What AI chat sites do with your chats

Yes. OpenAI may use chats from personal accounts to train its models unless "Improve the model for everyone" is turned off. Temporary chats are not used. A chat you rate with thumbs up or down may be used even after you opt out.

Chats stay in the account until you delete them. A deleted chat disappears at once and is removed from OpenAI's systems within 30 days, unless it must be kept longer for security or legal reasons. Temporary chats may be kept for up to 30 days.

Turn off "Improve the model for everyone" in Settings under Data controls. New chats are then not used for training and still stay in your history. Choosing "Do not train on my content" in OpenAI's Privacy Portal does the same.

A single chat can be deleted from the chat history, and all chats at once with "Delete all chats" in Settings under Data controls. Deleting is permanent and cannot be undone; OpenAI removes the chat from its systems within 30 days.

Anthropic asks you to choose: new users pick during signup, and the choice can be changed later. Chats are used to train its models only if you allow it. Chats flagged for safety review, and chats you rate with thumbs up or down, may be used either way. Incognito chats are not used.

Chats stay until you delete them: a deleted chat leaves your history at once and Anthropic's systems within 30 days. If you allow training, de-identified data may be kept for up to 5 years. Chats flagged as breaking the usage rules may be kept for up to 2 years.

In Settings, open Privacy and switch off "Help Improve our AI models". New chats are then not used for future training, and stored chats are left out of later training runs. Chats flagged by safety checks may still be used for safety work.

Chats can be deleted one at a time or several at once from the chat list. A deleted chat leaves your history immediately and is removed from Anthropic's systems within 30 days. Deleting the account removes all of its data.

Yes. "Keep Activity" is on by default for users 18 or over, and with it on Google uses chats to improve its services, including training its AI models. Human reviewers read some chats. Temporary chats are not used for training.

Saved chats are deleted automatically after 18 months by default; this can be set to 3 or 36 months, or switched off. Chats read by human reviewers are kept for up to 3 years even if you delete your activity. With "Keep Activity" off, chats are kept for 72 hours.

Turn off "Keep Activity" in Gemini Apps Activity. Future chats are then not used to train Google's AI unless you send feedback, but they are also no longer saved in your activity and are kept for only 72 hours. Temporary chats are not used for training either.

A single chat can be deleted from the side panel, which also removes it from Gemini Apps Activity. In Gemini Apps Activity you can delete everything or a chosen period. Chats already read by human reviewers are not deleted and are kept for up to 3 years.

Yes. With free use of AI Studio, Google uses what you submit and the responses to improve and develop its products and AI models, and human reviewers may read it. This does not apply if the account has a Cloud project with active billing, or in the EEA, Switzerland and the UK.

Their policy doesn't clearly say. Google states that it keeps prompts and responses for 55 days to detect misuse, but gives no period for saved AI Studio chats or for content it uses to improve its products.

The terms describe no setting that turns this off. Content is left out of product improvement only when AI Studio counts as a paid service: the account has a Cloud project with an active billing account, or is a Workspace enterprise account. The same holds in the EEA, Switzerland and UK.

Their policy doesn't clearly say. Google's terms and developer documentation for AI Studio mention saving a prompt to work on later, but do not explain where saved chats are stored or how to delete them.

Yes, unless you turn it off. On Free, Pro and Max accounts the "AI data retention" setting is on by default, and while it is on Perplexity may use your data to train its AI models. Enterprise data is never used for training.

Chats stay while your account is active. If you delete the account, your data is removed within 30 days. Incognito chats expire after 24 hours, and chats made while signed out expire after 14 days.

Turn off "AI data retention" on the Preferences page of your account settings. It only covers data collected after you switch it off; data already collected for training cannot be removed.

You can delete one chat or all of them from the History page. Deleting your whole account removes all account and personal data within 30 days.

Yes, unless you turn it off. DeepSeek says it may use what you type and the replies to train and improve its models, to a small extent and after removing details that identify you.

DeepSeek keeps what you type for as long as you have an account. It gives no period for how quickly deleted chats are removed, and says it may keep data longer for legal reasons or when its rules were broken.

