What happens to what you type into ChatGPT, Gemini, Claude and other AI chatbots

What you type into an AI chatbot leaves your phone or computer and goes to the company’s own computers, where the answer is written. The chat is then usually saved in your account until you delete it. Depending on the assistant and your settings, it may also be used to teach future versions of the program, and a few staff members may read some chats to check for misuse or to improve the answers. For seven of the nine assistants on this page, the maker’s own pages say chats are used for this unless you say no. For Copilot only the pages of the older app say so, and the new app’s pages do not; xAI describes a way to say no without stating the default. Deleting a chat removes it from your list at once, but copies can stay on the company’s systems for up to 30 days, and longer in some cases.

This site is written with AI tools, including Claude by Anthropic, one of the assistants compared here. The comparison uses only the makers’ published information, with links. Every setting name and time limit below is quoted as it appeared on the makers’ pages on 31 October 2026; they change often, so the “Sources checked” date above is the one that counts.

What happens to a message after you press send?

Four things can happen, in this order.

It travels to the company. The program that writes the answers, a large language model, runs on the company’s computers, not on your phone. Your message has to go there to get an answer.
It is saved. Regular ChatGPT chats stay in your account until you delete them. Google deletes Gemini activity older than 18 months by default, and you can choose 3 or 36 months instead, or no automatic deletion.
It may be checked. Automatic filters, and sometimes people, look for messages that break the rules (more on this below).
It may be used for training. Training is how the program learns: it is shown huge amounts of text and gradually gets better at writing answers. Your chats can become part of that text unless you switch it off. Switching it off stops future use, but Anthropic and Meta both say it does not take your data out of training that has already started, and Perplexity says training data it has already collected cannot be removed.

Which assistants train on your chats, and where is the switch?

The table lists, in alphabetical order, what each maker says and where it puts the setting. “Not stated” means the maker’s pages opened on 31 October 2026 did not say.

Assistant	Used for training? (as the maker states it)	Setting, and where it is	Private or temporary chat
ChatGPT (OpenAI)	Yes, on by default for personal Free, Plus and Pro accounts (OpenAI)	Settings > Data controls > “Improve the model for everyone”	Temporary chat: not used for training; a copy may be kept up to 30 days for safety
Claude (Anthropic)	Yes, unless you opt out: the plans page lists model training as “Opt-out” for Free, Pro, Max and Ultra; Anthropic’s help page says chats are used if you allow it	Settings > Privacy > “Help Improve our AI models” (Anthropic)	Incognito chat (ghost icon): not used for training; kept 30 days
Copilot (Microsoft)	Older app: yes, unless you opt out; not for under-18s or people not signed in, and Italy is not among the excluded countries (Microsoft). New app (since 18 August 2026): not stated	Microsoft says you can control it, but names no setting; the new app’s privacy controls list no training switch	Not stated
DeepSeek	Yes, unless you turn off “Improve the model for everyone”: inputs and outputs, de-identified, to “develop or improve the Services” (terms, point 4.3)	“Improve the model for everyone”; the path in the app is not described	Not stated
Gemini (Google)	Yes, if you are 18 or over: “Keep Activity is on by default”	gemini.google.com > Activity > “On” > “Turn off”	Temporary chat: not used for training; kept 72 hours (Google)
Grok (xAI)	“We may use your content”; default not stated (xAI)	App: Settings > Data Controls > “Improve the model”; grok.com: Settings > Data > “Improve the Model”	Private Chat (ghost icon): not used for training; deleted within 30 days
Le Chat, now Vibe (Mistral)	Yes: “users are not opted out by default” (Mistral)	Phone app: Settings > Data & Account Controls > untick “Enable data sharing”; web: “Allow your interactions to be used to train our models” in the Admin panel	Not stated
Meta AI	Yes: your interactions with Meta’s AI are listed among its training sources (Meta)	No switch: an objection request from your Facebook, Instagram or Meta account, and a separate one from your WhatsApp account	Not stated
Perplexity	Yes, on by default for Free, Pro and Max (Perplexity)	Account settings > Preferences > “AI data retention”	Incognito: sessions “expire after 24 hours”; training not stated

Two details catch people out. First, a thumbs up or thumbs down can count as permission: OpenAI says the whole conversation linked to that feedback may be used for training even after you opt out, and xAI and Google also say that feedback you send may be used even when the setting is off. Anthropic says it keeps the whole conversation behind a thumbs up or down for up to 5 years and may use it to train its models. Second, Meta says that private messages with friends and family are not used to train its AI unless you or someone in the chat shares them with Meta AI.

How long are chats kept after you delete them?

Only some makers give a number. In their words, as of 31 October 2026:

ChatGPT: removed from your list at once and scheduled for permanent deletion within 30 days, unless already de-identified or kept longer for security or legal reasons. It cannot be restored (OpenAI). Archiving a chat does not delete it.
Claude: removed from your history at once and from Anthropic’s storage within 30 days. If you allowed training, de-identified data may stay up to 5 years in the training process; chats flagged for breaking the rules are kept up to 2 years (Anthropic).
Copilot: the older app stores conversations for 18 months by default, and you can delete them at any time. For the new app no period is published.
Gemini: with Keep Activity off, chats are still saved for up to 72 hours. Chats already read by reviewers are not deleted when you delete your activity: Google keeps them up to three years, separated from your account.
Grok: removed from xAI’s systems within 30 days, unless de-identified or kept for safety, security or legal reasons.
Le Chat (Vibe): kept until you delete the conversation or your account, says Mistral’s privacy policy (in force since 3 September 2026). No period after deletion is given.
Meta AI: no period is given. Meta’s EU terms for its AI warn that deleting messages, whole chats or even your account may not delete Meta’s copy, and point to its Privacy Center for the right way to do it.
DeepSeek and Perplexity: their privacy policies (DeepSeek, Perplexity) say data is kept as long as necessary, without a period for deleted chats.

