Study of 225 pieces of COVID-19-related content rated false or misleading: 59% remain up with no warning label on Twitter, 27% on YouTube, and 24% on Facebook
Facebook and YouTube do better, but still leave some misinformation up — More than half of the misinformation …
Context & Ripple Effects
A month after reporting that coronavirus misinformation was persisting across all three platforms despite their partnerships with health organizations (the March New York Times account), the Post now has numbers: of 225 flagged COVID-19 items, 59% remain up with no warning label on Twitter, 27% on YouTube, and 24% on Facebook.
The ranking matters because each company had publicly framed its response as working — the study converts that framing into a measurable enforcement gap, with Twitter the clear laggard among the three.
First-order effects
- Twitter faces the most direct exposure: nearly six in ten flagged items stayed up unlabeled, undercutting its stated collaboration with health authorities and inviting scrutiny of its labeling pipeline.
- Facebook and YouTube can point to comparatively better rates, but both still leave roughly a quarter of flagged content untouched, so none of the three can claim clean hands on the same dataset.
Second-order effects
- Fact-checking groups gain ammunition to push from spot audits to systematic monitoring — the approach Avaaz later applied when it found 56% of fact-checked COVID misinfo in major non-European languages unactioned by Facebook (its European-language audit).
- Attention shifts toward high-reach accounts rather than volume alone, anticipating the 'superspreader' framing NewsGuard used when it found Twitter failing to act on large verified accounts (the NewsGuard report) and Facebook's later dispute over twelve such accounts.
Third-order effects
- If the pattern holds, platform moderation becomes a permanent measurement battleground: independent audits quantify enforcement gaps, platforms dispute the methodology, and the cycle repeats across topics — as later label reviews of midterm candidates' false claims showed the same Twitter-Facebook shortfalls persisting years on (the midterms label review).
- Engagement economics keep the incentive structure tilted against removal: the NYU finding that known misinfo publishers drew six times more engagement than sources like WHO suggests the gap is not just capacity but business model.
The trend: Platform content moderation is converging on an audit-driven accountability loop in which independent studies measure enforcement gaps faster than the platforms close them, making moderation performance a recurring public metric rather than a one-off controversy.