The Collapse: The Influencer Economy Automates Itself

The Business of Being Believed

I. The Aitana Problem

Aitana López has pink hair, eleven million impressions a month, and no body. She posts fitness content, fashion hauls, opinions about her day, from an apartment in Barcelona that does not exist, run by an eleven-person team at an agency called The Clueless. She earns the agency somewhere around €10,000 a month in brand deals. She has never been tired, never cancelled a shoot, never posted something at 2am that a brand’s legal team had to spend the next morning apologising for.

Her creator built her for one reason, and he said it plainly: real influencers were unreliable, and they were costing his firm money. Not a technological pitch about the future of content. A cost line, and a scheduling problem, solved by removing the person. That sentence is the entire thesis of what follows, and everything after it is working out what it means that the sentence turned out to be true.

This is not a story about deepfakes, and it is not a story about whether audiences can tell the difference. It is a story about what happens when a business figures out that the thing it was paying for — attention, trust, a face people feel they know — never actually required a person attached to it, and the person was simply the only available method of producing it until now. Aitana is not a trick played on the public. She is the plain version of a business model the public had already agreed to believe in.

II. From Wedgwood to the Feed

Somebody has always been paid to lend a face to a product, and the method for finding that face has changed roughly every fifty years since it began. Josiah Wedgwood is usually credited as the starting point: in the 1760s he supplied pottery to Queen Charlotte, then advertised the fact relentlessly, understanding before almost anyone that royal association could move volume no amount of craftsmanship could move alone. The eighteenth century didn’t need to be told that trust could be borrowed. It needed someone to prove the borrowing worked.

The nineteenth century industrialised the same instinct, because mass production had created a problem trust used to solve for free. When goods were made locally, by people you knew, quality was something you could inspect. Once goods were made in factories and shipped by rail, the buyer had no way to check anything before paying for it — and testimonial advertising arrived to fill the gap, a stranger’s word standing in for the inspection nobody could perform anymore.

The twentieth century professionalised it into the celebrity endorsement, and made it extravagantly expensive. Michael Jordan and Nike, Coca-Cola’s invention of the modern Santa Claus, a system in which a famous face cost a fortune and a company simply had to trust that the fortune was buying something real. Nobody could measure it with any precision. That was, for decades, the whole business model: pay enormous sums for borrowed trust, and don’t ask too many questions about the return.

Then the internet did to celebrity what the factory had done to the village grocer — it broke the monopoly on being known. Blogging in the early 2000s, YouTube from 2006, and alongside YouTube the arrival of multi-channel networks like Maker Studios and Next New Networks, businesses that functioned as talent agencies for people whose talent was simply that a camera liked them. YouTube bought Next New Networks outright in 2011, an early admission that whatever this was, it was worth owning rather than merely hosting. Instagram, from 2010, gave the whole apparatus a visual grammar — the curated feed, the lifestyle shot, the sense of a life being lived slightly better than yours — and by around 2015 “influencer” was something you could write on a Schedule C, filed in the same box as a market trader or a jobbing plumber, whether or not you were declaring it.

Notice what happened to who held the leverage at each stage of this history. When Wedgwood needed the Queen, the Queen held the leverage. When the mass-market advertiser needed a testimonial, the publisher who controlled the printing press held it. When the celebrity endorsement was king, the celebrity and their agent held it. And when the platforms handed ordinary people the tools to build an audience directly — a camera, a feed, a following nobody else owned — for the first time in this two-hundred-and-sixty-year history of marketing, the person being watched held real leverage over the person paying for the watching.

That arrangement lasted about fifteen years. Aitana is what it looks like when it ends.

III. Four Orders of the Sign

Jean Baudrillard, writing in 1981, described four stages a sign passes through on its way to losing any connection to a real thing at all, and the influencer economy has now produced a textbook case of the last one. The first stage is a faithful copy — a portrait that resembles the person sitting for it, an image that reflects a basic reality. The second is a copy that distorts the reality it represents — flattering, exaggerating, doing PR for the thing it depicts rather than describing it. The third pretends to be a faithful copy while having no reality behind it at all — the counterfeit that works precisely because it convinces you the real thing is still somewhere underneath it. The fourth stage has no relationship to reality whatsoever. It is, in Baudrillard’s own phrase, a simulacrum: not a copy of anything, a sign that generates and sustains itself.

