Microsoft Copilot privacy FAQ · PCWorld on Meta AI ads · xAI FAQ · DeepSeek privacy policy · DeepSeek bans (Wikipedia)

Quick answers

Does ChatGPT use my chats to train its models?

Yes, unless you switch it off — OpenAI’s pricing page says content is used to train its models on every personal plan, with an opt-out. On the web: your account menu → Settings → Data controls → turn off “Improve the model for everyone”. On the app: Settings → Data controls. Temporary Chats are never used for training and are deleted within about 30 days.

Does Claude train on my data?

Yes, unless you opt out. Anthropic’s pricing page lists model training as “Opt-out” on Free, Pro and Max, and its 2025 terms update asked every user to choose. Switch it off under Settings → Privacy → Model Improvement. If it is on, chats can be kept up to 5 years; if it is off, about 30 days. Incognito chats are never used.

Does Gemini use my data for training?

Yes by default. Gemini’s “Keep Activity” setting is on unless you turn it off, and Google says a sample of chats is read by human reviewers and kept for up to 3 years, even if you delete your activity. Turn off Keep Activity in the Gemini app’s activity settings, or use a temporary chat.

Is DeepSeek safe to use?

It works well, but its own privacy policy says your data is stored in the People’s Republic of China, and several governments have banned it on official devices. Don’t put work, school, health or financial details into it.

Does turning training off delete my old chats?

No. The switch only affects future chats. To remove old ones, delete them from your history; some services still keep copies for a while for safety and legal reasons.

Can I stop Meta AI using my chats for ads?

No. Since December 16, 2025 Meta uses your Meta AI conversations to personalize ads, with no opt-out. The only way to avoid it is not to use Meta AI.

Keep learning

Plan guide