How to delete your ChatGPT history, and what stays behind

When someone looks up how to clear their ChatGPT history, they almost always want one of these three things — and they are not the same thing:

1. For it not to show in the sidebar, because other people use the computer.
2. For it not to train the model.
3. For it to be gone from the company’s servers.

The first takes two clicks. The second is a toggle in settings. The third is a legal process with small print. Let us take them in order.

1. Removing conversations from the list

In the sidebar, the three-dot menu on any conversation offers Delete. To empty the lot, Settings → Data controls has an option to delete all conversations at once.

That removes them from your view instantly. What happens afterwards, on the company’s side, is the part almost nobody reads: OpenAI states that a deleted conversation is removed from its systems within roughly 30 days, unless it has to be retained for legal or security obligations.

That exception is not theoretical. The litigation surrounding AI companies has produced data-preservation orders which, while they are in force, require the company not to destroy information even when the user has deleted it. The specifics come and go with each case, but the lesson is permanent: the delete button is an instruction to the company, not a physical guarantee.

2. Cutting off the training

This is the setting people most often confuse with the previous one.

Under Settings → Data controls there is a toggle along the lines of “Improve the model for everyone”. Turn it off and your conversations stop being used for training. They do not stop being stored. They stay in your account, available to you and held by the company, until you delete them.

Put the other way round: training and storage are two separate taps. Closing one does not close the other.

There is a third route halfway between the two, temporary chat: the conversation never enters your history and is not used for training, but it is kept for a limited period for abuse monitoring. It is the browser’s incognito mode in a different coat, with the same virtues and the same misunderstandings — the parallel is direct enough that it earns its own article: what private browsing actually hides.

3. Making it disappear from the company

At this point we are no longer talking about an app setting, but about a right.

If you are in the European Union, the right to erasure under the GDPR lets you demand that your personal data be deleted. AI companies run a privacy request form for exactly this. It is the same machinery described in how to delete your personal data from the internet: you file the request, there is a response deadline, and the company may argue grounds for keeping part of it.

The bluntest option is deleting the account. It is irreversible, takes a few days to complete, and takes everything with it: history, settings and subscription.

And in Claude, Gemini and the rest

The pattern repeats under different names:

• Claude (Anthropic). Conversations live in the account and are deleted from there. The policy on using conversations for training has changed over time and involves a user choice, so the sensible move is to open the privacy settings and see how yours stands today, rather than trusting what you read a year ago.
• Gemini (Google). The relevant control is called Gemini Apps Activity, inside your Google account. From there you can delete it, schedule automatic deletion, or switch it off. Google states plainly that human reviewers may read conversations to improve the service, and that this is why confidential information does not belong there.
• Copilot, Perplexity and the others: look for “data controls” or “privacy” in the account settings. They all have an equivalent, because European rules require it.

Rule of thumb: if an AI service is free and will not let you see or delete your history, you are not going to like the answer to what it does with your data.

What no amount of deleting fixes

One part of this does not depend on settings. If in March you pasted a contract with names and account numbers, that text already left your computer, travelled across the network and was processed on someone else’s server. Deleting it in August lowers your future risk; it does not undo the past.

Which is why the thing that genuinely protects you is not typing it in the first place. The technique is set out in why you should never paste private data into an AI, and it comes down to substituting before you paste: names become initials, account numbers become a row of X’s, addresses become “a town up north”. The model understands the problem just as well.

What you can do today, in five minutes

1. Open Settings → Data controls in whichever AI you use.
2. Look at the training toggle and decide whether you want it that way.
3. Delete the conversations that contain real data.
4. If something serious is in there — a contract, a payslip, a client’s details — file an erasure request through the privacy form as well.
5. From now on, clean texts before you paste them.

The short version

• Deleting the conversation, turning off training and erasing the data are three separate things.
• A deleted conversation is usually removed within about 30 days, absent a legal duty to keep it.
• Turning off training does not stop storage.
• Temporary chat is discreet, not invisible.
• The only thing you never have to delete is what you never typed.