Do people at the company read your chats?

Some can, for limited reasons. OpenAI’s terms say it uses “automated systems and human review” to find content that breaks its rules. Google says “a subset of chats are reviewed by human reviewers”, and that reviewed data is disconnected from your account. In its FAQ for the older Copilot app, Microsoft says some conversations get automated and human review, and that you cannot opt out of human review. xAI says “a limited number of our authorized personnel” may review Grok chats. Meta’s EU terms say its checks of chats with its AI can be automated or done by people. Anthropic says that a small number of staff working on training can see the chats you allow it to use, after your user ID, such as your email address, is removed, and that chats flagged by its safety systems may still be used for its safety work even when training is off (Anthropic). Mistral’s privacy policy says authorized team members can access personal data when their job requires it. DeepSeek’s and Perplexity’s pages say nothing either way.

Sharing is a separate risk. A shared link to a chat can be opened by anyone who has it, and xAI warns that a link posted publicly may turn up in Google results.

What should you never paste into a chatbot?

The makers say it themselves. Google: “Please don’t enter confidential information that you wouldn’t want a reviewer to see.” xAI: “please do not share any personal or sensitive information”. Meta asks you not to share anything you do not want its AI to keep and use, such as information on sensitive topics. Microsoft’s Copilot FAQ says you shouldn’t provide “any confidential or sensitive personal data” you would not want Microsoft to use. DeepSeek says not to provide sensitive personal data, and Anthropic advises caution with financial details, health records, passwords and confidential documents.

In practice, leave out:

Passwords, PINs and one-time codes, and full card or bank account numbers.
Health records and test results with your name on them.
Identity documents: ID card, passport, tax code, scans of signatures.
Other people’s personal details, such as a friend’s diagnosis or a client’s address. DeepSeek asks you to get a person’s consent first.
Work secrets: contracts, customer lists, unpublished figures. At work, follow your employer’s rules; on a ChatGPT business account, OpenAI’s privacy policy says administrators can see your content.

If you need help with such a document, replace names, addresses and numbers with made-up ones before pasting it.

What rights do you have in the EU, and where can you complain?

The EU’s data protection law, the GDPR (Regulation (EU) 2016/679), gives you the right to see the data a company holds on you (article 15), to have it deleted (article 17) and to object to some uses of it (article 21). Italy’s data protection authority, the Garante, explains that you can write to the company without any special form, and that it must answer within one month, a period that can be extended by two months.

Start with the app: most of the nine let you download or delete your data from their settings.
If that is not enough, use the form or email address in the service’s privacy policy.
If the answer is late or unsatisfactory, you can file a free complaint with the Garante by certified email (PEC), registered letter or by hand. Outside Italy, your own country’s authority takes complaints.

The Garante has acted on a chatbot before: on 30 January 2025 it ordered the two DeepSeek companies to limit, with immediate effect, the processing of data of users in Italy, and opened an investigation. DeepSeek’s privacy policy says it stores personal data in the People’s Republic of China. For legal disputes, a lawyer or the Garante can tell you what applies to your case.

Where does this answer stop?

Pages are not always clear. Microsoft has not said, on the pages opened for this guide, whether the new Copilot app trains on chats. Grok’s pages describe the switch without stating in words whether it starts on. Open your own settings and look.
Work and school accounts follow other rules. OpenAI, for example, says it does not train on business workspaces by default; your organisation’s policy decides.
Rules change. Several of these pages were updated in the last weeks of October 2026. This page is checked every month.

Where to go next

AI for everyone: what AI is and how to start using it safely.
AI assistants compared: prices, apps and data settings of the same nine assistants side by side.
Why is ChatGPT asking for my ID or a selfie?: the age check, who runs it and what it keeps.
What is an AI agent?: what changes when the AI can also click and type for you.

Common questions

Does ChatGPT keep my chats after I delete them?

For a short time. OpenAI says a deleted chat disappears from your list at once and is scheduled for permanent deletion from its systems within 30 days, unless it was already de-identified or OpenAI must keep it longer for security or legal reasons. A deleted chat cannot be restored, not even by OpenAI support.

How do I stop ChatGPT from using my chats for training?

Open Settings, choose Data controls and turn off Improve the model for everyone. The choice applies to your account on the web and in the app, and it does not delete your saved chats. If you later press thumbs up or thumbs down on an answer, OpenAI says the whole conversation linked to that feedback may still be used for training.

Is ChatGPT safe to use for private matters?

It depends on what you write and on your settings. Your messages are saved on OpenAI's computers, may be checked by automatic systems and people for misuse, and are used for training unless you turn it off. For anything you would not want a stranger to read, such as health records, passwords or other people's details, several makers advise leaving it out.

Can my employer see my ChatGPT chats?

If you use a work account, yes, it can. OpenAI's privacy policy says the administrators of a ChatGPT Enterprise or business account may access your account, including your content. A personal account you pay for yourself is not managed by your employer, but your company's own rules on using AI still apply.

Does Meta AI on WhatsApp read my private messages?

Meta says it does not use your private messages with friends and family to train its AI, unless you or someone in the chat shares those messages with Meta AI. What you write to Meta AI itself is a different matter: Meta lists those interactions among its training sources, and you can object from your WhatsApp account.

Is DeepSeek safe to use in Italy?

DeepSeek's privacy policy says it collects, processes and stores personal data in the People's Republic of China. On 30 January 2025 Italy's privacy authority, the Garante, ordered the two DeepSeek companies to limit the processing of data of users in Italy and opened an investigation. Check the Garante's website for the current status before you use it.

Topics: privacy, ChatGPT, Gemini, Claude, GDPR, data protection