Skip to main content

Chapter 8: Who Bought the Mainstream

Part Three: Not Treating the Structure as Your Own Flaw ｜ 📖 Who's Collecting at the Door ｜ 📖 Full Edition

Chapter 8: Who Bought the Mainstream

Prologue: Most Shared, Most Banned​

In 2026 I wrote an article called "The Efficiency Trap." It argued that if the efficiency gains from AI don't create new demand, they end up as structural compression, and that the AI education on the market teaches exactly that compression, not creation.1

That article became at once my most shared and my most banned. Many of the people who shared it came from outside the AI world: real estate agents, embedded systems engineers, coaches who teach primary school children to code. The ones who banned it were several of Taiwan's largest AI groups on Facebook. Some posts waited for approval forever; some were approved when no one would see them any more; some were removed, and my page was blocked along with them.

During the same period, the feeds of these groups never lacked content: cheat sheets, prompt templates, courses on limited-time offer, the administrators' own seminars.

At first I thought this was just a few group administrators' likes and dislikes. Later I realized it was an ecosystem.

1. The Problem Isn't Whether the Courses Exist​

Courses that "harvest leeks," as Chinese internet slang puts it, cropping beginners again and again the way leeks are cut and left to regrow, have always been around. There used to be start-up courses on "becoming rich"; today there are AI courses that run a few hours and cost over ten thousand. In "Every Sentence Is True" I took one of them apart: the work shown was real, the tools were real, every sentence held up, and what students actually bought was only one of the five things behind the showcase.2

Nor is this unique to AI. Chapter 1 said that every round of technology comes with a story (see Chapter 1); every story comes with a crowd of people selling courses. The computerization era had computer classes and crash courses; the internet era had web design classes and e-commerce start-up courses; this round has AI courses. The tools keep changing; what is sold doesn't: the fear of "learn it or be left behind."

What such courses sell is one sentence: learn it or be left behind. Chapter 1 argued that when companies cut their technical staff, they are clearing out the veto, and AI is the name this round gives it (see Chapter 1); Chapter 7 argued that the people around you will describe your situation as your personality (see Chapter 7). The course market takes the next step: it describes structural layoffs as a deficit in your personal skills, then packages that deficit as a product and sells it back to you.

But this chapter isn't about the courses themselves. Whether a course is good or worth the money, buyers can at least judge for themselves; at worst, they lose the fee.

The real problem is the distortion it causes in the whole ecosystem once it appears. Course sellers don't just sell courses; they buy exposure. Once they've bought exposure, they start defining what is mainstream; once they define the mainstream, other voices get pushed where no one can see them. In the end, the question is no longer "have you been harvested?" but "apart from them, who else can you still hear?"

2. Two Queues​

Chapter 4 discussed a restaurant that put high-paying and budget customers into one queue (see Chapter 4). Social platforms didn't start out that way.

The original platforms had two separate queues. One was the ad queue: people who wanted to pay for exposure stood in it, their content was clearly labeled "sponsored," and readers knew at a glance it was commercial promotion. The other was the organic queue: people who wanted to talk freely and speak normally stood in it, and whoever wrote well or made sense got read and shared. They said their piece, we said ours, and neither interfered with the other.

It was much like slotting fees in supermarkets. Brands pay for the best shelf space; everyone knows it and can see it: the row at eye level has been paid for. It may be unfair in places, but at least it is transparent.

What happens now is something else. Leek-harvesting operators use huge ad budgets to take most of the ad queue's slots, which in itself is fair enough. The problem is that they have also pushed into the organic queue, dressed in its clothes:

  • following and liking one another to pump up volume;
  • paying for fake engagement and fake numbers so a post looks like it spread naturally;
  • running their own closed groups, where administrators pin posts, comment first and cross-post insiders' content, laundering exposure behind closed doors.

The result is that you can scroll your feed all day without being able to tell whether what you're looking at is a paid ad or a group of people posing as experts in various fields who, having bought volume, now stand on stage defining the world.

The restaurant in Chapter 4 at least labeled its fast lane as a fast lane. Here the fast lane is dressed as the ordinary queue.

