The Audience Becomes the Factory

Soon enough, the blank box looked back.

“What’s on your mind?” Facebook asked, in a tone of mild concern that almost hid the commercial ambition under it. Twitter wanted to know what was happening. YouTube offered an upload button and a progress bar. Instagram, a little later, made the whole proposition tactile: take a picture, improve it, publish yourself. The internet, which had spent its first popular decade arranging pages for you to read, began requesting copy.

You filled in the field. You cropped the photo. You tagged the friend. You tried one caption, then another. You watched the small red numbers arrive. Sometimes you deleted the thing and posted a better version an hour later.

Ad buyers could describe this, for a while, as better display. MySpace carried banners and takeover units. Facebook offered cleaner pages, a more upscale audience composition, and the irresistible promise of real identity. YouTube sold pre-roll against videos it had not paid to make. Twitter eventually inserted promoted tweets into a stream already composed of tweets that other people had written for free. The machinery was familiar enough that budgets could move.

Fine as far as it went. The missing line item was labour.

Old media companies had to buy or commission the thing that kept your eyes in place. Networks paid for pilots, series orders, talent, sets, crews, rights, distribution. Publishers paid for writers, editors, photographers, paper, print runs, delivery. Even search, for all its strange brilliance, had to crawl and index a web whose pages had been produced elsewhere. Social platforms did something more radical and more economical. They arranged matters so that the audience supplied the raw material.

This was the innovation. Not the banner, not the profile page, not even the targeting. The industrial trick lay in persuading millions of people to produce the content that would attract and retain millions of other people, while the platform owned the distribution, the measurement, and the margin.

Silicon Valley discovered a labour force willing to write, photograph, film, edit, caption, tag, sort, recommend, review, and moderate, provided the arrangement could be described as sharing.

That sounds severe, so it helps to look at the product choices. Early Facebook, before News Feed, was mostly a set of destination pages. You went to a person’s profile much as you went to any web page: on purpose. The site held identity in place. It did not yet fully weaponise activity. Then, in 2006, Facebook introduced News Feed and Mini-Feed. Suddenly the site stopped waiting for you to visit someone’s page and began telling you, continuously, what everybody had just done. Your changed relationship status, your new photos, your added friends, your comments, your affiliations—tiny units of behaviour that had existed as scattered facts on separate pages—were collected into a stream and delivered outward.

The backlash was immediate and, from the company’s point of view, revealing. Thousands protested. Groups formed demanding the feature’s removal. The complaint was partly about privacy, but privacy was not the whole of it. People felt, quite accurately, that a quiet archive of personal information had been turned into a broadcast medium overnight. The same actions were now more consequential because the distribution had changed.

That distinction sounds technical until you feel it. A page waits. A feed arrives. A page requires intent. A feed manufactures it.

Once you see that, a great deal of apparently miscellaneous product design begins to look like operations management. The Like button, introduced in 2009, was not a cute embellishment. It was a cheap, scalable signal of preference, easier than a comment and more informative than a glance. Retweet, first improvised by Twitter’s own population and then built into the product, did not simply make sharing convenient. It turned distribution into a reflex. Follow buttons, reaction buttons, autoplay, notifications, pull-to-refresh, infinite scroll: these were methods for increasing supply, reducing friction in circulation, and harvesting measurement.

The wrong model lingers because it flatters everyone. It flatters the platforms, who get to call themselves neutral hosts. It flatters the people on them, who get to feel like participants in a cultural commons. It flatters advertisers, who get to imagine they are buying adjacency to authentic human connection instead of renting access to a machine that has converted that connection into inventory.

For a time, the language helped preserve the illusion. “Social” sounded warm. “Sharing” sounded civic. “Community guidelines” suggested a parish hall with a clipboard. Even “friend” did useful work, turning a database relation into a moral category.

You could watch the euphemism fail at the edges. On YouTube, people who uploaded regularly began to behave less like hobbyists and more like programmers with deadlines. Thumbnails were tested. Titles were revised. Intros shortened. Release schedules fixed. Audience retention graphs studied with the pinched seriousness once reserved for shop-floor output. Facebook publishers learned that one style of headline would travel and another would die in place. Twitter taught people the performative value of speed and the commercial value of being legible at speed. Instagram turned the angle of a coffee cup into a distribution question.

