Facebook supplementing algorithms with direct questions to improve relevance of News Feed, currently testing with 600 people
Revealed: Facebook's Project to Find Out What People Really Want in Their News Feed — Every weekday beginning at around nine, 30 contracted white-collar laborers arrive …
Context & Ripple Effects
This 2015 trial — contracted staff asking a 600-person panel directly what they want in their feed — reads today as the origin point of a habit Facebook never dropped. A year later the News Feed team was talking openly about relevancy scores and giving users more control, and by 2019 the experiment had graduated into production: an algorithm update folding survey data on worthwhile links and close friends into ranking itself.
What makes the small test matter is where it led alongside the survey track: Facebook repeatedly paired machine ranking with human judgment, from the Today In local news tests blending curation and AI to hiring journalists for the News tab. The 2021 global ranking tests soliciting post-level user feedback are essentially this 600-person panel at platform scale.
First-order effects
- Facebook's News Feed team gets a preference signal engagement metrics can't provide — why people value or resent specific posts — from 600 users whose answers feed ranking experiments.
- Publishers and pages whose content scores well on clicks but poorly when asked about directly now face a ranking input they don't control and can't optimize for the way they optimized for engagement.
Second-order effects
- Rivals building feeds on pure behavioral inference face pressure to add their own explicit-feedback loops, since Facebook's hybrid signal can justify relevance decisions engagement data alone cannot.
- Once surveys become a ranking input, the incentive shifts toward content that survives scrutiny when named valuable — favoring the same mix of human-curated and algorithmic surfaces Facebook tested in Today In and later the News tab.
Third-order effects
- If the pattern holds, feed ranking structurally moves from inferring preferences purely from behavior to combining declared preferences with observed ones — making user research infrastructure a competitive moat rather than a design nicety.
- The recurring pairing of algorithms with human judgment (survey panels, contracted curators, hired journalists) points toward hybrid curation becoming standard architecture for social feeds, with platforms accountable for explaining not just what ranked but why.
The trend: Social feed ranking is evolving from purely behavioral algorithms toward hybrids that fold declared user preferences and human curation into what gets shown.