How Slow the Organic Queue Is​

The fast lane sells well because the ordinary queue is slow enough. My own numbers can serve as a reference: in June 2026, each post on my Facebook page averaged around twenty-odd impressions, and about one in ten of the people who saw a post engaged with it; the engagement rate wasn't bad, but reach was close to zero. During the same period, the ad manager's estimate told me that for a post with only eleven organic impressions, NT$100 a day would reach anywhere from zero to three hundred-odd people.

The slower one queue moves, the more valuable the other becomes. Whether this is the platform's deliberate design or a side effect of the algorithm can't be seen from outside; but for the people in the ordinary queue, the result is the same.

The platforms, features and figures in this chapter are as they were in 2026. Platforms will be renamed, features redesigned, numbers changed. What to take away is the two queues: one paid, one organic, and the slower the latter, the more valuable the former. When the next platform appears, look first for where its two queues are.

3. From Buying Exposure to Defining the Mainstream​

Buying exposure is in itself normal business. A course seller who buys plenty of ads recoups the cost as soon as one customer buys an annual plan; that is genuine capital efficiency. The problem is that once you buy exposure in large enough quantity, it is no longer just exposure; it becomes definition.

When you open your feed and most of the content about AI comes from the same circle, what that circle says becomes "the AI community's consensus." What counts as learning AI, who counts as an expert, which questions are worth asking, which attitudes count as positive: all are decided by the loudest group. Voices outside that consensus are not rebutted; they are only categorized: negative energy, unfriendly, amateur, troll.

Turning the Pressure Back​

Once the power to define is in hand, it can be turned to suppress others.

In large groups, suppression needs no formal ban. Administrators don't have to delete your post; they only have to not approve it, approve it late, or pin their own people's content next to it, and the effect is the same. Comments that press on costs or ask for details can be removed as "unfriendly"; accounts reported by a group of people are often restricted first and left to wait on an appeal. Blocking, not blocking, outright muting: the effects are really the same, only different in degree. Suppression is a continuous spectrum, not a series of isolated incidents.

The most hidden form is the so-called shadowban. I have run into this: a post is approved, the administrator even likes it, but other members can't see the post or its comments at all. On your own account everything looks normal; only by switching to another account do you find it has vanished.

What makes the shadowban most effective is that it notifies no one. No notification means no list; no list means no group of victims; no group means no policy pressure. In the three volumes of Attenuation: How Visibility Is Allocated, I set out this mechanism in detail, along with how it reverses the burden of proof and puts it on creators.3

What Was Also Taken: the Original Article​

This had a sequel I didn't expect.

I once posted an article in a large AI group. Later the group blocked my page. Looking from another account, the post could no longer be found in the group; but when I searched the article's title on Google, the top result was the group's copy, ranked above the original on my own website. Once my page was blocked, I no longer even had permission to delete that post.

Meta's terms of service state that the license users grant the platform over their content ends when the content is deleted from its systems; the terms also say users can delete their content at any time.4 But when the platform's own blocking takes away the power to delete, the license never ends. The people who can't see me can see my article; only it hangs under someone else's name, drawing traffic for someone else's group.

4. Fake People Get Blue Checks Too​

Verification is even more absurd.

In February 2023, Meta launched paid blue-check verification, Meta Verified. According to the official description, subscribers submit government ID matching their profile name and photo; what they get for subscribing includes not only the verification badge and impersonation protection, but also "increased visibility."5

In other words, the blue check had two meanings from the start: one is "there is a real person here"; the other is "this account has paid, so it will be shown a bit more."

The first meaning is failing. In 2026 I saw with my own eyes an AI-generated virtual person's account wearing a blue check. Someone underneath asked: isn't the blue check something you buy? The account's owner answered matter-of-factly: yes, you have to buy it. What the ID check actually stops in practice, I don't know; I only know that a person who doesn't exist can appear in your feed carrying "real person verification." I have also seen instructors selling anxiety courses advertise with AI-generated characters, buy blue checks for those characters, and then run them as ads.

What remains is only the second meaning. If even fake people can be verified as "real," the blue check no longer verifies what is real; it verifies whether you paid. So turn the question around: does a real person who won't pay for verification never get basic exposure?

