Five critiques of social media, weighed

Critiques of social media examples come up in almost every discussion of these services, and the same five keep returning. Each of them is partly right, which is why they survive, and each has a specific weak point that goes unexamined because the argument feels finished. Working through them one at a time is more useful than deciding which side to be on.

For each: what it claims, what is solid in it, where it overreaches, and what evidence would move it.

What to take away

One: it is designed to be addictive

The claim. The product is built to maximize time spent, using techniques known to produce compulsive use.

What is solid. The design techniques are real, documented and deliberate. Variable rewards, streaks, counts and interruptions are not accidents, and their purpose is to increase engagement. Snapchat streaks are a concrete example of a feature discussed in this debate. Critic Tristan Harris has argued that persuasive design can capture attention; that is a claim about design, not a clinical diagnosis.

Where it overreaches. The word carries a clinical meaning it is not entitled to here, and using it imports a whole framework of diagnosis and treatment onto a population that has not been assessed. There is a genuine question about compulsive use and it is not settled by the design intent, since designing for engagement and producing a clinical condition are separate claims. The general practice of adding scored progress to ordinary activity, gamification, is used in many places nobody describes this way.

What would move it. Evidence on within-person change: whether the same people, followed over time, change in the predicted direction when their use changes. A platform-specific test could randomly hide Snapchat streak counts for some users and compare their use and compulsive-use symptoms with a control group. The study would also need to separate age, cohort and period effects, a problem discussed under generational digital culture.

Two: it destroyed something that used to exist

The claim. Local institutions, shared attention, patience, or ordinary conversation existed before and do not now. The claim is often applied to platforms such as Facebook, Instagram and TikTok, but naming them does not establish that they caused the loss.

What is solid. Claims of cultural loss can point to declines in civic participation or face-to-face socializing. But a decline in a measure, by itself, does not show that a social platform caused it.

Where it overreaches. The counterfactual. Several long-running changes were already under way, and separating a medium's contribution from the rest requires comparisons that account for other causes. Robert Putnam's Bowling Alone is a prominent account of declining civic participation, not evidence that social media caused it. Memories of earlier platforms such as Myspace are not a reliable baseline either; the problem of reconstructing the past is discussed under early social networks.

What would move it. Compare local-group participation before and after access to a named service or feature changes in matched communities, while accounting for other changes happening at the same time. For example, a comparison centered on Facebook Groups could test a specific claim about local institutions; it would not settle every form of cultural loss.

Three: the algorithm decides what people see

The claim. Ranking determines attention, so whoever controls it controls the conversation.

What is solid. Ranking does determine what is shown, the criteria are not published, and they change. Nobody outside can audit it, which is a real accountability problem regardless of what the ranking does. Facebook's News Feed and YouTube's recommendations make the phrase “the algorithm” concrete, but they are distinct systems rather than one decision-maker.

Where it overreaches. It treats being shown as being persuaded, which skips every step in between. It also treats the ranking as a single decision made by someone, when it is usually an optimization nobody fully understands, including its operators. The narrower version of the concern, that the system produces separated informational worlds, is the filter bubble claim, associated with Eli Pariser's theory that personalization can narrow what users encounter. Evidence for that claim is genuinely mixed rather than absent.

What would move it. A test would compare randomized ranked and chronological feeds, record what users actually saw and did not see, and then measure recall or belief change. That would distinguish exposure, attention and persuasion. Independent researchers would need access to impression-level data, which is exactly the data nobody outside has, and which is itself the strongest version of the accountability complaint.

Four: it is the public square

The claim. These services are where public discussion happens, so exclusion from them is exclusion from public life.

What is solid. For many purposes, they are where the discussion is. Losing access has real consequences that are not adequately described as losing a service.

Where it overreaches. The phrase is doing legal work it cannot support: a square is a place with one set of rules, and these are several private systems with different ones. In the United States, the First Amendment generally restrains government rather than private moderation. Manhattan Community Access Corp. v. Halleck is relevant to the state-action distinction, though it concerned a public-access television operator, not social-media platforms. The claim also collapses two distinct questions, whether a service should be obliged to carry someone and whether it is wise for it to exclude them.

What would move it. A legal argument should name the governing rule and the precise duty being claimed; a civic argument needs a stated public expectation. Usage data alone cannot settle either question. The distinction between legal obligation and a service's desired public role is where arguments like these critiques of social media often stall.

Five: it is just a tool

The claim. The technology is neutral; the outcomes come from people.

What is solid. People do the acting, and the same service supports very different behavior in different groups. Any account that leaves out the people is missing most of it.

Where it overreaches. Defaults are not neutral. What is easy, what is visible, what is counted and what is deleted all shape behavior without anyone deciding, and a design that makes one action cheap and another expensive has taken a position. Langdon Winner's essay ‘Do Artifacts Have Politics?’ is a useful earlier statement of the question, not proof that every platform feature produces one outcome. The pure form of this argument is technological determinism turned inside out, and it fails for the same reason: it puts everything on one side.

What would move it. Compare the same population under different defaults—for example, a chronological feed versus a personalized one—and measure what people do, not just which settings they select. Such a comparison can test whether the interface changes behavior in that setting; it cannot show that people have no agency. The structures that produce these differences are covered under internet communities.

The pattern across the five

Three empirical, one normative, one conceptual. Arguments stall most often when two people are answering different kinds of question, which is worth checking before deciding anyone is being unreasonable.

Common questions

Which of the five is strongest?

The ranking accountability point, because it holds regardless of what the ranking turns out to do, and because its remedy is specific.

Do any of them contradict each other?

The first and the last, directly. Anyone holding both is describing design as powerful when criticizing and neutral when defending.

Why does naming a falsifier matter so much?

Because it separates a claim from a position, and both are legitimate. The trouble comes from positions presented as claims.

Are there arguments missing here?

Several, including whether personalized feeds narrow exposure through filter bubbles. That is distinct from proving that exposure persuades users. These five were chosen because they recur across discussions and because each fails in an instructive and different way.