Turn off the setting called "Improve the model for everyone". The official pages name it but do not say where it sits in the menu. The privacy policy also gives a right to opt out of training, which can be requested by e-mail.

You can delete your chat history yourself in the settings, and you can delete your account. A deleted account cannot be brought back, and DeepSeek may still keep some data where the law requires it.

Microsoft's pages say two things. Its current Copilot help says prompts, replies and file contents aren't used to train the foundation models Copilot runs on. Its general Privacy Statement says Microsoft may use your data to develop and train its AI models.

Their policy doesn't clearly say. The current help pages say Copilot keeps past chats so you can come back to them, and give no time limit. The pages for the older app version said 18 months.

Their policy doesn't clearly say. The current Copilot help pages describe no setting for turning model training off. The pages for the older app version described an opt-out; Microsoft says those no longer apply to the updated app.

You can delete single chats from the Chats list in Copilot. To delete everything at once, use "Delete all activity history" in the Microsoft privacy dashboard.

Yes, unless you turn it off. The company behind Grok says it may use your prompts and Grok's replies to train its models. Conversations in Private Chat are not used for training.

Chats stay in your account for as long as you want. Chats you delete, and Private Chat conversations, are removed within 30 days unless they have to be kept for safety, security or legal reasons.

On grok.com, switch off "Improve the Model" under Settings, then Data. It applies to new conversations only, and feedback you choose to send later may still be used. Private Chat conversations are also left out of training.

You can delete one conversation with the trash button on the conversation history page, or all conversations from Settings. Deleted data takes up to 30 days to be removed from the company's systems.

Yes. On personal Vibe accounts, what you type and the replies are used to train Mistral's models by default, until you opt out. Thumbs-up and thumbs-down feedback is used on every plan. Enterprise workspaces are opted out by default.

Chats are kept until you delete the conversation or your account. After a deletion, Mistral may still hold the data in its back-end systems for a time it does not specify, for example to enforce its usage rules or because the law requires it.

A setting turns it off. On the web it is in the Admin panel, under Vibe and then Privacy: "Allow your interactions to be used to train our models". In the mobile app it is the "Enable data sharing" checkbox. Thumbs-up or thumbs-down feedback is still used.

Yes, one conversation at a time: open the three-dot menu beside the conversation's title, choose Delete and confirm. It cannot be undone. Deleting the whole account is also possible from the account settings.

It depends on the bot. Poe says its official bots, and prompt bots built on OpenAI, Anthropic, Google and Meta models, do not train on your chats. Bots and apps made by third-party developers, shown with a half-shaded privacy shield, may store chats and use them for training.

Poe gives no time limit for ordinary chats: Quora keeps personal information for as long as it has a relationship with you, and chats you delete are removed from Poe's systems. Temporary chats are erased within 24 hours. The AI providers keep chats on their own schedules.

Poe describes no switch for this. Each bot or app has a privacy shield in its info card: fully shaded means the AI providers will not train on the chat, half shaded means it may be used for training. Giving a bot new permissions can change its shield to half.

Yes. "Delete all chats" in Settings clears the whole history, and single messages can be deleted inside a chat. Poe says deleted chats are removed from its systems. The account can be deleted in Settings too; logging in within 14 days cancels that.

No. Hugging Face says it does not store any user data for training purposes. The models are run by outside providers, chosen per model; each provider handles data under its own policy, which the notice tells you to consult.

Your conversation history is stored so that you can come back to past chats. The notice gives no time limit; you can delete any conversation at any time. Logs kept for debugging last up to 30 days and, Hugging Face says, hold no user data.

Nothing to turn off: Hugging Face says it does not store user data for training. The outside providers that run the models follow their own policies.

You can delete any conversation at any time from the HuggingChat interface. The notice does not describe the steps.

Yes. Character.AI says it uses what users submit, including chat content, to train and improve its AI models. It says it first separates the content from account details and removes directly identifying information.

Their policy doesn't clearly say. It says information is kept for as long as needed for the purposes it was collected for, and gives no period for chats.