The platforms did not need to issue employment contracts for any of this to look suspiciously like work.

You may have felt it without naming it. The small negotiations with yourself before posting. The moment of delay while you decide whether a joke is worth the risk of silence. The second attempt at the video because the first one looked flat. The glance at the analytics. The little lift when the numbers rise faster than expected. The unspoken embarrassment when they do not.

Plenty of social activity is just social activity. Friends do talk to friends. Families do share pictures. People do make things for reasons that have nothing to do with monetisation. The point is not that every post is a shift clocked by management. The point is that the platform’s economics improve when ordinary expression becomes machine-readable production. A birthday photo can be both sincere and useful to the system. A recommendation for a restaurant can be both genuine and unpaid commercial distribution. Your affection and the platform’s margin are perfectly capable of coexisting. That is one reason the arrangement proved so durable.

The old media analogy also fails on cost structure. Television had substantial content costs before it sold the first thirty-second spot. A network betting on a season could make a very expensive mistake. Social platforms shifted a remarkable portion of that risk outward. People bought their own cameras. They paid their own broadband bills. They wrote the copy, styled the room, trained their own voices, performed their own personalities, and endured their own flop risk. Even the unsuccessful content helped; it supplied data about what failed.

A failed sitcom is a write-off. A failed post is training data.

This is where the business starts to look less like publishing and more like a peculiar labour arrangement nobody negotiated. The platform supplies tools, rules, ranking, and occasional applause. You supply output. Compensation, for most, arrives in a strange blend of attention, status, affiliation, and intermittent possibility. A minority gets paid directly through revenue share, sponsorships, tips, subscriptions, or brand deals. The great majority receive something vaguer and, in aggregate, cheaper: visibility, social standing, the prospect of future visibility, and the variable emotional rewards attached to both.

Variable rewards matter. Behavioural psychologists call this reward prediction error: the gap between the reward you expect and the reward you get. When the outcome is better than expected, the system pays attention; the surprise carries information. Social platforms did not need to read a neuroscience textbook to discover the operational version. A post that gets no reaction is one kind of lesson. A post that unexpectedly takes off is another, and a more powerful one. When approval arrives unevenly—sometimes now, sometimes later, sometimes not at all—it keeps you checking. It is payroll without cash.

Television’s great commercial unit was the rating point. However rough and compromised, it offered a standard measure of audience size that could be sold in advance through the upfront market, defended with reach and frequency, and corrected with make-goods if the audience failed to arrive. Search added a sharper measure: the click, and then the conversion behind it. Social platforms needed a different proxy because they were not managing a fixed schedule or waiting for explicit queries. They were choosing, item by item, from a near-infinite supply of possible things to show you next.

The metric that rose to rule this environment was engagement.

Engagement sounds humanistic in the way “customer success” sounds caring. In practice it is a bucket of measurable actions—likes, comments, shares, replies, clicks, dwell time, watch time, completion rate, saves, follows—that a platform treats as evidence that the thing shown was worth the showing. Worth, here, means several related things at once. Did it keep you present? Did it provoke a response? Did it lead to another piece of content? Did it give the system clearer information about what to send you next? Did it create another opportunity to deliver advertising?

Engagement became the operating proxy because the feed had no common schedule. Every post, every clip, every image, every thread competed continuously against every other. The platform could not rely on a lead-in, a tentpole show, or audience flow in the television sense. It had to manufacture an individualised sequence from chaos.

That proxy spread through the product with remarkable speed. Likes made ranking easier. Comments signalled heat. Shares suggested transportability. On YouTube, raw view counts proved too crude; a video could win the click with a lurid title and lose the audience seconds later. So the platform shifted toward watch time and later audience retention. Suddenly the grammar of YouTube changed. Openings tightened. Narratives stretched. Suspense migrated into places where a sane editor would not previously have put it. “Wait till the end” ceased to be a request and became a business model.

