Short answer

Most AI chat services have a setting that controls whether your conversations help train their models, but it rarely covers everything. Google says Gemini chats with Keep Activity off are not used to train AI models but are still kept for 72 hours. Anthropic says consumer chats are used only if you opt in to its Model Improvement setting, with exceptions such as safety review.

Key takeaways

  • Opting out of training and deleting your chats are separate things. Check both settings.
  • Google says Gemini chats with Keep Activity off are not used to train AI models unless you send feedback, but are still kept for 72 hours.
  • Google says chats already reviewed by service providers are retained for up to 3 years even if you delete your activity.
  • Anthropic says consumer chats are used for model improvement only if you choose to allow it, with exceptions including safety review and feedback.
  • Free tools may use your inputs for training, and some paid or business plans let you opt out, according to UK DfE guidance.

Opting out is a set of choices, not a switch

Many AI services use conversations to improve their models unless you tell them not to, while others only do so if you opt in. The wording differs from company to company, and so do the exceptions. This article covers two services whose official help pages we opened: Google’s Gemini and Anthropic’s Claude. For other tools, find the service’s own privacy help page and look for the same ideas: a training or model improvement setting, a temporary chat mode, a history setting and a deletion tool.

UK Department for Education guidance makes the baseline point for schools: free tools may use inputs to train their models, while some paid tools let users opt out. The rule of thumb is to assume a free tool may learn from what you type until you have checked.

Google Gemini

Google’s Gemini Apps help page directs you to the Gemini Apps Activity page and its “Manage and delete your activity” article. It does not give click-by-click menu steps, so start there. It says the following.

  • With Keep Activity off, future chats won’t appear in your Activity and won’t be used to train AI models, unless you send feedback.
  • Those chats are still kept for 72 hours so Gemini can respond, process feedback and protect Google, its users and the public.
  • Some features stop working. Connected apps such as Google Workspace are unavailable, and remote browser and computer tasks require Keep Activity to be on.
  • Chats already reviewed by service providers aren’t deleted when you delete your activity. They are retained for up to 3 years.
  • Audio and Gemini Live videos or screenshares are not used to improve Google services by default. If you turn on the separate option, Google can use them, including for AI model training.
  • Your Gemini settings do not control processing of your chats to create anonymised data for improving Google services.

Anthropic Claude

Anthropic’s privacy page, dated 16 March 2026, covers consumer plans. It says chats and coding sessions may be used to improve Claude only in specific cases:

  • When you choose to allow it through the Model Improvement setting in Privacy Settings. The page does not give the steps, and links to a separate guide.
  • When a chat is flagged for safety review. These may be used, including to train models for Anthropic’s Safeguards team.
  • When you explicitly join a program such as the Trusted Tester Program.
  • When you use thumbs up or down. The page says the full conversation is kept for up to five years, de-linked from your user ID, and may be used for training.

It also says Incognito chats are not used to improve Claude even with Model Improvement enabled, and that raw content from connectors such as Google Drive or MCP servers is excluded unless you paste it directly into a chat.

What opting out does not do

Opting out of training is about future model improvement. It does not promise that no one ever sees your chat, that every copy is deleted, or that anything already collected is withdrawn. As both pages show, safety review, short-term retention and feedback have their own rules.

That is why the most reliable protection is still what you type. Our guides to what not to tell AI chatbots and AI chatbot privacy settings and data retention cover the habits. If you connect an assistant to your files, email or tools, read about AI agent permissions and risks too.

If sensitive information has already gone into a chatbot, delete the chat, review the service’s retention rules, and follow our data breach checklist if it was something like a password or ID number.

Where we could not verify

We could not open the equivalent help pages for several other services, so we have not described their settings. Settings and policies change, so check the vendor’s help page for the current steps.

What you can do

  1. 01Open the privacy or data controls in each AI service you use and find the setting about improving or training models.
  2. 02Turn that setting off for chats you do not want used, and read what the service says it does not cover.
  3. 03Use the service's temporary or incognito mode for sensitive questions, where the service offers one.
  4. 04Avoid giving thumbs up or thumbs down or sending feedback on private chats, because feedback can be handled differently from normal chats.
  5. 05Delete old chats you no longer need, and check separately how long the service keeps them.
  6. 06Do not type in what you cannot afford to expose. Opting out reduces use, it does not erase risk.

Related reading: AI voice clone scams and the data breach tracker.

Frequently asked questions

If I turn off training, is my chat deleted?

No. Google says that with Keep Activity off, future Gemini chats are still kept for 72 hours so it can respond, process feedback and protect Google and its users. Deleting and opting out are separate controls.

Can humans read my AI chats?

Google says chats may be reviewed by service providers, and that chats already reviewed are not deleted when you delete your activity but are retained for up to 3 years. Anthropic says chats flagged for safety review may be used, including to train models for its Safeguards team.

What about incognito or temporary chats?

Anthropic says its Incognito chats are not used to improve Claude, even with Model Improvement enabled. Other services have similar modes, so check each service's own help page for what the mode covers.

Do business and API accounts follow the same rules?

Not necessarily. Anthropic's page covers consumer plans and refers readers to a separate article for commercial products. The DfE notes some paid tools let users opt out. Check the terms for your plan.

Sources

  1. Gemini Apps privacy and Keep Activity settings, Google Gemini Apps Help
  2. Is my data used for model training?, Anthropic Privacy Center, 2026-03-16
  3. Generative artificial intelligence (AI) in education, UK Department for Education (GOV.UK)

This article reflects public information as of 2026-11-06. AI tools and rules change quickly; we update articles when new facts are published. Spot an error? See our editorial and corrections policy.