- Trains on your data by default
- Yes. Meta says it uses people's interactions with its AIs, including content and messages exchanged with AI features, to improve AI at Meta across the Meta AI and Vibes apps and websites, and the pages checked describe no tier that is excluded.
- How to opt out
- not documented
- Data retention
- Meta keeps information as long as it needs it to provide its products, meet legal obligations or protect interests, decided case by case with no fixed period; users can delete all chats and media in the Meta AI app, but public vibes are not deleted that way.
- Admin controls
- not documented
- Certifications
- not documented
- Data region
- Meta transfers, stores and processes information globally, including to and from the United States, and no choice of region is described.
- Free versus paid
- not documented
What it does with your data
Meta uses what people type into Meta AI to improve its AI. The vendor states this directly for the Meta AI and Vibes apps and websites.
Meta’s Privacy Policy lists the information it can use to develop and improve AI at Meta. That list includes content and messages exchanged with AI features, plus related metadata. Meta says this work serves both Meta Products and third parties. So treat anything a colleague types, pastes or uploads as material Meta may use this way.
Your text can also travel a second way. When Meta AI cannot answer a question, Meta shares the messages sent to it with select partners. It also passes on general information, such as the user’s region.
Meta keeps information for as long as it needs it to run its products, meet legal obligations or protect its interests. It decides this case by case and gives no fixed period. Voice recordings stay in history only for a while, but the text transcript stays. The pages checked do not split any of this by tier.
Settings to change first
The vendor pages checked do not describe any setting that stops Meta from using your chats for AI training. The only control they document is deleting history. Here is what you can do:
- Delete chats regularly. In the Meta AI app, go to Menu > Settings > under App settings, Data & privacy > Manage your information > Delete all chats and media > Delete all. After that, chats, prompts and media no longer appear in History or Media.
- Watch for public vibes. The delete step above does not remove public vibes. Ask staff not to make any public.
- Avoid posting voice prompts. A voice recording posted with a prompt stays available until someone deletes the post.
The page does not say whether deleting history removes content that Meta has already used.
Free and paid tiers
The vendor pages checked do not cover free versus paid tiers for Meta AI. They do not describe a paid plan, and they do not say that any plan handles data differently.
Other tools often offer a business tier that is excluded from training. Do not assume Meta AI has one. Nothing in the pages checked says so. Everything in this profile applies to Meta AI as the vendor describes it today: the training use, the partner sharing and the case-by-case retention.
If you hear about a business or paid version, ask Meta for its terms in writing before you rely on them. Get the answer to 3 points: whether it trains on your content, how long it keeps chats, and who at Meta can see them.
Admin controls
The vendor pages checked do not cover admin controls. They describe no single sign-on, no central user management and no organization-wide data settings. In practice, each colleague would be using a personal account that you cannot manage or audit.
They also do not cover certifications or audit reports. We found no named security certification or audit report for Meta AI in the material reviewed.
On location, Meta transfers, stores and processes information globally, including to and from the United States. This covers its own offices and data centers as well as partners and service providers. Information from AI interactions is one of the categories listed. The pages checked describe no way to choose a data region.
Using it with donor and client data
Do not use Meta AI with donor or client data. The vendor pages checked offer no tier or setting that keeps your content out of AI training.
Never paste in:
- Donor names with gift amounts, giving history or contact details
- Client case notes or anything that identifies a person you serve
- Health, payment or bank details
- Grant applications or board papers that are not public yet
If colleagues want to use it, limit them to tasks with no private data. Good examples are rewording a public event notice or brainstorming headline ideas for a newsletter that has already gone out.
Before you allow wider use, get these points in writing from Meta, because the pages checked do not cover them:
- How to opt out of AI training
- Whether a paid or business plan exists, and how its data handling differs
- What admin controls an organization gets
- Which certifications or audit reports apply
Questions people ask
Does Meta AI train on what my staff type?
Yes. Meta says it uses people’s interactions with its AIs to improve AI at Meta. The pages checked describe no plan or setting that changes this.
Can we delete what we have entered?
Partly. The Meta AI app has a “Delete all chats and media” option that removes chats, prompts and media from History and Media. Public vibes are not deleted by this. The page also does not say whether content already used for training is removed.
Does anyone outside Meta see our chats?
It can happen. When Meta AI cannot answer a query, Meta shares the messages with select partners, along with general information such as the user’s region. Meta also stores and processes information globally through its partners and service providers.
Verified on Published Updated
Sources
- How Meta uses information for generative AI models and features, checked October 27, 2026
- Remove posts, chats and media from Meta AI and Vibes, checked October 27, 2026
- Privacy Policy, checked October 27, 2026
- Information used for AI at Meta, checked October 27, 2026