Docs: Facebook tested turning off News Feed ranking for 0.05% of users in Feb. 2018 and found engagement dropped, Groups content rose, and ad views increased
New leaked Facebook documents show what happened when Facebook tried to remove feed ranking. It didn't go well.
Context & Ripple Effects
This leak completes an arc that related coverage has been building all year: after Facebook ran global News Feed ranking tests soliciting user feedback on what posts are valuable in April, and Protocol documented its practice of withholding protections from a subset of users in "holdout" experiments, we now have the results of the most extreme version — ranking switched off entirely for 0.05% of users in February 2018.
The finding lands differently because Facebook has been here before structurally: its six-country test moving non-promoted Page posts to a secondary feed cut media engagement badly enough that Facebook killed it months later. The new documents show why the company keeps choosing ranked feeds — unranking hurt engagement while ad views still rose, meaning the algorithm is load-bearing for attention quality, not just distribution.
First-order effects
- Users in the holdout saw a degraded experience — lower engagement overall — while Groups content surfaced more, revealing which content types depend least on ranking to find an audience.
- Advertisers technically got more ad views without ranking, giving Facebook internal evidence that ranking's value is concentrated in keeping users engaged rather than in raw ad inventory.
Second-order effects
- Publishers and Page owners face renewed confirmation from the Explore Feed episode and this test that any de-ranking move shifts reach away from media and toward community formats like Groups — shaping where they invest audience-building effort.
- Internally, the result arms defenders of the ranking system within Facebook against critics pushing for chronological or less-optimized feeds, since the company's own holdout data shows what removal costs.
Third-order effects
- As Meta's later admission of a "massive ranking failure" that boosted harmful content views up to 30% shows, ranking is now so deeply embedded that both deliberate removals and accidental glitches produce measurable harm — making feed algorithms a natural target for regulator-mandated transparency and researcher access.
- If holdout-style experimentation keeps surfacing in leaks and hearings, platform curation choices shift from internal product decisions to externally audited ones, with companies expected to document what they tested and what they learned.
The trend: Algorithmic feed ranking is hardening into infrastructure that platforms can't practically unwind — every test against alternatives confirms its centrality while raising calls for outside oversight of how it works.