Meta adds LLM-powered feed controls to Facebook

Meta is expanding its AI-powered "Your Algorithm" personalization feature to Facebook and Facebook Reels, incorporating large language models much more heavily into its core content ranking engines. The update allows users to directly configure and manage the specific topics they wish to see across their main feeds, giving individuals tighter, immediate control over their recommendation experience.

The new rollout follows similar implementations by Meta earlier this year, including the initial launch of "Your Algorithm" on Instagram and the comparable "Dear Algo" feature on Threads. On Facebook, users can activate the new controls by selecting "interested" or "not interested" choices found directly beneath individual posts or inside the three-dots action menu. In certain instances, the platform will display follow-up queries that ask users to describe their precise preferences in their own words.

Underneath the user interface, Meta is leveraging LLMs to parse the precise context, intent, and nuance behind these natural language inputs. By moving beyond traditional rigid category tags, the language models allow the feed algorithm to process complex directions and update recommendations in real time. Meta indicates that content adjustments take effect immediately, adapting within the exact same browsing session to reflect whatever matters to a user at that moment.

For machine learning practitioners and system architects, Meta's expansion demonstrates how modern content discovery architectures are shifting toward interactive, model-driven feedback loops. Rather than inferring user intent purely through passive behavioral signals such as watch time, clicks, or scroll speed, recommendation pipelines are integrating generative AI capable of understanding explicit instructions. This strategy provides consumers with clear steering controls while offering systems rich, high-intent feedback to refine algorithmic ranking.