Feeds, Friction, and Attention

Read

The Queue You Cannot See

1. The crowded square

Harbour Forum was designed as an online meeting place for a small town. Residents could announce events, ask for advice, debate local decisions, and share photographs. Its home screen was called the Square. No one saw every new post there. A ranking system selected and ordered items for each account.

When the council proposed moving a bus stop, the Square filled with argument. Lina opened the app and saw a video titled *They Have Already Decided*. Her neighbour Dev saw a route map and a careful explanation from a transport planner. Both believed they had entered the same public conversation. In practice, they had entered different queues.

This did not mean that the system had secretly assigned them fixed opinions. It meant that visibility had been organised before either person began reading. The design had helped determine which claims felt urgent, common, or absent.

2. A prediction is not a purpose

The forum team explained that the ranking system predicted which posts a person might engage with. Past actions supplied signals: opening an item, pausing, replying, hiding, sharing, or returning later. The system could also use features of the post and patterns among accounts. It then produced an ordered feed.

But a prediction such as “likely to receive a reply” is not the same as a public purpose such as “likely to support an informed local decision”. A measurable signal becomes a proxy when it stands in for something harder to measure. Replies may indicate usefulness, confusion, delight, anger, correction, or conflict. The number records activity without explaining its quality.

The important question was therefore not only whether the prediction worked. It was what the system had been instructed to predict, and what human value that target was supposed to represent.

3. The difference made by friction

The video reached Lina after several neighbours had shared it. Its final image showed an empty pavement beside the words *No one asked us*. A bright button invited her to pass it on immediately. Sharing required one tap; opening the linked council document required leaving the forum and finding the relevant section.

This imbalance is a form of design power. Friction is the effort, time, uncertainty, or interruption attached to an action. Removing friction can make a service accessible. Adding a pause can prevent an impulsive mistake. Neither low nor high friction is automatically good. Its effect depends on where it appears and whose goal it serves.

Harbour Forum had made reaction easy and context comparatively difficult. The interface did not force Lina to misunderstand the proposal, but it changed the path of least resistance.

4. From signal to feedback loop

As the video attracted replies, the ranking system treated the activity as evidence that more people might engage. Wider visibility produced more reactions. Those reactions generated stronger ranking signals, which produced still wider visibility. A feedback loop had formed.

The loop did not create the original concern. Residents genuinely disagreed about distance, safety, and access. Nor did every viewer respond in the same way. Some challenged the video's implication; others shared it as an example of misleading rhetoric. Yet their corrective activity could still increase the video's measured importance.

Cause had become distributed across a system: the creator's framing, the audience's responses, the ranking target, the sharing interface, and the timing of the dispute. Blaming only “the algorithm” would hide human choices. Blaming only users would hide the environment in which those choices gained reach.

5. Privacy beyond secret messages

At a public meeting, Dev asked whether the feed used private information. The product manager answered that posts on Harbour Forum were public and that the system did not read private messages. The answer sounded reassuring, but it addressed only one part of the question.

Privacy also concerns patterns inferred from ordinary behaviour. Repeated pauses on housing posts, late-night visits, location settings, group membership, and hidden items may suggest interests or circumstances that a person never stated. An inference is not necessarily true, yet it can still shape what the system shows or withholds.

The issue was not simply whether one data point was secret. Residents needed to ask what was collected, what could be derived, how long signals remained useful, who could act on them, and whether people could meaningfully refuse without losing access to the town's main forum.

6. Virality changes the context

A week later, an account outside the town reposted the bus-stop video with a mocking caption. The clip moved into larger networks whose users knew nothing about the route. Local phrases became slogans. Replies arrived faster than the forum's small moderation team could interpret them.

Virality is often described as if content spreads because it contains an irresistible property. Content matters, but spread also depends on network connections, recommendation, timing, copying tools, and the ease of moving material between contexts. A statement made for neighbours may be judged by strangers under different assumptions.

Reach can expose neglected concerns and connect people who would otherwise remain isolated. It can also detach a claim from the evidence, relationships, and opportunities for repair that gave it meaning. Scale is not just a larger audience. It can be a different communicative situation.

7. Community is also governance

The team first treated moderation as rubbish collection: remove threats, obvious abuse, and repeated spam after users reported them. The bus-stop dispute revealed a wider task. Rules also shaped who could speak repeatedly, how corrections travelled, whether coordinated disruption was visible, and what happened when technically permitted conduct drove others away.

No policy could guarantee a comfortable conversation. Strong disagreement was part of the forum's purpose. But “leave everything up” was not neutral. It allocated the costs of attention, verification, and harassment to individual users. Aggressive enforcement also carried risks: vague rules could silence unpopular positions or be applied unevenly.

Governance required inspectable definitions, proportionate responses, appeal routes, and evidence about effects. It also required admitting that a platform rule is a decision about participation, not merely a technical setting.

8. Polarisation is not a screen effect alone

Some residents said the feed had divided the town. Others replied that the conflict already existed. Both claims contained part of the picture.

A media system can repeatedly foreground the sharpest contrast, reward identity-signalling language, or make opponents appear more uniform than they are. These conditions may harden interpretation. Yet people bring histories, institutions, interests, and offline relationships to the screen. A feed does not manufacture every division, and changing a ranking rule cannot repair every cause.

This distinction matters because a broad diagnosis can produce a shallow cure. If the problem is only described as “polarisation”, designers may suppress visible disagreement while leaving unequal access or unresolved policy conflict untouched. The team needed narrower questions: Which voices were overexposed? Which misunderstandings persisted? Which design choices rewarded escalation? Which disagreements were legitimate and needed a better forum rather than less visibility?

9. Negotiating a better target

Harbour Forum invited residents, moderators, council staff, accessibility advocates, and the product team to propose changes. One group wanted a chronological feed. Another wanted verified local information placed first. A third feared that official priority would turn public communication into government control.

The meeting moved from positions to criteria. A workable policy should preserve access to dissent, reduce contextless resharing, make ranking influences more legible, protect sensitive behavioural data, and remain usable for people with limited time or digital confidence.

The resulting proposal combined measures: a clear choice between recent and ranked views; a prompt before sharing unopened linked material; visible context on rapidly circulating local posts; shorter retention for selected behavioural signals; and a public explanation of moderation and appeal. None was a complete solution. Together they made the system's purposes easier to question and revise.

10. Attention is an allocation

When Lina next opened the Square, the video was still available. Beside it appeared the route document, a summary of unresolved questions, and an explanation of why the post was receiving extra context. She could disagree with the council, challenge the summary, or leave the discussion. The redesigned forum had not delivered a correct opinion to her.

It had changed the conditions of attention. Every feed allocates limited visibility: now rather than later, this voice rather than another, reaction before context or context before reaction. Even a chronological list reflects choices about membership, posting, repetition, and notification.

The democratic question is therefore not whether a platform can become perfectly neutral. It is whether its allocations can be seen, contested, and connected to defensible purposes. Users remain responsible for what they create and share. Designers and institutions remain responsible for the environments that amplify, interrupt, remember, and rank those actions.

> Keep this idea: A media system shapes attention through targets, data, ranking, interface friction, network effects, and governance. Responsibility is distributed, but it does not disappear.