Why the loudest voices online can look like the majority

A political argument can look overwhelming online long before we know how many people actually support it.

You open X and see the same position repeated across posts, replies, and reposts. The same accounts appear again and again. One side seems more active, more confident, and more visible than the other. After enough exposure, it becomes easy to turn that visibility into a much bigger claim: this must be what most people think.

That is where the problem begins.

Social media gives us an enormous amount of behavioral data, but visibility on these platforms is not distributed evenly. Some people participate occasionally. Others post constantly, attract large audiences, or repeatedly enter the same political conversations. When activity is highly concentrated, a relatively small number of users can shape a large share of what everyone else sees.

I wanted to measure how large that gap could become in practice.

Using Arabic-language political discussion in Lebanon, I examined who was actually generating and attracting engagement rather than assuming that every visible interaction represented a different voice. The results of the study showed just how concentrated an apparently large political conversation could become.

The question, then, is not whether social media tells us anything useful. It clearly does. The more difficult question is what exactly we are measuring when a political position appears dominant online—and how far that can be from majority opinion.

When a minority starts to look like the majority

The illusion does not require thousands of people expressing the same opinion. It can emerge when a much smaller number of people participate often enough to occupy a large share of the visible conversation.

One user might comment once on a political issue. Another might reply to dozens of posts, repost similar arguments, and appear repeatedly under journalists, politicians, and influential accounts. Both count as individual participants, but their presence on the platform is completely different.

This becomes especially important during political conflicts and periods of intense disagreement. The people most motivated to participate can also become the people most visible to everyone else. Someone following the discussion may encounter the same arguments hundreds of times without necessarily noticing how many of those interactions come from the same set of accounts.

That creates a simple but powerful illusion: frequency begins to look like popularity.

The problem is not that highly active users are somehow less legitimate. Their opinions are real, and their activity can tell us a great deal about mobilization, organized communities, and the arguments generating the strongest reactions.

The mistake happens when visible activity is treated as if it were a representative sample of the wider public.

I explored this same problem in a Global Voices analysis on the illusion of online public opinion in Lebanon’s political conflicts. Here, I wanted to look underneath that broader argument and ask something more basic: how concentrated was the activity behind what people were seeing?

Instead of treating posts, replies, and reactions as though each represented a different voice, I looked at how engagement was distributed across users.

What the data showed about who dominates the conversation

I examined 15,767 Arabic-language posts on X related to Hezbollah, published by 8,148 different users over eight days in March 2026. For each post, I measured engagement as the combined number of likes, reposts and replies it received.

At first glance, more than 8,000 users sounds like a large and diverse conversation. But once the engagement was grouped by user, the distribution looked very different.

The top 1% of users accounted for 61.5% of all engagement in the dataset.

Expanding the group only slightly made the concentration even clearer. The top 5% accounted for 90.6% of engagement, while the top 10% accounted for 96.2%.

In other words, the overwhelming majority of engagement was concentrated around a relatively small fraction of the users taking part.

This kind of concentration is not unique to Lebanon. Pew Research Center found that 97% of political tweets from U.S. adults in its study were produced by just 10% of users.

That matters because totals can hide the structure underneath them. Two discussions might each contain thousands of posts and interactions while representing very different patterns of participation. In one, attention might be spread relatively widely. In another, a small group of accounts might repeatedly generate or attract most of what becomes visible.

The dataset looked much more like the second pattern.

Media accounts also had a disproportionate presence near the top of the engagement distribution. They represented about 10% of users in the dataset but made up almost 30% of the most highly engaged 1%. Their posts also received more interactions on average than posts from non-media accounts.

That finding matters when interpreting what appears prominent online. Part of the visible conversation is shaped not only by highly active individuals but also by institutional accounts with larger audiences and a greater ability to attract engagement. For journalists observing social media, this can further concentrate attention around a relatively small number of already visible sources.

None of these numbers tells us what Lebanese people as a whole believe about Hezbollah or any other political question.

The dataset was not designed to answer that question. There was no representative sample of the Lebanese population, no polling and no attempt to translate likes or reposts into votes or political support.

What the numbers show is something narrower but important: visibility was extremely unequal.

When someone looked at the most active posts, the most repeated arguments or the opinions attracting the most interaction, they were not looking at an evenly distributed cross-section of thousands of participants.

They were looking at a conversation whose visible surface was heavily shaped by a small minority.

I examined the broader implications of this concentration in a Fair Observer analysis on why loud online voices can be mistaken for majority opinion.

Why the loudest voices become the most visible

Part of the explanation is straightforward: people do not participate at the same rate.

Some users post occasionally. Others treat political discussion as a constant activity. They reply more often, repost more material and appear in more conversations. Over time, those differences accumulate.

