Meta's AI App Was Generating Fake News. Literally.

Meta spent years telling Congress, regulators, and the public it was cracking down on misinformation and low-quality clickbait on Facebook. Then it built its own AI-powered clickbait machine and stuffed it inside its app.

According to The Verge's Robert Hart, Meta's standalone AI app — the one launched in April 2025 — had a 'For You' section that generated entire fake 'stories' on demand. Tap a prompt, get a full article. No reporters. No sources. No accountability.

The feature had apparently been running for at least a few months before anyone in the press caught it.

What These 'Stories' Actually Looked Like

Hart, a London-based reporter, tested it himself. His feed served up prompts like 'A royal butler finally settled the milk first debate,' 'The psychology of joining a queue without knowing why,' and 'Inside the extreme sport of visiting every UK pub.'

The algorithm had clearly profiled him as British and went full stereotype.

His colleague got a different treatment — the system apparently pegged him as a luxury watch guy and served up pieces like 'My fake Rolex experiment' and 'The brutal math behind the Rolex waitlist illusion.'

Hart described the AI-generated text as 'puffy filler' that offered little beyond restating the premise of the prompt. Sourcing? ZERO. None. The stories cited nobody and nothing.

Hart tried to trace one story — the royal butler tea piece — and found it appeared to originate from a 2018 BBC Three comedy series about a fictional character. The AI apparently laundered a joke from a sitcom into a 'news story' presented as fact.

Meta Pulled It the Moment Questions Were Asked

Meta said it would remove the feature after The Verge contacted them for comment.

Not before. Not proactively. After a reporter showed up.

A company that caught a mistake internally fixes it without prompting. A company that gets caught removes it when questioned.

Meta has not publicly explained how long the feature ran, how many users saw it, what guardrails — if any — were in place, or whether the AI-generated content was labeled as AI-generated in any meaningful way. As of June 6, 2026, no Meta spokesperson has issued a statement with those specifics.

This Is Bigger Than One Bad Feature

For years, Facebook's algorithm supercharged clickbait publishers — the kind of low-effort, high-sensation garbage that crowded out real journalism. Meta repeatedly promised to fix this. They changed the algorithm. They added 'news feed quality' rankings. They eventually killed their News Tab entirely in the United States and Australia.

Now they've cut out the middleman. Why pay a clickbait farm in Macedonia when you can just generate the slop yourself?

The business logic is straightforward. AI-generated content is essentially free to produce. It fills screen time. It keeps users engaged. The app needs reasons for people to open it beyond just asking the chatbot questions. A personalized content feed solves that — and AI makes it infinitely scalable at near-zero cost.

The problem is that the content is fabricated. Not exaggerated. Not spun. Made up from scratch, with no factual basis, no sources, and no accountability.

What Mainstream Coverage Is Missing

The Verge did the legwork on catching this. But the broader media coverage of Meta's AI products consistently treats these missteps as isolated technical glitches rather than a pattern.

Meta's entire AI product strategy involves generating and surfacing synthetic content at scale. The 'Discover' feed in the original Meta AI app — which showed AI-generated images and conversations from users who often didn't realize they were public — followed the same pattern. Feature launches. Users engage without full understanding. Press asks questions. Feature gets modified or pulled.

Rinse, repeat.

Meanwhile, Meta CEO Mark Zuckerberg has been on a year-long media tour presenting himself as a free-speech defender and standing against 'censorship.' A platform that won't tolerate certain kinds of human speech but will happily auto-generate fake articles without disclosure presents a stark contradiction — one that has largely escaped notice in coverage of Meta's AI push.

Labeling Questions

Was any of this AI-generated content labeled as AI-generated at the point of consumption?

If a user tapped a prompt and read a 'story,' did they see a clear disclosure that a language model fabricated every word? Or did it look like an article?

Based on Hart's reporting, it looked like an article. That matters enormously — both legally, given ongoing EU AI Act enforcement, and ethically, given Meta's stated commitments to transparency.

The Aftermath

If you use Meta's AI app, you were potentially reading fabricated 'news stories' for months without knowing it. Meta didn't tell you. They didn't label it clearly enough that a tech reporter noticed the labeling. They killed it when a journalist showed up.

The feature is gone now. Until the next one.