Meta will deprecate a For You feed inside its standalone Meta AI app that generated entire clickbait-style articles on demand, with images and text fabricated by the model. The company confirmed the shutdown after questions from a reporter who found the feed produced fake first-person essays, invented expert sourcing, and AI images of real public figures, none of them carrying an AI label. The feature had been live for at least several months across multiple users, despite Meta's framing of it as a limited test.
The Meta AI app launched in April 2025 with a public Discover feed of user prompts and images, an experiment that drew criticism after users appeared not to realize their conversations were public. That feed is gone. In its place, the app paired a standard chatbot interface with a For You page that surfaced cards of suggested article prompts; tapping one generated a full story tuned to the user's apparent interests, served up like a news item.
Targeting profiles were crude and stereotype-driven. A London-based reporter saw cards built around tea, queuing, pubs, soccer, and the royal family, with prompts including "A royal butler finally settled the milk first debate" and "Inside the extreme sport of visiting every UK pub." A colleague was sorted into a luxury-watch bracket and served cards like "My fake Rolex experiment" and "The brutal math behind the Rolex waitlist illusion." The same card tapped twice produced variant stories on the same theme; the same headline typed into a fresh chat produced something different entirely.
Key facts
- 01Meta's standalone AI app added a For You feed that generated entire clickbait-style articles on demand, with text and images created by AI.
- 02One generated image of the British royal family featured two Queen Elizabeth IIs, despite her death several years ago.
- 03Articles carried no visible AI label, even though Meta has publicly said it wants users to know when posts are made with AI.
- 04A 'Rolex experiment' first-person story appeared to be fully fabricated, generated in-app without a byline or sourcing.
- 05Meta said the feature was a limited test and will be deprecated with no plans to move forward.
The generated text leaned on vague references to unnamed experts or fictional research and offered little beyond restating the prompt. The royal butler tea claim appears to trace back to a 2018 BBC Three comedy series, Miss Holland, featuring former royal butler Grant Harrold, but the Meta AI article cited no source. The Rolex piece read as a first-person narrative with no byline and appeared to be wholly invented.
“We're testing a daily feed that proactively shares tips, content, and recommendations tailored to your interests.”— Tracy Clayton, Meta spokesperson
Image output had similar problems. An article titled "Who really pays for the royal family in 2026?" was illustrated with a group portrait featuring two Queen Elizabeth IIs — despite her death several years earlier — alongside distorted approximations of Princess Kate, Prince William, and King Charles. Other images carried the usual artifacts: impossible hands, unnatural posture, and at least one animated GIF of a couple making movements no human body could make.
Meta has previously said it wants "people to know when they see posts that have been made with AI" and that it automatically labels some user-generated content when AI is detected. The For You feed carried no such label on either the generated articles or their images. Meta's own rules on generating images of real people are opaque, and the app produced recognizable depictions of the royal family without apparent restriction.
Spokesperson Tracy Clayton initially described the feature as a daily feed that shares tips and recommendations, with the goal of suggesting what is most relevant to a user "before you even have to ask." A second, near-identical statement quietly dropped the word "proactively." A third statement followed later in the day, confirming the feature will be pulled.
“This was a test for a limited number of users and it will be deprecated. Meta has no plans to move forward with this feature.”— Tracy Clayton, Meta spokesperson
The reversal raises more questions than it answers. Meta declined to say whether it considers the output news or fiction, what safeguards were in place, or whether AI-generated images of public figures complied with its own content policies. The company also did not explain how a "limited" test reached at least four reporters at the same outlet, or why a feed that produced fabricated articles and unlabeled synthetic images of the royal family shipped to users in the first place.
The episode lands at an awkward moment for Meta's AI strategy. The company is pouring capital into model training and consumer AI surfaces while simultaneously trying to thread a regulatory needle on synthetic media in the EU and UK, where AI labeling and likeness rules are tightening. Shipping a feature that fabricates news-style articles about real public figures, without labels, is the kind of product decision regulators cite when arguing platforms cannot self-govern generative output. It also follows a separate incident covered here in which Meta's AI support agent handed over Instagram accounts to attackers who simply asked, suggesting the company's AI product reviews are not catching obvious failure modes before launch.
The bigger problem for Meta is not this one feed but the pattern it reveals: a consumer AI app that defaults to generating plausible-looking content with no provenance, no sourcing, and no labels, then gets walked back only when a reporter calls. Generative feeds are an obvious product direction — TikTok-style engagement loops powered by infinite synthetic content — and Meta will try again in some other form. Whether it can ship one that distinguishes "fitness tips" from a fake article about the monarchy, with the right labels attached, is the actual test, and this attempt failed it.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
