Factcheck.org debunks less than one post per day, as human fact-checkers' primary role is seen to be helping Facebook's algorithms to understand fake news sites
The social network's algorithms scan billions of posts each day in hopes of flagging misinformation before it goes viral; humans can't keep up
Context & Ripple Effects
Facebook's fake-news machinery has always been a hybrid of unpaid human judgment and machine scale. The original design from December 2016 had third-party checkers like Snopes and AP flagging stories that got warning labels and lower News Feed ranking, but a 2017 review found those labels arriving too late or not at all — the initiatives were judged largely ineffective. By mid-2018 Facebook responded by doubling down on machine learning and expanded fact-checking partnerships.
The WSJ reporting closes that loop: Factcheck.org debunks fewer than one post per day against billions scanned by algorithms, so the humans' real function is teaching the classifiers which fake-news sites to catch. That reframing matters because later audits found fewer than half of debunked posts actually bore labels, and by 2023 the number of fact-checking sites worldwide was declining as platforms pulled back.
First-order effects
- Factcheck.org and partner checkers are effectively demoted from gatekeepers to training data: their verdicts feed Facebook's classifiers rather than directly policing what users see.
- Facebook gains cover for scale — algorithms scan billions of posts daily while the human layer's tiny throughput makes manual review structurally irrelevant to virality.
Second-order effects
- Unpaid partners like Snopes and AP are doing work whose value accrues to Facebook's models, not their own audiences — an arrangement that pressures the economics of independent fact-checking.
- Because ranking decisions rest on classifier judgments calibrated by sparse human input, errors in site-level classification propagate at machine speed across the News Feed.
Third-order effects
- If platforms keep treating fact-checkers as algorithm tutors rather than publishers of corrections, the profession hollows out — consistent with the later worldwide decline in fact-checking sites even as AI-generated falsehoods multiply.
- Moderation authority consolidates inside the platform's own systems, with external organizations reduced to advisory inputs — a governance model where accountability for false content sits almost entirely with Facebook.
The trend: Platform content moderation is shifting from human fact-checkers as arbiters to humans as calibration data for machine classifiers, concentrating enforcement power inside the platforms themselves.