This is the greatest irony of the whole business. Creators work hard writing articles that bring the platform content, readers and time spent; that traffic isn't rewarded but pressed down by the algorithm, then sold back to them monthly under the name of "increased visibility."

5. Symbiosis​

So "loopholes in social platforms' mechanics" and "the spread of leek-harvesting courses" are not two problems but two halves of one vicious circle.

Walk through it step by step:

  1. The platform suppresses organic reach, so the fast lane can be sold.
  2. The businesses most willing to pay for the fast lane are those with the highest margins and fastest conversion, and leek-harvesting courses are among them. Course sellers do the math: as soon as one student buys an annual plan, the ad spend is recouped.
  3. Having bought enough exposure, course sellers define the mainstream; those outside it who press on costs or raise objections are categorized, reported and removed.
  4. With fewer voices warning buyers, the next wave of buyers grows; with more buyers, course sellers have more money for the next round of ads.
  5. The platform takes in more ad revenue and has even less reason to let the organic queue move faster. Back to step one.

I have seen a sales-coach page with a blue check and tens of thousands of followers whose post stated outright: you have to pay for your lead lists. That line is a slogan course sellers teach their students, and it is also the platform's own business model: your customers can only be found if you pay.

The circle has one more lock. Facebook groups and pages are often recommended to newcomers through "mutual friends." If the handful of friends you first added were already in these groups, every later spread of yours is locked inside the same circle; however it is passed around, it reaches the same people. On this platform there is almost no road left to readers outside the circle.

As with Chapter 1's loop, each step of this circle is a rational response to the step before (see Chapter 1). The platform wants revenue, course sellers want customers, administrators want to keep their groups "positive." No step needs anyone to mean harm; but once the loop is complete, the only side that can be heard is the side that paid.

6. An Industry Chain Spinning in Place​

If this circle stayed on social platforms, the losses would be at most some readers' time and some students' fees. The most alarming step is when the "professional definitions" built up by capital are packaged one layer higher, into government-endorsed projects and company-backed training courses.

The Transfer of Legitimacy​

The route is no mystery. The loudest people on social media are the easiest to invite: workshops co-organized by government, speaker lists for corporate training, adjunct appointments at colleges, partner titles at vendor events, officially recognized skills certifications. Each layer of endorsement carries over to the next course, so that "took a few courses, plus self-study" gradually becomes a credential. What students see is someone endorsed layer upon layer; what they can't see is that none of these endorsements ever measured the ability to deliver, only volume.

Subsidies Are the Amplifier​

Under Taiwan's Ministry of Labor's "Industrial Talent Investment Program," as of 2026, employed workers are subsidized for 80% or 100% of training fees, up to NT$100,000 per person every three years, for courses run by private training providers.6 The policy's intention, to encourage continuing education, is not wrong; but once it connects with the circle above, the effect changes.

In "The Efficiency Trap: Sequel" I wrote that subsidies are the amplifier of the whole model: some groups' gatherings are nominally paid, but participants get in free as long as they hand over their personal details so the organizer can claim the subsidy.7 Subsidy programs measure course hours, numbers of graduates and certificates issued, not whether students can make anything once they finish. This is exactly the measurement problem discussed in Chapter 6: what gets measured is what is easy to count (see Chapter 6). A few hours to teach a set of tools produces precisely hours and certificates of completion. What the system purchases and what such courses sell line up perfectly.

Same Name, Different Contents​

Put the whole chain together: students' fees and government subsidies flow to the courses, course profits flow to ads, ad spending flows to the platform, and the platform sells exposure back to the courses. The money keeps moving, and every cell in the reports has a number, hours, headcounts, certificates, reach, nothing missing. Only, after one full loop, nothing has been made.

Chapter 1 said an R&D department can keep its name while its contents are swapped: the headcount is still there, the spending is still there, but what goes on inside is no longer R&D (see Chapter 1). This chain is the societal version of the same thing. The words "AI talent development" are still there, the budget is still there, and completion certificates increase every year; but what flows inside is ad spending and subsidy money.

