Social media platforms, video sites and search engines have become central places where people get news, argue about politics and spend their free time. They raise two linked sets of questions: how they are designed to capture attention, and how they decide what speech to allow.

The attention economy

In 1971 the economist and computer scientist Herbert Simon observed, in paraphrase, that a wealth of information creates a poverty of attention. When content is abundant, human attention becomes the scarce resource. Many online services are free to use and earn money from advertising, so they benefit when people spend more time on them. This is often called the attention economy.

Engagement design

Designers use many techniques to keep people engaged: infinite scroll, autoplaying videos, streaks, badges and frequent notifications. Some draw on variable rewards, a pattern studied by the psychologist B.F. Skinner, in which rewards arrive unpredictably, as when refreshing a feed sometimes produces something exciting and sometimes nothing. As with a slot machine, unpredictability can make a habit hard to break.

Recommendation systems learn what keeps each user watching or clicking. Critics argue that optimizing for engagement can reward outrage and sensational content and crowd out goals such as accuracy or well-being. Defenders reply that people enjoy and value these services, that recommendations help them find things they like, and that companies increasingly measure satisfaction as well as clicks.

Filter bubbles and mental health: the evidence

The activist Eli Pariser warned in 2011 of filter bubbles, in which personalization shows people only views they already hold. Research since then has been more mixed than the phrase suggests: many studies find most people see a wider range of sources than feared, while a smaller group consumes very one-sided media. Political polarization in some countries began rising before social media and has followed different paths in different countries.

A similar debate surrounds young people's mental health. Some researchers argue that smartphones and social media helped cause a rise in teenage anxiety and depression in several countries from the early 2010s. Others find only small average associations and argue that the causes are not yet established. Many agree that effects probably vary by person and by how platforms are used.

Moderation trade-offs

Every platform removes some content, such as spam, fraud and illegal material. Beyond that, choices are hard. Moderation at the scale of billions of posts relies on automated tools and large teams of human reviewers, and both make mistakes. The central trade-off: removing more harmful content also means removing more legitimate speech by mistake, while removing less leaves more harmful content up. Rules about harassment, misinformation or extremism require judgments about context, satire and news value.

Law shapes these choices. In the United States, the First Amendment limits what the government can do to restrict speech; it does not by itself bind private companies. Section 230 of the 1996 Communications Decency Act generally shields online services from liability for what their users post, and protects their good-faith decisions to remove content. Supporters credit it with enabling the open web; critics from different political directions want it changed, some because they see too much removal and others too little. The European Union's Digital Services Act (2022) takes another approach, requiring very large platforms to assess systemic risks, explain their moderation decisions and offer users ways to appeal.

Worked example: a viral health claim

Imagine you help set the rules for a mid-sized online community. A post claiming that a common kitchen spice cures a serious illness is spreading fast. It links to a small study, misreads it and has thousands of shares. Some readers say it helped them; another reports that a relative stopped prescribed treatment after reading it.

Moderators have more options than “leave it” or “delete it.”

1. Leave it up. Protects open discussion and avoids treating adults as unable to judge. Risks real harm if people abandon treatment.
2. Add context. Attach a note linking to reliable medical sources, or let users add community notes. Keeps the speech while informing readers, but notes can arrive too late.
3. Reduce spread. Stop recommending the post without removing it. Less drastic than deletion, though critics say quiet demotion lacks transparency.
4. Remove it. Stops the spread but may look like censorship, and if the moderators turn out to be wrong, they have suppressed a true claim.

Now apply the trade-off. A strict rule (remove any health claim that contradicts official guidance) will catch harmful posts but also remove legitimate debate, since official guidance sometimes changes. A loose rule (remove only illegal content) leaves the harmful post spreading.

Good practice includes clear published rules, consistency, an explanation to the poster, a way to appeal and openness about how often each action is used. Design matters too: if the feed rewards posts that provoke strong reactions, moderators will always be playing catch-up. Many communities choose the middle options for uncertain claims and removal for clear, dangerous falsehoods.