A setting called "Improve the Model for Everyone" stops new chats being used to train the generative AI models. The help page gives the steps only for users in the EEA and UK; the notice for US, EEA and UK residents says the opt-out can be used at any time.

The official pages describe deleting the whole account, which is permanent, and asking Character.AI to delete personal data. They do not describe a way to delete one chat or the whole chat history on its own.

Yes. Qwen says it may use chat content and feedback to train and improve its AI models, after removing details that could identify the person. It says raw input is not used directly for training.

Their policy doesn't clearly say. It says personal data is kept only as long as needed to provide the service, with no period for chats, and that a chat rated with like or dislike is stored as feedback.

Only by e-mail. The terms say a user can opt out of model training by writing to [email protected]; after that, chat content is no longer used for training, but feedback given on answers still is. No setting in the product is described.

The official pages describe only deleting the whole account, from the Qwen Settings page, or asking for personal data to be deleted at [email protected]. They do not describe deleting a single chat or the chat history on its own.

Yes. Kimi's international policy says what users type, upload or generate is processed to provide and improve its services, including training and optimising its AI models.

Their policy doesn't clearly say. It says personal information is kept only as long as needed for its purposes, with no period for chats, and is deleted or anonymised after a request to delete the account.

Their policy doesn't clearly say. The international policy describes no setting or form that keeps chats out of training. On the mainland-China service (kimi.com), the policy says a user can ask customer service to stop it.

Only in part. The international policy says to write to [email protected] about deleting personal information, and that the whole account can be deleted for good. It names no way to delete one chat; the kimi.com policy mentions a "Delete Chat" button in "Chat History".

Yes. Yandex says it may train Alice's AI models on data about how you use it, such as your chat history, and its terms let it use your requests to improve its services. This applies unless you switch it off in Yandex ID.

On the free plan, chat history is kept for 14 days. With the paid Алиса Плюс option, chats are kept with no time limit and generated images for 30 days. Yandex's privacy policy sets no fixed period for other data: it is kept as long as needed for its purpose.

Yes. In Yandex ID, open the data access page ("Доступы к данным"), choose "Алиса" and switch off "Помогите Алисе стать лучше". Only chats after the change are left out of training.

Yes, but only one chat at a time: the help page says all chats cannot be deleted at once. Each chat is deleted from its menu in the chat list. Yandex ID also has a data management section for deleting data that Yandex services hold.

Their policy doesn't clearly say. The giga.chat user agreement gives Sber a broad licence to use what you send but never mentions training. Sber's separate terms for GigaChat inside its banking apps do allow training on requests that contain no personal data.

Their policy doesn't clearly say. The help page says a signed-in user's chats are kept in one place so they can be reopened, but none of the official pages we could read gives a storage period.

Their policy doesn't clearly say. The user agreement and help pages describe no setting or form for keeping chats out of training; the agreement only says the service is not meant for requests that contain personal data.

Yes. A chat can be deleted with the bin icon inside the open chat, or from the three-dots menu on it in the chat list. The help page does not mention deleting all chats at once, or say when deleted chats are removed from Sber's servers.

No. DuckDuckGo says chats are not used to train AI models, either by itself or by the model providers it sends them to (such as OpenAI and Anthropic). The one exception is a chat you choose to share when giving feedback: that one is stored and reviewed to improve its products.

By default chats are saved only in your browser, not on DuckDuckGo's servers. Model providers must delete them after replying, within 30 days at most, with exceptions for safety and legal reasons. Optional Sync & Backup keeps encrypted copies, deleted after 18 months without use.

There is nothing to turn off: Duck.ai chats are not used for training. DuckDuckGo only stores a chat if you choose to share it when giving feedback.

Yes. One chat can be deleted from the icon or menu beside it, and the Fire Button inside Duck.ai deletes all chats at once. Pinned chats are not removed by the Fire Button; they have to be unpinned or deleted one by one.

Yes. Inflection AI, the company behind Pi, says it uses the information it collects, including what you type or say to Pi, to develop and train the AI models behind its services. This applies unless you opt out in your account settings.

