Every major chatbot you use, ChatGPT, Claude, Gemini, Meta AI, has some version of a setting that controls whether your conversations get used to train future models. The problem is that these settings live in different menus, have different defaults, and in at least one major case do not exist at all for a huge chunk of users. If you have been meaning to lock this down but never got past the search results, here is what each toggle actually does, where to find it, and the parts of the process that quietly do not work the way you would assume.
ChatGPT: two separate settings, not one
In ChatGPT, go to Settings, then Data Controls, and turn off "Improve the model for everyone." This stops new conversations from being used in training going forward. It does not touch conversations you already had before you flipped it off.
There is a second, separate setting worth checking: "Reference chat history" under Personalization. Turning that off clears what ChatGPT remembers about you across sessions, but it is a different switch from the training toggle, so turning one off does not turn off the other. If you want a conversation to leave zero trace, use Temporary Chat instead. OpenAI has said temporary chats are not used for training, though the company has also said it may retain that data internally for a window of around 30 days for safety and abuse monitoring before deleting it.
Claude: the toggle also changes how long your data is kept
Claude's version lives under Settings, then Privacy, where you will find a toggle called "Help improve Claude." Since Anthropic opened this choice to Free, Pro, and Max users in August 2025, the setting does two things at once: it decides whether your chats are used for training, and it decides how long Anthropic keeps them.
If the toggle is on, Anthropic can retain your conversations for up to five years to use in training future models. If you turn it off, you drop back to a 30-day retention window used only for safety and technical purposes, then your data is deleted. One thing no toggle can undo: if your account had this setting on at any point in 2025 or 2026, that data may already be baked into models that have already been released. Turning the switch off stops future use, it does not pull your words back out of a model that already learned from them.
Google Gemini: turning it off does not mean turning it off completely
For Gemini, the control is called Gemini Apps Activity, found in your Google account's activity settings. You get two options: "Turn off" or "Turn off and delete activity." Picking the second one is the one that actually removes your past conversations instead of just stopping new logging.
Here is the catch that is easy to miss: even with Gemini Apps Activity fully off, Google has stated it may still retain conversations for up to 72 hours to generate responses and for safety checks, and it can still process your chats to create anonymized data used to improve its services more broadly. "Off" reduces what happens with your data. It does not guarantee nothing happens with it.
Meta AI: the setting most people cannot actually use
This is the one that surprises people. If you are a Meta AI, Facebook, or Instagram user based in the United States, there is currently no toggle or form that lets you opt out of having your public posts and interactions used for AI training. It simply is not offered to US accounts.
Users in the EU have had an "objection" form available since 2024, buried inside Meta's Privacy Center rather than exposed as a simple settings switch. Meta quietly extended the same form to Australian users in July 2025, more than a year after the EU version existed, and did not announce the change. Reports of it only surfaced later through users comparing notes. Where the form is available, private accounts and messages are already excluded automatically, and submitting it gets you an email confirmation that your public data will not be used going forward, not retroactively.
What none of these settings actually erase
Across every one of these companies, the same pattern repeats: the toggle controls future data collection, not what has already happened. If a chatbot already used a year or two of your conversations to help train a model that has since shipped, turning the setting off today does not reach back into that model and remove what it learned. Companies describe this as technically very difficult to undo, which is a polite way of saying it is not really undoable with current methods.
So the realistic goal is not "erase my AI footprint." It is "stop adding to it starting now," and separately, delete what visible chat history you can through each platform's history or activity controls, since that at least removes the readable record even if it cannot unteach a model.
The takeaway
The honest version of this guide is less satisfying than "flip these four switches and you are done." Three of the four major AI companies give you a real, if imperfect, opt-out. The fourth, Meta, simply does not offer US users one at all right now, which is worth knowing before you assume every platform works the same way. Flip the toggles that exist, do not assume any of them are retroactive, and do not be surprised if the Meta one is missing entirely when you go looking for it.