Meta AI
What the default settings do
Meta trains its generative AI models on public content from adult users across Facebook and Instagram: public posts, photos and their captions, and comments, alongside licensed data and public information from the open web. Private messages between you and friends or family are excluded from this unless you choose to share that content into a conversation with Meta AI yourself. Content from accounts belonging to people under 18 is also excluded.
Your conversations with Meta AI itself are treated differently from your messages to people. Since December 16, 2025, Meta uses your text and voice chats with Meta AI to personalize the content and ads you see across Facebook, Instagram, and its other apps, alongside signals like likes and follows. There is no setting to opt out of this other than not using Meta AI. Meta says it does not use conversations about religious views, sexual orientation, political views, health, racial or ethnic origin, philosophical beliefs, or trade union membership to show you ads. The change does not apply to people in the EU, the UK, or South Korea, where privacy law requires a different approach.
Whether you can object to AI training depends on where you live. People in the EU and UK have a legal right under data protection law to object to their own public content being used for AI training, and Meta offers an objection form for it. That right had real effect: Ireland’s Data Protection Commission pushed Meta to pause training on EU and UK users’ data in mid-2024, and Meta only resumed in 2025 after adjustments, including a more prominent objection process. Outside those regions, including in the United States, Meta does not offer a way to object to your own posts being used. The only form available there covers information about you that Meta obtained from third parties, such as the public web or licensed data, not your Facebook or Instagram content.
The standalone Meta AI app adds a social layer. It has a Discover feed where prompts you choose to share are public, and Meta can suggest those shared prompts on Facebook and Instagram. In 2025, reporters found many people had posted personal conversations without realizing they were public. Settings in the app let you keep your prompts visible only to you and stop them being suggested elsewhere.
In September 2026, Meta also began rolling out Muse, an AI agent in the US that can do tasks like sending email, booking travel, and filling in forms on your behalf. You connect each service to Muse yourself, one at a time. Meta says Muse does not share people’s conversations or data with its ads systems, which is a different rule from regular Meta AI chats.
The Meta AI assistant itself, whether inside Instagram, Facebook, Messenger, WhatsApp, or the standalone Meta AI app, cannot be fully removed. Meta’s help documentation covers muting its presence in a chat thread or minimizing where it appears, but there is no setting that deletes the assistant from the apps entirely the way you might uninstall a separate app.
A reasonable privacy baseline
Most of these settings are in the Privacy Center and Accounts Center, reachable from Settings & Privacy inside Facebook or Instagram. Meta AI app settings are under your profile icon in that app.
Meta AI chats and ads
Avoid sharing anything sensitive or personal with Meta AI → Practical habit rather than a setting
Outside the EU, UK, and South Korea, what you type or say to Meta AI can shape the ads and content you see, and there is no switch to stop that. Meta excludes some sensitive categories from ad use, but the simplest protection is to treat a Meta AI conversation more like a public post than a private message, and to use a different assistant for questions you'd rather keep out of your ad profile.
Accounts Center → Ad preferences → Review ad topics and the information used for ads
This won't stop Meta AI chats from being used, but it lets you see and hide ad topics and manage some of the other information used to show you ads, which can blunt the effect. It is managed separately from anything AI-specific.
Meta AI app
Profile icon → Data & Privacy → Manage your information → Make all your prompts visible to only you → On
This keeps your prompts out of public view. You should also be careful with the Share and Post buttons in a chat, which are what publish a conversation to the Discover feed.
Profile icon → Data & Privacy → Suggesting your prompts on other apps → Turn off for Facebook and Instagram
Stops Meta from suggesting prompts you've shared publicly on its other apps. If your prompts are already set to visible only to you, this is a second layer of protection.
Training objection (EU and UK)
Settings & Privacy → Privacy Center → Privacy topics → AI at Meta → Submit an objection if you're in the EU or UK
This is the path to Meta's form for objecting to your public content being used to train its generative AI models. In the EU and UK, the objection is backed by a legal right. Outside those regions this objection is not offered for your own content; the form you may find there concerns information from third-party sources only.
Account audience settings for posts → Set sensitive content to Friends or narrower
Meta's stated policy scope for AI training is public content. Content that isn't public to begin with is the most reliable way to keep specific posts, photos, or captions out of the training pool, and outside the EU and UK it is the only real lever you have.
Meta AI in other apps
Mute or hide the Meta AI chat thread → In Messenger, WhatsApp, or Instagram DMs where it appears
This is the practical limit of what's controllable: you can mute notifications from Meta AI and reduce where it surfaces in your chat list, but there is no toggle that removes it from the apps entirely.
Muse (US)
Connect only the services you need → Review connections in Muse before granting each one
Muse asks you to connect apps individually, so you can keep its reach narrow. Meta says Muse data is kept out of its ads systems, but an agent that can send email or make purchases still sees a lot, so start with the one or two services you actually want it to handle.
Trade-offs
- Not using Meta AI is the only way to keep your AI conversations out of ad personalization outside the EU, UK, and South Korea, which means giving up a tool that is built into apps you may use every day.
- Keeping your Meta AI app prompts private means you can’t share a useful conversation to the Discover feed without changing it back.
- Setting posts and photos to Friends-only or narrower reduces your presence in Meta’s training data, but it also limits the reach and visibility you may otherwise want for public content, such as a public profile used for professional purposes.
- Muting Meta AI reduces how much it intrudes on your daily use of the apps, but you lose quick access to it in the threads where it’s muted.
- Treating Meta AI conversations as semi-public means being more guarded in exactly the kind of casual, exploratory conversations the assistant is designed to invite.
- Adjusting ad preferences may reduce how relevant ads feel to you without reducing how many ads you see overall.
Who should not change these settings
- If you live outside the EU and UK, there is no training objection to submit for your own posts, so skipping that step is simply accurate, not an oversight.
- If you rarely use Meta AI, or only ask it everyday questions you’d be fine seeing reflected in your ads, the AI-chat ad change affects you little and there is nothing to switch.
- If your account and posts are already public by design, such as a creator, business, or public figure account, keeping content public is likely central to why you use the platform, and narrowing audience settings would undercut that purpose.
- If you find Meta AI genuinely useful inside Messenger or WhatsApp threads, such as for quick questions without leaving the conversation, keeping it unmuted is a reasonable trade for that convenience.
- If you’re comfortable with your public content contributing to Meta’s AI development and don’t view it differently from the visibility you’ve already accepted by posting publicly, leaving the defaults as they are is a coherent position.