Chats are kept while your account is active. After you delete them, what you typed is generally erased within 15 days, though it can be kept longer for security, fraud, legal or financial-record reasons. Pi's replies may be kept indefinitely for limited purposes.

Yes. The privacy policy says model training can be turned off on the Account settings page, and the help centre says this can be done at any time. Neither page gives the name of the setting.

Yes. Single messages can be deleted from the conversation. To erase everything, delete the account in Settings or type "!delete" in the chat; that removes the whole chat history together with the account.

Yes. Z.ai's terms let it use what individual users type to improve its services and to develop its AI technology, and its privacy policy lists training and improving its models as a purpose. Content sent through the API is not used this way unless the customer agrees.

What a user types is kept for as long as they have an account. A conversation the user deletes leaves the chat history at once and is then deleted from Z.ai's servers automatically; the policy does not say how long that takes.

Their policy doesn't clearly say. It mentions a right to object where local law gives one and invites users to contact Z.ai with concerns, but describes no setting or form for keeping chats out of training.

Yes. Single conversations can be deleted, and the whole chat history can be deleted in settings. A deleted conversation leaves the history at once and is then removed from Z.ai's servers automatically.

Yes. Arena passes chats to the companies whose AI models appear on it, to evaluate, improve and develop their products, and may publish them in public research datasets. Its FAQ says shared data is de-identified; its policy warns against entering anything you would not want public.

Their policy doesn't clearly say. It keeps personal data for as long as it needs it for its purposes, with no period given; a chat you delete is usually removed from its systems within 30 days.

Their policy doesn't clearly say. The choices it lists cover marketing e-mail and cookies; it names no setting or form that keeps chats from being passed to AI providers or published.

Yes, one chat at a time: archive the chat, then delete it from Search. Arena says the data is then queued for deletion from its systems, usually within 30 days, though it may keep limited information where the law requires or for security and fraud prevention.

No. Proton says Lumo does not use chats to train AI models. The one exception is feedback you choose to send: an anonymised copy of that prompt and reply goes to the Lumo team and, for the Apertus model, may be shared with the universities behind it for training.

On the Free plan, saved chats are kept for 7 days only; Lumo Plus keeps more history, with no period stated. Saved history is encrypted so that Proton cannot read it. Guest chats are erased when the session ends, and chats in "Ghost Mode" are not saved at all.

Nothing to turn off: Lumo does not train on chats. A prompt and reply are passed on only when you choose to send feedback on an answer, so not sending feedback keeps them out.

Their policy doesn't clearly say. It says personal data can be accessed, edited, deleted or exported directly from the account interface, but no official page read explains how to delete a chat or the whole history.

No. Venice says it does not train models on what users type and does not store prompts or replies on its servers. If you pick a model marked "Anonymous" (outside providers such as GPT, Claude or Gemini), that provider receives the text and handles it under its own rules.

Venice says it does not keep chats on its servers: the history is stored only in your browser on your device. Exceptions it names: generated videos are held until you download them, optional Pro encrypted backups expire after 90 days, and providers of "Anonymous" models may store prompts.

Nothing to turn off: Venice says it does not train on chats. Models marked "Private" (the default) run where no prompt or reply is stored; models marked "Anonymous" send the text to an outside provider that applies its own rules.

Yes. Chats exist only in your browser, so deleting your chat history in Venice, or clearing the browser's data, removes them for good. Venice's pages give no menu path for this. The account itself can be deleted in Settings.

No. T3 Chat says it does not train AI models itself and has switched off training with the model providers it uses wherever it can. Some new models can only be used with training allowed; it says these are clearly marked so you can decide whether to use them.

Their policy doesn't clearly say. It lists prompts and uploaded files among the data it collects, but gives no period for how long chats are kept.

Nothing to turn off: T3 Chat says chats are not used for training. Models that only work with training allowed are marked, so avoiding those keeps chats out. With your own API key, the provider's training setting is yours to manage.

Their policy doesn't clearly say. The terms mention that deleted content may be irretrievable and that the account can be cancelled, but no official page read explains how to delete chats or what happens to them afterwards.