Meta withdraws AI-generated clickbait feed following accuracy concerns
The social media giant has ended a limited test of an automated news feed that produced AI-created articles and images of public figures, citing the feature’s lack of substance and sourcing.
Meta has removed a feature from its standalone Meta AI app that generated AI-created, clickbait-style articles and images. The "For You" section, which populated feeds with fabricated stories and images of public figures, was pulled after The Verge raised questions about the content's accuracy, lack of sourcing, and absence of AI labelling. Meta stated the feature was a limited test and has no plans to proceed.
The AI-generated content included fabricated stories, such as a tale about a royal butler settling a “milk first” debate (which appeared to trace back to a 2018 BBC Three comedy series) and a first-person narrative about a fake Rolex experiment. Images generated by the feature depicted real public figures, including multiple instances of Queen Elizabeth II, despite her death several years prior, alongside other royals like Prince William and King Charles.
The images contained significant errors, including impossible hands, unnatural body angles, and one image that was actually a GIF of an older couple. Internal prompts revealed in the chat history showed the AI was instructed to act as a “helpful conversational assistant” responding to proactive feed cards. Meta spokesperson Tracy Clayton issued three statements, with the second mysteriously removing the word “proactively” from the description of the test.
The feature targeted users with region-specific content, such as aggressively British topics for a London-based reporter and luxury watch content for another colleague. The standalone Meta AI app launched in April 2025 with a public “Discover” feed showing AI-generated images and conversations, which has since been removed.
The current app interface includes a standard chatbot and the “For You” page, which had been present for at least a few months prior to the withdrawal. Meta has previously stated its intention for users to know when posts are AI-generated and has implemented automatic labelling for some user-generated content. The Verge reported that the generated text read like “puffy filler” with little substance and no bylines.
