How YouTube's recommendation algorithm can promote divisive clips and conspiracy videos and may have spread disinformation during the 2016 election
An ex-YouTube insider reveals how its recommendation algorithm promotes divisive clips and conspiracy videos. Did they harm Hillary Clinton's bid for the presidency? Tweets: @zeynep and @profcarroll Tweets: Zeynep Tufekci / @zeynep : A fascinating, important in-depth investigation of how YouTube's recommendation algorithm apparently functioned in the 2016 election: no matter which political side the researcher started from, the platform pushed pro-Trump, anti-Clinton videos. http://www.theguardian.com/... http://twitter.com/... David Carroll / @profcarroll : “YouTube was six times more likely to recommend videos that aided Trump than his adversary.” http://gu.com/... “Roughly half of the YouTube-recommended videos in the database have been taken offline or made private since the election...” http://gu.com/... http://twitter.com/... Expand More For Next 2 Unexpand More For Next 2
Context & Ripple Effects
This investigation is the opening entry in a multi-year argument over whether YouTube's recommendation engine shapes elections. An ex-insider account, amplified by Zeynep Tufekci and David Carroll, claimed the algorithm pushed viewers toward pro-Trump, anti-Clinton videos regardless of their starting point. Later reporting cut both ways: a report on right-wing content's spread argued supply and demand, not the algorithm, drove proliferation, while YouTube trained its recommender to demote conspiratorial videos — only for organic sharing and bot promotion to blunt the fix.
The follow-through matters because each intervention displaced rather than ended the problem: a study found cutting recommendations to fringe channels worked but made Fox News the most recommended source for election videos, a Platformer report tied YouTube's 2020 misinformation role to its popularity and permissive policies, and a former alt-right creator described deliberately engineering confrontations for the echo chamber.
First-order effects
- YouTube's core product — its recommendation system — is publicly tied to 2016 election disinformation, with David Carroll citing a six-fold likelihood of pro-Trump pushes, creating immediate reputational exposure for the platform's ranking choices.
- Zeynep Tufekci and David Carroll convert insider testimony into a flagship case study, putting algorithmic amplification of divisive video in front of policymakers and advertisers.
Second-order effects
- YouTube responds over the following years by teaching its recommender to recognize and demote conspiratorial videos (its demotion system), yet organic link-sharing and bot promotion circumvent it, and removing fringe channels from recommendations redirects election-video traffic toward mainstream outlets like Fox News.
- Counter-research forces the debate onto creators' incentives: if right-wing content spreads through supply and demand rather than the algorithm, YouTube's liability case weakens while the economics of outrage-driven production come under scrutiny.
Third-order effects
- Recommendation systems get treated as election infrastructure subject to audit and moderation, sharpening the question of distribution-layer liability for algorithmically amplified content rather than merely hosted content.
- If the pattern holds, platform interventions will keep displacing amplification rather than eliminating it — suppressing fringe channels while concentrating recommendation power in whichever mainstream outlets remain, leaving the structural driver contested between code and audience demand.
The trend: Platform recommendation algorithms are shifting from opaque engagement optimizers into moderated, politically audited distribution infrastructure whose interventions reshape, rather than remove, problematic amplification.