Facebook went through its own metric evolutions, often in public and usually through the side effects. Publishers learned that posts optimized for likes behaved differently from posts optimized for comments, and that posts engineered for shares had a special talent for making everyone worse. A tweak toward “meaningful” interactions sounds wholesome until you discover what kind of content reliably produces people typing at one another. Twitter, a platform whose ad business never fully justified its cultural centrality, nonetheless demonstrated the same principle in miniature: what spreads is shaped by the mechanics of spread. Make reply friction low and you get one kind of discourse. Make repost friction low and you get another. Add visible public metrics and everyone becomes a little more theatrical.

The platform did not need to want anything in particular. It just needed to keep score.

Each engagement signal did two jobs. It expressed a social reaction between people and supplied a machine label to the ranking system. A like was applause and metadata. A comment was conversation and training input. The platform’s genius lay in getting you to perform these interpretive tasks for free and then treating the aggregate as infrastructure.

You can see why feed design became the most consequential UX work in history. A feed is not decoration. It is the allocation mechanism for attention on a system with effectively infinite inventory and finite human time. Tiny interface decisions acquire ridiculous power when billions of people encounter them every day. If autoplay raises watch time, more video gets made in forms suited to autoplay. If stories disappear after twenty-four hours, communication shifts toward the disposable and urgent. If the reshare button sits one thumb movement away, distribution accelerates toward whatever can survive that speed. When TikTok’s For You feed proved that short videos could be distributed according to behavioural response rather than social graph alone, it reorganised not only one app but the strategic plans of every rival.

This is why product managers ended up with influence once reserved for editors, commissioners, and schedulers. A television executive could move money by changing a slot in the schedule. A feed team could do it by changing the weighting on a share, a pause, or a swipe-away. One adjusted inventory around programmes. The other adjusted culture around signals. No other branch of design has given a button shape, a loading delay, or a model parameter such power over what millions of people decide is worth saying.

And because the metric was public enough to be reverse-engineered, the population learned.

One of social media’s funnier and less encouraging truths is that audiences develop format literacy faster than platforms can refresh the format. Banner blindness was the first lesson. People learned where the ad was likely to sit and trained their eyes around it. Social media produced a more advanced version. People learned the cadence of clickbait headlines. Then they learned the visual vocabulary of thumbnail shock-face sincerity. Then the particular grain of sponsored authenticity on Instagram. Then the opening gambits of a TikTok hook. “I didn’t expect to share this…”; “No one talks about…”; “Watch till the end…”; the confessional camera angle; the brand mention placed exactly where spontaneity never is.

Once you can read the pitch, some of the spell breaks. Not all of it—good technique remains good technique—but enough. The population gets savvier. The old trick underperforms. The platform introduces a new affordance. Stories, reels, shorts, lives, duets, stitches, carousels—each one a new door opened just long enough for creators to rush through before the audience learns the floor plan.

This churn is often misdescribed as creativity. Some of it is. Much of it is adaptation to measurement.

Everybody is improvising against the dashboard.

You can see the pressure most clearly in the rise of the creator, a word that did useful ideological work because it fused artist, entertainer, entrepreneur, freelancer, and ad inventory into one flattering noun. Platforms needed this figure. Traditional media had stars, certainly, but stars were expensive and institutions owned much of the surrounding apparatus. The creator economy promised something leaner: distributed programming made by self-motivated individuals operating on spec.

YouTube formalised part of this with the Partner Program, sharing advertising revenue with uploaders and thereby turning more ambitious amateur production into a rational economic gamble. The arrangement looked generous compared with old gatekeepers and was, for a fortunate minority. For the platform it was better than generous; it was capital efficient. Pay out a share only after the audience arrives. Let millions audition themselves on spec. Keep the library. Keep the demand. Keep the ranking system that determines who rises.

Instagram and TikTok pushed the model further by making direct creator-brand deals central. Here the creator ceased to be merely the supplier of attention-generating content and became the ad unit itself. A brand could buy a placement inside the creator’s voice, inside the trust relation, inside the program environment that creator had built. The old display unit sat politely to the side. The sponsored integration moved into the middle of the speech.

