The entire visible culture of social media — the debates, the trends, the influencers, the outrage — is generated by roughly 1% of users, while the other 90% watch in silence, which means the picture we have of what people think online is profoundly wrong

Open a social platform and it appears that millions of people are speaking at once. Arguments rise, phrases repeat, a video format spreads and an opinion begins to feel unavoidable. Yet the feed is not a census of what users think. It is a display assembled from the activity of a highly unequal minority, selected again by ranking systems that decide which contributions become visible.

The familiar shorthand for this imbalance is the 90–9–1 rule: 90 per cent of users observe, 9 per cent contribute occasionally and 1 per cent account for most contributions. It is a useful warning about participation inequality. It is not, however, a universal measurement showing that exactly 1 per cent of today’s users generate the entire culture of every social platform.

The percentages vary substantially by platform, period and definition of participation. The durable finding is the concentration itself. A small, unusual group produces a disproportionate share of the material from which everyone else is invited to infer what “people online” believe.

The 1% figure began as a rule of thumb

Jakob Nielsen popularised the 90–9–1 formulation in 2006, drawing on earlier observations of online communities. His original account of participation inequality used examples including Wikipedia editing and online discussion boards. It described a recurring skew, not a law of human behaviour.

The categories were also more nuanced than creator versus spectator. The 9 per cent contributed intermittently, while the 1 per cent did so heavily. The remaining 90 per cent were described as lurkers, but even that term can conceal several behaviours: reading, watching, searching, forwarding privately, following links or using information in conversations elsewhere.

Later research found distributions that resembled the rule without reproducing it exactly. A 2014 observational study of four digital health communities reported that fewer than 1 per cent of users were “superusers”. Those members produced between roughly one-quarter and two-fifths of posts, depending on the community. Occasional contributors produced more, and lurkers produced none. The study in the Journal of Medical Internet Research supported the broad idea of participation inequality while showing why the headline ratio should not be treated as a precise allocation of all content.

Modern platforms remain extremely unequal

Large social platforms have changed since 2006. Posting is easier, creation tools are built into phones, and recommendation feeds routinely show content from strangers. None of that has made production evenly distributed.

In a 2019 analysis matching a representative survey with account data, Pew Research Center found that the most prolific 10 per cent of adult US Twitter users generated 80 per cent of tweets. The median user posted twice a month. Those active users also differed from the broader population in age and political composition, as detailed in Pew’s comparison of Twitter users with the US public.

A follow-up during the 2020 election period found an even steeper distribution: 10 per cent of US adult users produced 92 per cent of tweets in the study window. It would still be inaccurate to turn 10 per cent into 1 per cent, but it would be equally inaccurate to treat the resulting stream as a balanced cross-section of all users.

TikTok shows a different ratio and the same underlying structure. Pew’s study of US adult TikTok accounts found that about half had never posted a video. Among publicly accessible videos, the most active quarter of users produced 98 per cent. The TikTok analysis also found that a typical user follows far more accounts than they have followers, reinforcing the division between a large audience and a smaller production class.

Production inequality is only the first filter

Posting does not guarantee visibility. A second concentration occurs after content is created. Recommendation systems, follower networks, reposts and news coverage distribute attention unevenly. An occasional post from an ordinary account may reach a few dozen people, while one professional creator can appear in millions of feeds.

This means “the 1 per cent” should not be imagined as a stable club of individual hobbyists. Visible social media includes full-time creators, political campaigners, celebrities, journalists, companies, institutions, automated accounts and users who post compulsively without a commercial role. Their incentives differ, but they share one relevant characteristic: they produce enough public material to become legible to the platform.

The ranking system then learns from viewers as well as posters. Watching, pausing, liking and sharing affect distribution, so silent users are not without influence. Yet their influence is indirect and aggregated. The public object remains the creator’s post, while the audience’s reasons for viewing it are largely invisible.

Visible opinion is a selected sample

The distortion becomes most consequential when posting behaviour is mistaken for public opinion. People who choose to speak about politics online are not randomly selected from either a platform’s users or the population beyond it. They may be more politically interested, more certain, angrier, more professionally invested or simply more comfortable with public disagreement.

In 2024, Pew surveyed more than 10,000 US adults about political activity across TikTok, X, Facebook and Instagram. Many users avoided political posting, and among those who did, fear of criticism or harassment was a common reason. The cross-platform report shows that silence can reflect the anticipated cost of speaking, not indifference or agreement.

Research also suggests that participation can be shaped by what users think others believe. A 2025 field experiment on Reddit began from the observation that political discussions are often dominated by power users while most participants remain silent. The published study of participation in online political discussion examined how users select into those roles and how visible discussion conditions affect willingness to contribute.

A feed can therefore create a feedback loop. Highly motivated users establish the apparent tone. Less certain users read that tone as evidence of the local norm and decide whether speaking is worth the cost. Their silence leaves the original signal looking more representative than it is.

The silent majority cannot be assigned an opinion

There is a tempting mistake on the other side. If vocal users are unrepresentative, the quiet 90 per cent must secretly believe the opposite. The data do not justify that conclusion either.

Silence is missing information. A person may agree, disagree, feel ambivalent, lack knowledge, dislike public posting or use the platform only for entertainment. Even behavioural signals can be ambiguous. A view may indicate curiosity or dislike. A share may express endorsement, ridicule or a desire to discuss something privately.

The responsible conclusion is not that visible online opinion is false in a predictable direction. It is that its relationship to broader opinion is unknown until measured through a better sampling method. Representative surveys, carefully designed platform studies and transparent behavioural data can test that relationship. Trending lists, comment sections and viral posts cannot.

The feed is a stage, not the room

Social media’s visible culture is real. Trends alter language, creators build businesses and online outrage can impose material consequences. A minority can shape culture without representing the majority.

What the participation figures change is the meaning of what we see. The feed records what a selected group chose to make public and what a platform chose to amplify. It does not reveal the full distribution of belief among those watching, still less among society beyond the platform.

The 1% rule survives because it captures that asymmetry in one memorable number. Its literal precision has not survived contact with different platforms. The deeper warning has: the loudest available sample is usually the easiest one to observe, and often the least defensible one from which to infer what everybody thinks.