Meta has changed the prompts its AI chatbot suggests to users after a viral video showed the assistant asking a woman to identify her young daughters and surfacing photos she says she had deleted. The company told The Verge it «missed the mark» and that the feature «never should have prompted the individual with questions like that.» The episode matters because Meta AI sits inside Facebook, Instagram, WhatsApp and Messenger, where suggested prompts shape what hundreds of millions of people see next to their own posts.
What the video showed
Instagram user Kalie Robins posted a video describing what happened after she cross-posted a clip to Facebook. Beneath footage of her and her child singing in a car, Meta AI displayed the prompt «Who is the child passenger?» When she selected it, the assistant, according to Robins, combined details about her daughters from her earlier posts and from posts published by her relatives. It then offered further prompts about her children's ages and where she lives with her family, and showed photos of her daughters on the platform, including one she claims she deleted years ago.
Meta spokesperson Dina El-Kassaby said the company has fixed the issue that caused Meta AI to suggest prompts on personal topics. She described the feature as a way to help people get more information about posts that interest them, and said the generated response draws on information the user already has access to. In this case, Meta AI may have pulled from videos in which Robins named her children, according to El-Kassaby. The correction follows a similar retreat in July, when Meta removed a feature that let people create AI deepfakes of other Instagram users after a backlash.
What this means for businesses
For companies that publish on Meta's platforms, the incident shows how far an assistant can reach into content that was posted for a different purpose. A brand video with a child, a customer's face or a home address can become raw material for a suggested question, and the answer is assembled from posts the account and its contacts have already made. Marketing and support teams that treat social profiles as a low-risk channel now have to assume that anything visible there can be recombined by the platform's own AI.
Two limits are worth keeping in mind. Meta says the response is based on information the user already has access to, so the change addresses which prompts appear rather than what data the assistant can retrieve. The company has not said how the fix works, whether it applies to all languages and regions, or how quickly it reaches every surface where Meta AI runs. A single corrected prompt does not by itself prove that similar suggestions will not appear around other personal details, such as location or family relationships.
The signal to watch is whether Meta publishes the rule behind the fix and reports further prompt-related incidents in the coming months. If no comparable cases surface after the change, the correction can be treated as effective; a repeat would suggest the problem lies in how the assistant ranks personal content, not in one prompt.
