Which social networks train AI on your data?
Of the 10 social networks on file, 9 are graded: 8 train on your data by default and 1 doesn't. One more has a policy that doesn't say either way. The best graded is Bluesky (B) and the worst is X (F).
Do they train on your posts, your Pins and your public stories?
Every story about AI and social media is about scraping: who took the posts, and whether anyone was asked. The apps here are graded on a narrower question: does the company itself train on what you post, and do its own documents give you a way to stop it. What recurs, read together, is where the answer sits. Often not in the privacy policy at all. In a support article, a company post or a separate AI page.
The ones that say it, and name no way out for most
Pinterest says it uses information “to train, develop and improve our technology such as our machine learning models, regardless of when Pins were posted”: a permission reaching back to Pins uploaded before the policy existed. A right to object where generative-AI features rest on legitimate interests is stated, but it opens “If you live in the EEA, the UK, or Switzerland”. For everyone else the document names no way out, which is what an F is.
X says it may use “the information we collect and publicly available information to help train our machine learning or artificial intelligence models”. The policy does carry an opt-out that mentions AI training, and it points outward: it governs what third-party recipients of your data may do, including using it “to train their artificial intelligence models”. No control over X’s own training is named anywhere.
The ones with a way out you must go looking for
Meta’s UK-served policy lists among its legitimate interests the purpose “to develop and improve artificial intelligence technology while ensuring appropriate safeguards”, with public content you or others share as its input. The policy names no setting, no form and no objection route. The route is on Meta’s separate generative-AI page: an objection request through your account. A form to go looking for is a D. The US-served text is its own tier, unresolved.
Instagram has no privacy policy of its own: its privacy centre serves the Meta Privacy Policy under Instagram’s name, and the entry follows Meta’s. Rendered from the UK, the text carries the same AI purpose and names no way out of it. Meta’s generative-AI page counts Instagram accounts among those that can object, and that form is the route.
Threads has a supplemental policy of its own. It covers what Threads collects and defers the rest to the Meta Privacy Policy; it has no training language and no control. So the grade is Meta’s: the UK-served clause, and the objection form on Meta’s generative-AI page. A form to find, not a setting, is a D, as the Meta entry is.
LinkedIn may use your personal data to “develop and train artificial intelligence (AI) models”. No setting attaches to training. The one control that could reach it is the objection right, “where we rely on legitimate interests, you have the right to object”, and the policy never says which lawful basis training rests on. A route the document doesn’t settle is still a route.
TikTok’s EEA policy, the only version this site can read, puts training among its legitimate-interest purposes. One clause runs from “monitoring interactions and usage across your devices” to “training, testing, and improving our technology, such as our machine learning models and algorithms”. It names the legal basis for each purpose and grants a right to object over legitimate-interest processing. That’s a route, a remedy to exercise and not a switch. Nothing in the product turns training off.
Snap answers in a support article, not in its privacy policy. Content “posted publicly on Snapchat (such as posts to Spotlight, Public Stories, and the Snap Map)” may train its generative models, with “a combination of automated and human review”. A switch is named and located, under Privacy Control, but the article never says which way it ships, and opting out “does not affect training that has already taken place”. Private snaps and chats are a second tier, unresolved.
The one that says no
Bluesky’s privacy policy has nothing on training. The answer is a post by the company’s official account: “We do not use any of your content to train generative AI, and have no intention of doing so.” Off by default is what a B is. That the answer is a post and not the policy is why the confidence is medium. Two limits the company states: the commitment is about generative AI, and it can’t bind anyone else scraping an open public network.
The one that never reaches the subject
Reddit’s policy names ChatGPT and says third parties pay licensing fees for bulk content. The first says your public posts may be “available in search results on internet search engines like Google or in responses provided by an AI chatbot”. That’s where your words can end up, not what Reddit does with them. The second describes a commercial arrangement without saying what the buyers may do. Licensing isn’t training, and the document never uses the word train.
Side by side
3 pairs of social networks people choose between, with both verdicts on one page.