How to Limit What Facebook Trains Its AI On

Direct link to Meta's AI data request form

The form is otherwise nine steps and three pop-ups deep.

Before you spend the time: outside the EU and UK, submitting this obliges Meta to nothing. The sections at the end explain what does work.

How to reach the form from inside Facebook

Nothing in Settings points at it, and nothing on the Privacy Center front page does either. You have to already know it exists. The walkthrough is here so you can see how far down they put it, and so the route still works if Meta moves the direct link.

1. Open Settings & privacy

Click your profile photo in the top right, then Settings & privacy.

2. Open Privacy Center

3. Under Privacy topics, choose the AI at Meta card

4. In the pop-up, choose Other types of information

Worth reading rather than clicking past. It is where Meta lists what it trains on: your public content, your interactions with AI features, and everything else.

5. Choose Rights related to how your information is used

This screen names the outside sources: public information scraped from the internet, information from partners such as search engines, and licensed third-party content.

6. Expand Privacy and generative AI

7. Click Learn more and submit requests here

The last line of a long wall of text about Meta's privacy principles, and the only thing on the page that goes anywhere.

8. Tick the checkbox at the bottom of the form

Note what the single available option says: that you have a concern about your personal information appearing in a response from an AI at Meta model. That is the only request this form accepts. There is no option to delete anything.

9. Fill in your details and press Send

What the form tells you about itself

Meta explains that third-party information is not linked to any Meta account and is not stored in a way that lets it identify whose data is whose, then concludes:

we don't provide the ability to access, download, correct or delete personal information from third parties that may be used to develop and improve AI at Meta.

It adds that requests are reviewed "consistent with your local laws". In the EU or UK, GDPR gives you a right to object that Meta must honour in most cases. In the US it depends on your state, and roughly half have no comprehensive privacy law to invoke.

Send it anyway. It costs about ten minutes, and an unfiled request achieves nothing with certainty. Just do not treat pressing Send as being done.

What actually works, wherever you live

These are not opt-outs, but they change what there is to train on.

• Make your account private. Meta trains on public posts and photos, so going private takes future posts out of scope.
• Lock down your back catalogue. Under Settings & privacy → Settings → Audience and Visibility → Posts, limiting who can see past posts converts years of public posts to friends-only in one action. For most people this is the highest-value step here.
• Stay out of the Meta AI chatbot. What you type at it is logged and feeds later models. Ordinary private messages to friends and family are not used for training.

Meta will not un-train on what it already holds, and says plainly it cannot even find your third-party data to remove it. Deleting old posts, comments and photos is the one step that does not depend on Meta agreeing to anything.