A human influencer, however curated, operates somewhere between the second and third stages, and that distinction matters more than it looks. The lifestyle is real, even if the version shown is flattering past the point of accuracy — a stage-two operation, PR for an actual life. Or the persona has drifted so far from the person that maintaining it has become a performance the influencer gives even to themselves — stage three, the counterfeit sustaining itself on the audience’s continued belief there’s a real person underneath. Either way, there is, somewhere, a referent. A body that gets tired. A life the content is at least nominally about.

Aitana has no referent, and never has, and this is the precise sense in which she is a purer case than anything the theory previously had to work with. There was no real Aitana who got glamorised into a marketing version of herself. There was a decision, a rendering pipeline, and a business plan. She is the sign without the thing it supposedly signifies, generated whole, and — this is the detail that would have interested Baudrillard specifically — the fact of her unreality has not weakened her performance in the market. When The Clueless publicly confirmed she was AI-generated, her engagement did not collapse. It rose. The audience did not need to be fooled. It needed, apparently, only to be interested, and unreality turned out to be a source of interest rather than an obstacle to it.

Which suggests the smart French and German theorists of the 1970s were not describing a distant, abstract danger — they were describing a coming line of business, and business has since obliged them by building it to specification. Baudrillard argued that in a media-saturated society, the distinction between the real and its representation would eventually stop mattering to anyone, not because people would be deceived, but because the distinction would simply stop doing any useful work. Fifty years later there is a Barcelona apartment with no rooms in it, generating a return on investment precisely because nobody involved — not the agency, not the brands, arguably not even the followers — is especially bothered about whether it’s real.

IV. The Spectacle Was Already There

Baudrillard tells you what Aitana is. Guy Debord, writing two decades earlier, tells you what she is for. His argument in 1967 was that under advanced capitalism, social life itself becomes mediated by images rather than lived directly — that the relationships people believe they are having are increasingly relationships with representations, and that this is not a distortion of social life but has become its actual, dominant form. Where Baudrillard is diagnosing a sign that has come loose from any reality behind it, Debord is diagnosing a relationship that has come loose from any person behind it. Aitana is a fourth-order simulacrum in Baudrillard’s sense and a spectacle in Debord’s, and the two claims are doing different jobs: one tells you the image is not real, the other tells you the follower’s sense of connection to it is being sold as though it were.

Debord also gives us a mechanism Baudrillard doesn’t: recuperation, the spectacle’s ability to absorb its own criticism and sell it back as more of itself. When it emerged that Aitana was AI-generated, the obvious expectation is that the revelation would function as critique — proof of the con, grounds for the audience to withdraw. Instead, and this is the same rise in engagement already noted a moment ago, now examined for what it actually demonstrates rather than merely what it disproves: the unmasking did not damage the spectacle, it became content inside it, generating exactly the kind of attention the account exists to convert into brand deals. This is recuperation with a date stamped on it rather than a general theoretical claim: the criticism of the fakery was not merely ignored, it was the single best-performing piece of content the account had produced.

This is the point at which the two theorists earn their place side by side rather than functioning as a decorative double act. Say only that Aitana is a sign with no referent, and the reader learns something true but slightly inert — an observation about a category of image. Say that she is also a spectacle in Debord’s sense, and the reader gets the part that actually explains the money: the “relationship” a follower feels with her is the product being sold to brands, not a side effect of it, and the fact that no person exists to have a relationship with turns out to be irrelevant to whether the relationship can be monetised. Baudrillard explains why nobody needed to be deceived. Debord explains why nobody needed a person there at all — because the thing capital was actually buying was never the influencer. It was the feeling of a relationship, and that feeling has just been shown to survive the total absence of the person it was supposedly with.

