Most people did not choose Meta AI. It arrived in an app update, and one day there was a new chat and a new button inside WhatsApp, Instagram or Facebook.
That makes it different from other AI tools. Nobody downloads ChatGPT by accident. Meta AI, on the other hand, is something a billion people found waiting for them.
Because it was never really chosen, the questions people ask about it are different too. Less "what can it do" and more "can I get rid of it" and "is it reading my messages".
This guide answers all of it in plain language: what Meta AI actually is, what it helps with, whether it costs anything, whether it can be turned off, and what happens to the things typed into it. No jargon, no hype, just the honest picture.
What Meta AI actually is
Meta AI is the AI assistant built by Meta, the company behind Facebook, Instagram and WhatsApp. It appears inside WhatsApp, Instagram, Facebook, Messenger and Threads, in places like search bars and group chats, and it also exists as a standalone app and the meta.ai website.
Like other chatbots, it is built on a large language model. The standalone app launched on 29 April 2025 built with Llama 4, Meta's own model family, according to Meta's announcement.
In April 2026 Meta announced a newer flagship model called Muse Spark, the first from its Meta Superintelligence Labs division. Meta says it powers the Meta AI app and meta.ai first, with a rollout to WhatsApp, Instagram, Facebook and Messenger following.
The scale is enormous. Mark Zuckerberg announced 1 billion monthly active users across Meta's apps at the company's May 2025 shareholder meeting, per CNBC. Most of that billion did not seek out a chatbot. The chatbot came bundled with their messaging app.
What Meta AI can help with
Meta AI handles the same everyday tasks as other chatbots: answering questions, explaining things, drafting messages, and generating ideas.
Based on Meta's own announcements, it can help with:
- everyday questions and chat inside apps people already use
- generating and editing images from a description
- answering questions in a group chat when someone mentions @Meta AI
- web search on the meta.ai website
Its genuine advantage is convenience. There is no new app to download, no account to create, and no cost. Someone halfway through a WhatsApp group chat about dinner plans can ask it a question without leaving the conversation.
That convenience is the whole pitch. It is also, as covered below, the source of most complaints about it.
Is Meta AI free?
Yes. The base assistant is free across all the places it appears.
Meta has started testing paid plans, but only in three countries. Meta One Plus at $7.99 per month and Meta One Premium at $19.99 per month entered testing in Singapore, Guatemala and Bolivia from late May 2026, according to TechCrunch, which also confirms that base Meta AI remains free.
Those paid tiers are not available in the UK as of July 2026. For a UK reader, Meta AI is simply free, and there is no upgrade decision to make yet.
Free is worth pausing on, though. Meta's business model is advertising, and as covered in the privacy section, conversations with Meta AI now feed into that in most regions. The product costs nothing because the conversations have value.
Can Meta AI be turned off?
This is the question people search for most, and the honest answer is no. There is no setting that fully removes Meta AI from WhatsApp, Instagram, Facebook or Messenger.
The ExpressVPN guide to this puts it plainly: Meta does not provide a single setting to remove Meta AI across its apps, and some AI features stay built into the interface even when the assistant is muted or avoided. WhatsApp's own privacy page frames it as optional to use, not optional to see: "Using Meta AI is completely optional on WhatsApp. You have to take action to start the conversation."
What can be done instead:
- mute the Meta AI chat and its notifications
- delete or archive the Meta AI chat thread, though it can reappear when tapped again
- on Facebook, disable "Ask Meta AI" in search and turn off comment summaries per post
- on WhatsApp, turn on Advanced Chat Privacy for a specific chat, which disables some Meta AI features in that chat
- simply never tap the button and never message it
That last option matters more than it sounds. Meta AI only receives what is deliberately sent to it, per WhatsApp. Ignoring the button entirely is a genuine form of opting out, even if the button itself never goes away.
Does Meta AI read private messages?
This is the biggest fear, and Meta's official answer is no. WhatsApp states: "your personal messages are end-to-end encrypted and are not seen by Meta AI" (whatsapp.com/privacyquestions). Meta AI cannot start a conversation by itself. Only a person in the chat can bring it in.
However, there are caveats, and they matter.
Anything sent to Meta AI is a different story. When someone mentions @Meta AI in a chat, that part of the conversation is not end-to-end encrypted, as the Electronic Frontier Foundation explains. Meta receives whatever is shared with the assistant. In a group chat, any participant can invoke it, so one person's choice can expose what gets shared with it.
There is also the question of training. In the EU, Meta trains its models on adults' public posts and comments plus interactions with Meta AI, with private messages and under-18 data excluded, per Meta's announcement. EU users get a formal objection form, found in Meta's Privacy Centre, and Meta says it will honour all objections. The objection stops training use. It does not remove Meta AI from the apps.
Then there is advertising. From 16 December 2025, Meta uses Meta AI conversations to personalise content and ads across its apps, with no blanket opt-out, per Meta's announcement. Meta says sensitive topics such as health, religious views and political views are excluded from ads. Meta's post says the change applies in "most regions"; TechCrunch reported that the EU, UK and South Korea are excluded.
One incident is worth knowing about. In June 2025, the standalone app's Discover feed publicly exposed chats that users did not realise they were sharing, including personal and medical matters, as reported by CNBC and Malwarebytes. Nobody was hacked. People pressed "share" without understanding it meant publishing to the world. Meta later added a warning screen before sharing.
For the privacy-conscious, WhatsApp announced Incognito Chat with Meta AI on 13 May 2026, promising that "no one can read your conversation, not even us", with messages excluded from training and personalisation. Worth noting: Malwarebytes points out that this is Meta's own promise on Meta's own infrastructure, not the same technical guarantee as end-to-end encryption between two people.
Limitations and things to watch for
The first limitation is the one covered above: it cannot be removed, only muted, deleted per chat and avoided.
The second is that, outside the EU, UK and South Korea, conversations with it shape the ads a person sees, with no full opt-out. What gets typed into Meta AI is a signal, not just a question.
Third, like every chatbot, Meta AI can be confidently wrong. It generates plausible text based on patterns, which means it can state wrong facts while sounding certain. Answers should be treated as a starting point rather than a finished answer, and anything important should be checked separately. Meta markets Muse Spark's health answers as "developed with physician input", but that is Meta's claim, not a reason to skip a doctor.
Finally, independent comparisons such as Zapier's consistently find that dedicated chatbots like ChatGPT are stronger for complex work: professional writing, coding, and multi-step tasks. Meta AI's strength is being free, instant and already there, not being the most capable tool available.
How Meta AI compares to other AI tools
The practical difference comes down to convenience versus control.
Meta AI wins on convenience. It costs nothing, needs no setup, and lives inside apps a billion people already open every day.
A dedicated chatbot gives more control. Tools like ChatGPT offer clearer privacy settings, paid tiers for heavier use, and stronger performance on complex tasks. They are also easy to simply not have, which Meta AI is not.
In most cases, the sensible split is using Meta AI for quick, low-stakes questions in the moment, and a dedicated chatbot for anything involving real work or information worth protecting.
Final thoughts
Meta AI is unusual among AI tools because the relationship starts backwards. Instead of someone choosing it, it chose them, arriving inside apps they were already using.
Judged as a free assistant, it is genuinely useful for quick questions, image generation, and everyday help without switching apps. Judged as a choice, it is one nobody got to make, and the honest answer to "can it be turned off" is no, only quietened.
The safest approach is to treat it deliberately. Use it for casual questions if it is useful, keep anything private or sensitive out of it, and remember that whatever is sent to it is received by Meta. It works best as a helper, not the final authority, and the things worth keeping out of any AI chat are doubly worth keeping out of this one.
