Which media apps train AI on your data?
Of the 6 media apps on file, 3 are graded, and every one of them trains on your data by default. The other 3 have a policy that doesn't say either way. All 3 graded entries are F, at the same exposure score.
Do they train on what you read, watch and listen to?
Every story about AI and the media is about what the models took from the publishers. The other direction is what these verdicts read for. What a newspaper, a streaming service, an audiobook store or a publishing platform does with the record it holds of you: what you finish, what you abandon, what you write and to whom. Read together, the documents divide by how far they get. Some say it outright, with nothing to switch off. One raises the subject and sends the reader to another page. The rest never mention AI at all.
The ones that say it outright
The New York Times says the direct messages it collects in some of its apps “may be used to train AI models” for the moderation it describes. It says it again of recorded support conversations: “Your chat session will be recorded for quality assurance and training and may be used for training AI models.” The note on the verdict records who is saying it: a newspaper in litigation with OpenAI over models trained on its journalism, reserving the same use of its readers. No opt-out is named. The Times was written to. Nothing has come back.
Twitch’s is a single sentence near the end of a long paragraph about operating and improving the service: “This includes using data to develop or deploy our generative AI models and services.” What Twitch holds is chat, streams and viewing behaviour, much of it from young audiences. The document names no way to decline and no plan on which the sentence doesn’t apply, which is what an F is on this site.
The training in Medium’s clause is done by its vendors, not by Medium: they may use what you write “to train, test, and improve technology, including machine learning models and algorithms”. That the training is done by third parties doesn’t make it narrower. The same passage says Medium and those vendors “may scan, analyze, and review your content, messages, AI interactions, and associated metadata”, and a publishing platform holds drafts as well as posts, and private notes between writers and editors. No opt-out is named.
The one that raises it and points elsewhere
The Guardian is unusually direct for a newspaper, and then stops. Its policy says it “may use generative AI functionalities in our software and systems, including in our use of personal data”. Then it sends the reader to a separate statement of generative AI principles instead of settling the question here. No form of the word train appears in the document. A reader learns that AI touches their data, and not whether their data builds the AI.
The ones that never mention AI at all
Audible knows every book you have listened to and where you stopped, and its notice is precise about collecting that history: what you finish, what you abandon, what you replay. It never mentions training, machine learning, artificial intelligence or AI, in any form. Whether any of it builds a model isn’t a question the document takes up. Silence is a finding about the document, not an accusation.
Netflix has described recommendation as central to its business for years. The recommendation engine is the product. Its privacy statement is long and specific about what it collects, including what you watch and when you pause, and never mentions training, machine learning, artificial intelligence or AI, in any form. Its help centre comes closer: viewing signals “continually update our algorithms”, alongside what members with similar tastes watch. A system that learns from viewing, described without the word train. Netflix is being asked.
Side by side
One pair of media apps people choose between, with both verdicts on one page.