The internet still produces sensations that appear to be everywhere. Yet the numbers beneath a viral post can now say less about shared culture than they once did. In an interview with WIRED, cyber-ethnographer Ruby Thelot argues that online attention has fragmented into highly visible but isolated “digital islands,” where a trend can dominate one community without reaching far beyond it.
That distinction matters because platforms make scale look like cultural importance. Millions of views may reflect an algorithm repeatedly serving content to a narrow audience, a coordinated attempt to manufacture momentum or both. The result is not the end of online culture, but a widening gap between what is measurable and what people genuinely carry into their lives.
Virality can be engineered without becoming culture
Thelot describes today’s internet as intensely localized. People may inhabit overlapping platforms while seeing very different feeds, jokes and anxieties. A post can therefore appear unavoidable to one group and remain invisible to another. Older ideas of a single mainstream internet are poorly suited to this environment.
He connects the problem to Goodhart’s law: once a measure becomes a target, it stops functioning as a reliable measure. Views, likes and shares began as imperfect signals of audience response. Creators, brands and platforms have since learned to optimize for them directly. High engagement can now indicate successful distribution tactics rather than lasting cultural influence.
That does not make metrics meaningless. They still describe activity inside a system. The mistake is treating that activity as proof that a trend changed how a broad public thinks or behaves. Thelot’s account suggests a more demanding test: whether an idea survives outside the feed, spreads between communities and remains relevant after the recommendation cycle moves on.
The loudest narrative may not reflect the evidence
One example is “heteropessimism,” a label for bleak attitudes toward heterosexual dating. Online discussion can make disillusionment seem universal. Thelot says he examined roughly 1,000 Instagram and TikTok videos with more than one million views and found that negative dating content hovered around 25 percent.
That finding, as he presents it, does not show that dating frustration is invented. It suggests that the dominant online narrative may exaggerate its prevalence. Negative material can be more memorable, more discussable and more likely to be amplified, creating a sense of consensus that the underlying sample does not support.
The same lens applies to prediction markets. Their rise might be read purely as a financial development, but Thelot sees participation and group identity as central attractions. People are not only pricing an outcome; they are joining a community organized around watching events, displaying judgment and comparing beliefs. The market becomes a social object as well as a wager.
Both cases show why counting visible behavior is not enough. Researchers and editors need to ask what motivates participation, which group is participating and whether platform design is magnifying one emotional register over others.
AI forecasts risk repeating the same narrow assumptions
Thelot is also skeptical of the most sweeping AI-apocalypse narratives. His argument is not that powerful AI presents no danger. It is that intelligence alone does not automatically solve every practical problem or produce control over complex societies. Cooperation, coordination and embodied experience remain constraints that raw analytical capability may not erase.
He also warns that technological debates can undervalue forms of intelligence associated with emotion, the body and artistic practice. That is partly a philosophical position rather than an experimentally verified conclusion, but it highlights a familiar measurement error: treating the capacities easiest to benchmark as the only ones that matter.
Online trends and AI predictions share this weakness. Both can convert a narrow signal into a totalizing story. A viral metric becomes “what everyone thinks”; performance on selected tasks becomes a forecast of inevitable dominance.
The better response is not to abandon data, but to examine its boundaries. Who generated the signal? What system rewarded it? Does it travel beyond its original digital island? In a fragmented internet, cultural significance must be demonstrated through context and persistence—not inferred from a large number on a screen.
