Facebook's News Feed team on algorithms, relevancy scores, and the importance of giving users more control over their feeds
How Facebook's news feed algorithm works — A small team of engineers in Menlo Park. A panel of anonymous power users around the world. And, increasingly, you.
Context & Ripple Effects
This 2016 profile catches the News Feed at an inflection point: a small Menlo Park engineering team running relevancy scores, increasingly supplemented by asking users directly rather than only inferring from behavior. The groundwork was already laid a year earlier, when Facebook was supplementing its algorithms with direct questions to improve relevance, starting with just 600 people.
The arc since then runs in two directions. In 2018 the feed's deprioritization of passive content forced a reckoning among publishers, who were told that journalism engineered to be picked by an algorithm is marketing, not journalism. Yet by late 2021, internal documents showed Facebook still defending algorithmic ranking on the grounds that its data suggests the algorithm knows what users want better than they do themselves — the same tension between stated user control and algorithmic confidence this interview surfaces.
First-order effects
- Publishers and media companies dependent on News Feed distribution now have a clearer picture of how relevancy scores are built — including that a panel of anonymous power users helps tune them — making the ranking logic less of a black box but no more negotiable.
Second-order effects
- As Facebook scales feedback collection into global News Feed ranking tests soliciting user judgments on post value, publishers face a moving target where editorial choices are increasingly optimized against reported user sentiment rather than raw engagement signals.
Third-order effects
- If the pattern holds, the structural question shifts from 'how does the algorithm work' to 'who decides relevance': platforms will keep expanding explicit user-input mechanisms while their own research continues to argue the algorithm outperforms self-reported preferences, leaving regulators and publishers to press on transparency.
The trend: Social feed ranking is moving from purely behavioral inference toward hybrid systems that blend algorithmic scores with explicit user feedback, while platform-internal conviction that the algorithm knows best keeps true user control contested.