Which is, if you think about it for slightly longer than the algorithm wants you to, the part that should have been obvious about the whole industry from the start. Nobody following a stranger’s skincare routine believed they had a friendship in any ordinary sense. But they were behaving, economically, exactly as if they did — responding to recommendations the way you’d respond to a friend’s, extending trust an advertisement alone could never earn. Debord’s argument was never that people are fooled into thinking the spectacle is real. It’s that they act on it as though it were, which is a different and more useful thing to be right about, and Aitana simply proves the point can be pushed all the way to zero and the behaviour doesn’t change.

V. Who Actually Gets Paid

V.1 Sizing

Global advertising spend runs to roughly $1.06 trillion in 2026. Digital advertising accounts for somewhere between $740 and $836 billion of that. Influencer marketing, on the more generous estimates going around, sits at $32 to $48 billion — call it 3 to 4% of everything spent on advertising worldwide, or 4 to 6% of the digital slice specifically. Whichever end of the range you take, the influencer economy is a footnote next to the industry it sits inside.

The number that matters is not the size, though. It’s the gap between the size and the growth rate. Influencer marketing spend has been growing at somewhere north of 30% a year, against low single digits for advertising generally. A category that’s 4% of digital spend and compounding at ten times the rate of the market it belongs to does not stay 4% for long. This is the shape of every disruptive slice of an economy in its early years — small enough to be dismissed, fast enough that the dismissal has a shelf life. Streaming looked like a rounding error against broadcast television in 2010.

And the range itself — $32 billion at one end, $48 billion at the other, a spread of 50% on the low estimate — nobody actually agrees on the number. Different research houses count different things as “influencer marketing spend” — some include only paid brand-creator deals, others fold in platform-native ad products bought by creators themselves, others count agency fees as part of the spend rather than a cost sitting on top of it. A market this large, this fast-growing, and this economically consequential to the people working inside it, and the basic question of how big it actually is comes with a spread the size of a mid-cap company’s entire annual revenue. A problem not uncommon in fast growth markets, as everyone catches up to reality.

V.2 Who’s Actually Working

The top 10% of creators capture somewhere between 60 and 80% of all the money paid out to creators, and that share has been rising, not falling. Most of what looks, from outside, like a broad and thriving economy of ordinary people building an audience is in practice a small number of people doing very well, sitting on top of a very long tail making next to nothing. Median earnings for creators generally have been drifting down even as the total market grows — an unusual combination, and the clearest sign yet of a winner-take-all structure rather than a rising tide.

Nobody can agree how many people are even in this workforce, and the disagreement is not trivial, it’s an order of magnitude. Estimates for the total number of “content creators” worldwide range from around 50 million up past 300 million, depending entirely on how loosely the term gets used. Narrow it to people who would specifically call themselves an influencer and the range drops to somewhere between 50 and 127 million. Narrow it again to people earning a full-time living from it, and the number falls to something like 2 to 4 million.

Which means somewhere between 95 and 99% of the people who consider themselves part of this workforce are not, in any economic sense, employed by it. They are performing the activity of a job — a posting schedule, a media kit, a brand-outreach folder — without the underlying relationship that would make it one. That is not a story about individual failure or bad luck. It is a category doing something categories aren’t supposed to do: describing an aspiration and an activity for the overwhelming majority who hold it, while describing a genuine occupation for a small minority at the top. A sign of employment, circulating freely among people it was never actually attached to — which is a fair description of what we previously called a simulacrum, aimed this time not at a product, but at the workforce making it.

V.3 The Toll Booth

TikTok’s original Creator Fund, running from 2020, paid $0.02 to $0.04 per thousand views — a video with a million views, the kind of number that feels like real success to whoever made it, earned somewhere between $20 and $40. Its replacement, Creator Rewards, launched in 2024, pays better: $0.40 to $1.00 per thousand qualified views, putting that same million views at $400 to $1,000. An improvement of ten to twenty-five times, but still built on a definition of “qualified” that filters out roughly half of a creator’s raw view count before the payout formula ever sees it. Instagram, the platform where influencer culture is most visibly performed, publishes no comparable rate at all — its creator bonus programmes are, in the industry’s own description, largely inactive, and the platform pays almost nothing directly to the people whose content fills it.