This solved a problem for advertisers. Ordinary platform ads often felt like interruptions, and interruptions are now judged against a much harsher standard than they once were. A creator could smuggle the pitch inside something the audience had already elected to receive. The recommendation came in a familiar face, with the tone calibrated to the format, and with just enough self-mockery to flatter the audience’s awareness of being marketed to. The sales line was no longer adjacent to the content. It had taken residence inside it.

For creators, this produced a peculiar status. They were entrepreneurs in the sense that every invoice depended on them. They were employees in the sense that a platform ranking change could wipe out a livelihood by Friday. They were inventory in the sense that agencies bought them in bundles and dropped them when the numbers softened.

The creator economy is a staffing agency that calls itself a stage.

Still, some of it worked. Small businesses that could never have afforded national television found efficient, high-trust ways to reach exactly the people they wanted. Individuals who would once have needed institutions found audiences directly. Whole genres emerged that old commissioners would never have touched and audiences plainly wanted. The system paid out real value. That matters. A serious account of the attention economy has to concede the dividend before examining the rake.

The discomfort enters when you trace the mechanism all the way through. If the platform measures engagement and allocates distribution according to engagement, then creators make for engagement. If creators make for engagement, everybody else learns from what works. Friends pick up the cadence of the platforms they live on. Brands imitate creators; creators imitate one another; ordinary people imitate both.

A family holiday photo arrives with the pose pre-solved by prior platform success. A political opinion is shaped for shareability before it is shaped for truth. A teenager learns which version of her own face produces the right kind of comments. None of this requires conspiracy. It requires only a distribution system that rewards certain forms of legibility more than others.

You have seen smaller versions in yourself. The post that sounded too plain, so you sharpened it. The video you abandoned because the hook felt slow. The photo you liked better privately than publicly. The temptation to phrase a sincere thought in the accent of whatever format is currently clearing.

The system keeps the learnings even when you do not keep the post. Delete it if you like. The lesson has already been taken.

This is where the old story—ads around content—starts to look almost quaint. Social media did not merely give advertisers new places to appear. It built a production system in which attention could be extracted from social life itself, measured continuously, and sold at multiple layers: by the platform, by creators, by brands borrowing creators, by publishers adapting to platform rules, by agencies arbitraging performance across all of them. A person could be audience, contributor, distributor, data source, and target in the same afternoon, while feeling only that they had been online.

The business incentives were too clean for accident to be the right word. If you can get the population to produce the thing that keeps the population occupied, your content costs fall, your measurement improves, your inventory expands, and your product becomes harder to leave because your friends, habits, unfinished conversations, and self-presentation all now live inside it. The platform does not need to own the content in the old studio sense. It needs to own the terms of visibility.

That last point matters more than it first appears. Ownership of distribution used to mean presses, trucks, affiliates, spectrum, shelf space. In the feed era it meant ranking. Whoever controlled ranking controlled commercial outcomes for everyone upstream. Publishers that once negotiated from relative strength found themselves rewiring headlines, layouts, and editorial calendars around Facebook’s priorities, then doing it again when those priorities changed. Video publishers hired whole teams because the platform signalled that video would travel, then watched the traffic disappear when the signal changed. Creators chased one recommendation system after another across platforms, each new feature introduced as opportunity and experienced, about six months later, as compulsion.

The audience became the factory, yes. Factories still have managers.

The population did not merely produce the goods. It learned to supervise itself. Analytics dashboards turned output into self-discipline. Public metrics turned self-discipline into social comparison. Advice industries sprang up to explain the changing rules of visibility, which were then changed again.

The next strategic question was how to keep the factory running all day, everywhere, with shorter intervals between prompt and response. The answer was not a new business model. It was the same one, made portable: a factory that fits in a pocket.

The television and search entries carry forward; social adds a factory floor.

paid out: free publishing, global distribution, new forms of affiliation and entertainment, direct audiences for small businesses and creators, and the intoxicating chance to be seen without permission took in: unpaid content production, behavioural data, social graphs, free labour in tagging and moderation, and advertising sold against attention generated by the audience itself measured: engagement—likes, comments, shares, watch time, completion rate, follows, and every other proxy that could rank a feed and price a person’s visibility left behind: the question of whether a distribution system that rewards legibility over truth, and buys creators the way media once bought slots, can really be separated from the outcomes it predictably produces