If one person comments once and another comments fifty times, both still count as one participant. But they do not leave the same footprint. The second person has fifty opportunities to become visible where the first has one.

Platform design can reinforce that imbalance. A large-scale study of Twitter’s recommendation system found that algorithmic ranking can amplify some political content relative to a chronological feed. Content that attracts visibility can then receive further opportunities for interaction, making visibility partly self-reinforcing.

This does not mean platforms deliberately manufacture a false majority. It means that the signals they are built to respond to — activity and engagement — are not the same signals we would use to measure representative public opinion.

Self-selection can add another layer.

People who care strongly about a political issue may be more willing to discuss it repeatedly than people who are indifferent, uncertain or simply uninterested in arguing online. Someone can hold a political opinion without producing much visible evidence of it at all.

The result is an imbalance in what can be observed. Highly motivated participants generate more posts, replies and reactions. Quieter users generate less data. When we later look at the platform, we naturally know much more about the people who chose to participate most intensely.

That can make intensity look like prevalence.

A position can appear dominant because its supporters are more active, more organized or simply more willing to spend time discussing it. Another position might be held by many people who rarely post about politics at all. Social media makes the first group much easier to see.

For researchers and journalists, this is why raw engagement should be interpreted carefully.

A large number of replies can tell us that a subject is generating intense participation. It does not tell us how many distinct people support the position being expressed, and it certainly does not tell us what proportion of the wider population agrees with it.

That is enough for a minority to occupy a much larger share of the visible conversation than its numbers alone would suggest.

What social media can and cannot tell us about public opinion

The answer is not to stop using social media as evidence. It is to be more precise about what kind of evidence it can provide.

Social platforms are very good at showing us where attention is concentrated. They can reveal which topics are producing strong reactions, which narratives are spreading quickly, which accounts are influential and which arguments are generating unusually intense participation.

That is valuable information.

What they are much less suited to doing is answering a different question: what does the broader public believe?

Those two questions are easy to confuse because the same numbers are often used for both. A post receives thousands of likes, a slogan appears across hundreds of replies, or one political position seems to dominate a discussion. It is tempting to read that visibility as evidence of widespread support.

But engagement is not a representative sample.

This is a well-established methodological problem. Researchers writing in Public Opinion Quarterly have noted that people who post on social media — especially about particular topics — do not necessarily reflect either the platform’s full user base or the wider population.

A like is not a vote. A repost does not necessarily mean agreement. A reply can express support, opposition, sarcasm, confusion or simply a desire to join the conversation. And, as the concentration in this study showed, a large share of visible engagement can accumulate around a very small proportion of users.

This does not mean social media data is unreliable. It means the research question matters.

In another study, I used data from X to examine whether uncertain language makes people more likely to reply. There, I was not trying to infer what society believed from social-media activity. The behavior on the platform was itself the subject of the research.

I compared replies, likes and reposts to understand how users reacted to uncertainty, and found that uncertainty was associated much more strongly with replies than with the other forms of engagement.

For that kind of question, social media is exactly where the evidence should come from. If I want to understand what makes users reply, what spreads, how attention is distributed or how people interact with certain kinds of language online, platform data gives me direct observations of the behavior I am trying to study.

The problem begins when we take those same observations and ask them to answer a different question: how many people in society hold a particular opinion?

The problem exists even before differences in posting activity are considered: Pew Research Center has shown that Twitter users themselves can differ systematically from the broader population.

That distinction is particularly important for journalists.

Phrases such as “Lebanese people believe” or “social media is angry about” can make a platform conversation sound more representative than the evidence allows.

Often, a more accurate description would be narrower: a particular argument is receiving unusually high engagement, a group of highly active accounts is pushing a narrative, or a topic is dominating the visible discussion.

That may sound like a minor change in wording, but analytically it is a very different claim.

The same caution applies to researchers and policymakers. Social media can be extremely useful for identifying emerging narratives, mobilization, polarization, sudden changes in attention and patterns of online behavior.

But when the question is how common an opinion actually is across a population, other forms of evidence are needed: representative surveys, polling, demographic data and, depending on the question, other population-level indicators.

The useful distinction is simple: social media can be a strong measure of visibility, intensity and behavior, but a weak measure of representativeness.

The problem is not using social-media data.

The problem is asking it to answer a question it was never designed to answer.

And once that distinction is kept clear, the loudest voices online can still tell us something important — just not necessarily what the majority thinks.

Research

• Attention Concentration in Online Political Discussions on X — read the full paper on arXiv

Related coverage

• Global Voices — original analysis, with translations in multiple languages
• Fair Observer — related analysis based on the same research
• Eurasia Review — related op-ed on political discourse and unequal visibility on social media
• Al Mayadeen English — independent coverage of the study