This is a completely inefficient industry chain spinning in place. What it consumes is not only individuals' fees but a whole society's energy: working people take classes after work, unemployed people spend subsidies on classes, governments spend budgets buying hours, companies spend training budgets buying titles. That time and those resources could have gone into genuinely learning something and making something. I believe this does deeper damage to a society's long-term competitiveness than any single bad course.

7. Invisible Is Not the Same as Nonexistent​

This chapter belongs in the part titled "not treating the structure as your own flaw" for a simple reason: the biggest victims of this circle often don't feel like victims; they only feel they aren't good enough.

Someone who writes seriously, with each article seen by only twenty-odd people, can easily conclude: my writing is bad, nobody cares about my topics. Someone who wants to learn AI and finds only one kind of voice in their feed can easily conclude: this is the industry consensus, and if I can't keep up, that's my problem. Both conclusions mistake a designed visibility for their own ability.

This is the positional sampling of Chapter 4 (see Chapter 4). Standing in the suppressed queue, you think the world is this small; seeing only the group that paid, you think they represent everyone.

How to Spot Bought Volume​

From outside, bought volume and genuine expertise often look alike. But a few questions can be asked without any inside information:

  1. Is this content labeled as an ad? If not, are the groups, pages and seminars behind it the same business?
  2. In their community, do comments that press on costs or raise objections survive? If they stay and get serious answers, that is one answer; if they don't stay, that is another.
  3. Are the people endorsing them also endorsing each other? When people in one circle take turns recommending one another, the recommendation is no longer evidence.
  4. An account that claims influence and follows thousands of people: is it following to read, or to be followed back?
  5. Does what they say have results that can be tested? Or only certificates, titles and student testimonials?

Summary​

Looking back over this chapter, the leek-harvesting courses are only the entrance. The real problem is that they have grown together with the platform's mechanics: the platform slows the organic queue, course sellers buy the fast lane and then dress it as the organic queue; bought exposure becomes the definition of the mainstream, and that definition is packaged again as government and corporate endorsement, finally flowing back into the same chain. The money keeps moving, while the voices that can be heard grow fewer.

The public's precious energy and resources should not be used up by fake tasks, bad courses and empty slogans wrapped in money. And those pressed out of sight by this chain should not mistake being invisible for a flaw of their own.

Chapter 7 looked at how the people around you describe your situation as personality; this chapter looked at how markets and platforms describe visibility as ability. The next chapter leaves the screen, crosses a bridge from Linkou and walks to a beef noodle shop in Gongxi, Guishan. The old woman who ran it was never on any platform, yet she and her children's generation are judged by the same story: those who hold on are called resilient; those who don't are called not hardworking enough.

← Chapter 7: Psychological Manhunt: The Answer Was Given Three Times, and No One Heard It📖 Back to ContentsChapter 9: One Bridge Apart →

Footnotes​

  1. Nebula Walker (星忘塵), "The Efficiency Trap: When Everyone Is Learning AI, Who Is Creating New Demand?" (〈效率陷阱：當所有人都在學AI，誰在創造新需求？〉), Mythogen Engine [in Chinese], June 27, 2026. ↩

  2. Nebula Walker (星忘塵), "Every Sentence Is True: The Cost Ledger of an AI Video Course" (〈每一句都是真話：一堂 AI 影片課的成本帳〉), Mythogen Engine [in Chinese], September 24, 2026. ↩

  3. Nebula Walker (星忘塵), Attenuation: How Visibility Is Allocated (《衰減：可見度如何被分配》), three volumes, Mythogen Engine, August 2026 [in Chinese]: vol. 1; vol. 2; vol. 3. ↩

  4. Meta, Terms of Service, Section 3 (the license users grant over their content ends when the content is deleted from Meta's systems). [Version current at publication and link to be checked] ↩

  5. iThome, "Meta Launches Paid Subscription Service Meta Verified," February 21, 2023 [in Chinese]. ↩

  6. Ministry of Labor (Taiwan), "I Want to Take Vocational Training" (training for employed workers: Industrial Talent Investment Program) [in Chinese]. ↩

  7. Nebula Walker (星忘塵), "The Efficiency Trap: Sequel — You Think You Won" (〈效率陷阱・續章：你以為你贏了〉), Mythogen Engine [in Chinese], July 28, 2026. ↩