In 2018, a team led by MIT Media Lab researchers Soroush Vosoughi, Deb Roy, and Sinan Aral published a study in Science examining roughly 126,000 stories that had spread on Twitter over more than a decade. Their finding, widely cited since, was blunt: false news traveled faster, farther, and deeper than true news, and human behavior — not bots — was the primary driver, because novelty and emotional charge are what people reliably pass along.
The mechanism behind that finding is simple and well understood: recommendation systems used by platforms like YouTube and Facebook's News Feed are optimized to maximize engagement, and engagement correlates most reliably with strong emotional response — outrage, fear, tribal validation — rather than with accuracy. Nothing in the design of a ranking algorithm distinguishes between a claim that is emotionally activating because it is true and one that is emotionally activating because it is false and simple.
What spreads, under this optimization, is not the same category of thing as what holds up. A claim can hold up for years without ever spreading widely, because careful, qualified, or boring claims are usually the accurate ones and accuracy is not what the ranking function rewards.
The people most exposed are not the sophisticated consumers of media who already discount viral claims by habit. It is everyone else, for whom "this is everywhere" functions as an informal credibility signal that engagement-driven ranking has quietly hijacked.
The European Union's Digital Services Act, which took full effect for the largest platforms in 2024, is the first regulation to attack this at the mechanism level rather than the content level: it requires the largest platforms to give regulators and vetted researchers access to how their recommender systems actually rank content, and to offer users at least one feed option that is not based on profiling.
Platforms had already responded to years of criticism with labels, friction, and fact-check partnerships — all of which treat the problem as a content moderation issue rather than an incentive issue. None of those fixes touched the underlying optimization target, which continued to reward the same category of content regardless of what got flagged after the fact.
A ranking system audited under the DSA could, in principle, be required to show whether it is rewarding durability and correction rates alongside raw engagement, rather than raw engagement alone. That is a harder engineering and business problem than a fact-check label, which is precisely why platforms resisted it until a regulator made resistance the more expensive option.
Whether that audit power translates into a different ranking function, or just produces a more precisely documented version of the same one, is the actual test of the next few years. Until there is an answer, the safest assumption for any reader is that virality still measures resonance, not truth.