A parent opens an app to check whether a school meeting has changed time. Before finding the notice, the screen offers an argument they never followed, a run of short videos, and several notifications delivered together even though they were written hours apart. None of those items needs to be unlawful or abusive to reshape the visit. Their order has already made choices on the person’s behalf: what deserves attention first, what emotional temperature the session will have, and how much time seems normal to spend there. That is the part of the social-media debate that often disappears. Asking what should be removed matters, but so does asking what gets pushed forward, what gets delayed, and who can challenge that allocation.
Removal remains necessary when a service is dealing with threats, scams, or targeted harassment. Yet a digital community is not made safe simply by taking away the most visibly harmful material. Think about a public square. It also matters whether people can sit down, whether directions are legible, whether quiet voices have room, and whether essential movement is blocked by noise. Distribution performs that kind of civic planning on a screen. It gives some signals priority, turns others into background, and uses every pause, reply, or exit as evidence for arranging the next visit.
Ranking is a continuing judgment#
“Relevant content” sounds neutral until we ask relevant to whom, for what purpose, and at what cost. A ranking system can help someone find a message from their sister. It can also detect a fleeting curiosity and turn it into a stream of increasingly similar material. The distinction is not merely technical. It lies in the objectives people set, the trade-offs they accept, and the way success is counted. If the main measure is time spent, the product will learn to extend time spent. If it also values reaching useful information, returning to an intended task, or avoiding unwanted interruption, it has to account for very different outcomes.
“Relevant content” sounds neutral until we ask relevant to whom, for what purpose, and at what cost.
That is why treating feed order as a cosmetic convenience is misleading. The order is the consequential choice, repeated thousands of times a day. It shapes which subject feels urgent, which creator finds an audience, and which rumour takes on an outsized presence. A post may circulate without being recommended, but its reach changes dramatically when a service puts it in front of people who neither sought it nor know its author. That expansion is not a natural event. It is a specific way of editing a shared experience.
Calling it editing does not suggest that an employee manually reviews every screen. It means that someone chose goals, measurements, data sources, thresholds, and exceptions. When those decisions vanish behind the word “personalization,” public discussion loses its footing. A user sees only an individual result: these videos appeared for me. But the accumulation of individual results distributes attention among neighbourhoods, languages, and styles of speech. It rewards formats that fit the machinery and sidelines information that needs a little context before it can make sense.
An ordinary example#
Imagine the account of a local health clinic. It posts a changed schedule for a vaccination session. On the same day, a neighbourhood group advertises a free workshop. Both notices are useful to particular people for a short time. They nevertheless compete with videos made to produce instant surprise, comments engineered to heat an argument, and accounts posting around the clock. If the ranking is driven by immediate reaction, the health notice may sink from view. Nobody has banned it. A system has simply decided that something else belongs in the front row.
This example helps us leave behind a poor choice between a bare chronology and an all-knowing algorithm. A service can offer several meaningful ways in: recent activity from chosen accounts, a local channel with understandable rules, a practical-information digest, and a personalized selection that can be paused. The essential point is that those choices must be real, visible, and capable of producing recognizably different results. An alternative buried at the bottom of a settings menu does not return control. It merely makes control a test of endurance.
We should also distinguish recommendation by proximity from recommendation by capture. Suggesting a library event to someone who follows that library or lives nearby may have a clear purpose. Chaining together posts that intensify the same insecurity because a person paused on one for a few seconds is something else. The first practice can be defended through usefulness and the control retained by its recipient. The second needs a much stronger explanation: what concrete benefit does it offer, and why cannot that benefit be achieved through a less intrusive intervention?
Counting the cost of interruption#
The usual measures are attractive because they are easy to count: clicks, plays, comments, minutes on screen. But they do not show the cost of being interrupted. Someone may tap a notification because they fear missing out, not because they want to open the service. They may comment while angry and regret it later. They may watch twenty videos in succession without remembering one. When a design interprets every response as automatic approval, it mistakes captured attention for attention freely given.
A more responsible approach begins by measuring the frictions that have traditionally been ignored. How often does a notification break into a task a person has identified as important? How long does it take to find a saved post? What share of recommendations is immediately hidden? How many people voluntarily switch ranking modes? Do some groups receive more alarming material, more dubious advertising, or more content that fuels comparison? No single metric settles the question, but together they prevent wellbeing from becoming a cheerful label at the bottom of a privacy policy.
This perspective can change ordinary product work. Before launching a feature, a team can state a testable claim: “This reminder helps people return to conversations they chose without multiplying unwanted interruptions.” It can then be tested with a small group, frequency limits, and an expiry date. Numbers should be paired with conversations about lived experience. If the evidence shows that the reminder distracts more than it helps, it should be withdrawn. Learning to let a feature go is a design capability, not a commercial failure.
Explanations that lead somewhere#
Useful explanations are neither technical treatises nor reassuring slogans. When a recommendation appears, a person should be able to see in ordinary language what shaped it: they follow similar accounts, someone they talk to shared it, the subject is local, or the item is generating broad activity. That explanation must lead to an effective action. “Show me less of this,” “I prefer the newest posts,” “do not use my activity beyond this topic,” and “reduce these alerts” should change what happens on screen, not merely add a note to an invisible record.
Transparency also has a collective dimension. Researchers, consumer groups, and journalists do not need the identity of every user to examine whether an update concentrates visibility, changes what young people encounter, or buries essential public information. They need aggregated data, safe procedures, and enough continuity to compare changes over time. If the company alone interprets its experiments, the public learns about effects only after habits and harms have taken root. Independent observation is not hostility to technology; it is what makes correction possible.
The same principle applies inside organizations. A team should be able to name who owns a distribution decision, what harm threshold triggers a review, and how an affected person can report an unexpected result. “The model did it” is not an answer. Models are deployed within budgets, interfaces, targets, and release schedules selected by people. Clear responsibility makes it possible to stop a damaging experiment before it becomes routine, instead of turning every complaint into a request for a generic help centre.
Making room for a person’s own time#
Setting rules for distribution does not mean freezing social media or deciding from an office what everyone ought to find interesting. It means setting a boundary of dignity: no one should have to surrender disproportionate attention to recover basic information, and no one should have to accept intensive personalization in order to take part in their community. Platforms can compete to be more imaginative, useful, and pleasant without competing to occupy every available minute.
The simplest design question comes before an app is opened: “What did I come here to do?” If the answer is to check a message, the route should not pass through an obstacle course built to lengthen the session. If the answer is to explore ideas, the person deserves an exploration mode that explains its rules and can be left without losing access to their contacts. Designing exits is as important as designing entrances.
This approach also gives institutions a more practical agenda than vague calls for ethical algorithms. Require meaningful ordering choices. Limit the frequency of attention-grabbing prompts. Audit effects on access to public-interest information. Make explanations actionable. Support independent research with safeguards for privacy. These are not promises that every online conversation will become calm or fair. They are ways to make power visible where it currently hides in default settings.
The next time a controversy calls for a post to be removed, that request may be justified. But another question should accompany it: how did this material arrive here, and what system made it feel unavoidable? The answer will include incentives, interfaces, measurements, notification schedules, and human choices. Precisely because it is a chain of choices, it can be changed. Distribution stops looking like fate once we recognize it for what it is: a contestable architecture that should be explainable, auditable, and revisable by the people whose days it helps to arrange.