Platform payments of this kind now supply only 10 to 20% of a working creator’s total income. Brand sponsorships supply the rest. The platforms have arranged things so they are structurally indispensable — nothing reaches an audience without their feed and their algorithm — while remaining almost irrelevant to the actual payment for the content they host. That is the toll booth in its purest form: the toll itself is nearly zero, because the platform’s business was never paying creators. It was renting out their audience to brands, and leaving creators to negotiate whatever they could get directly.

And the platform is now being paid a second time for the same content, in a transaction the creator isn’t party to at all. Every time an ad appears next to a piece of content, that moment of the viewer’s attention has already been sold forward to an advertiser, through an instant, automated auction that happens before the page even finishes loading. The trust a follower has built up in an influencer simply makes that moment worth more in the auction — a better prospect for whichever brand wins it.

And platforms are now also licensing the accumulated content itself, retrospectively, as training data for the AI systems being built to do this kind of work more cheaply. Reddit’s deal with Google is reportedly worth around $60 million a year. X changed its terms of service to explicitly permit using posted content to train its own AI. Research into deals like these found that only a small minority pay anything at all to the people who actually created the material being sold. The platform gets paid once, in the moment, for the audience’s attention. It gets paid again, later, for the content that built that audience in the first place. The creator, in both sales, is the asset — never the seller.

V.4 The Middlemen Are Being Cut Out Too

WPP was, for years, the largest advertising company in the world. It is now dismantling itself. Revenue fell around 8% year-on-year through 2025, with the decline accelerating toward the end of the year, and the business has since announced a plan — internally called Elevate28 — to break up the decades-old holding-company structure entirely. Omnicom, a rival roughly its size, responded by buying a third major competitor, Interpublic Group, for $13.5 billion, specifically to survive at the new scale the market now demands, and is cutting a billion dollars of jobs and overhead in the process. Only Publicis, among the major players, is actually growing — and its own explanation is that it bet early on AI tools rather than headcount.

The stated reason for all of this is bigger than influencers alone, but influencers are a real part of it. Google, Meta, and Amazon between them now take somewhere around 70 to 80% of all digital ad spending directly, with no agency involved in the transaction at all. That’s the primary wound. But the agencies’ other traditional job — being the trusted middleman who finds and manages talent on a brand’s behalf — is being cut out from the other direction too, as more brands deal with individual creators directly, or through small specialist shops built for exactly that relationship, rather than paying a legacy agency’s overhead to broker it.

Which leaves the old agency squeezed from both ends of a transaction it used to sit in the middle of. Upstream, the platforms take the ad money without needing an agency to place it. Downstream, the brand can now reach the creator’s audience without an agency to arrange the introduction. The agency’s entire value proposition was standing between two parties who couldn’t easily find each other. Once they can, there’s very little left for the middleman to do except restructure, merge, or bet everything on a piece of software that can do the one job that’s still left.

V.5 The Wedge

The virtual influencer market is valued at around $11.7 billion in 2026, and it’s growing at 41 to 45% a year — faster than the human influencer market it sits alongside. The $11.7bn doesn’t cleanly map onto the $32–48bn total influencer marketing figure we have already identified — different research houses, different methodology, no published number anywhere for what share of brand spend is actually flowing to synthetic rather than human creators. An industry that publishes engagement rates to two decimal places has, so far, shown no interest in publishing the one figure that would show how fast it’s replacing the people it profiles as its talent.

What is measured cleanly, and is startling on its own terms: AI-generated influencer content gets a 5.67% engagement rate, against roughly 1.89% for human creators — nearly three times the response, at a fraction of the cost and none of the risk. Another virtual influencer, , reportedly generated over $11 million in 2023 across brand deals and licensing, run by a company that was itself acquired by a crypto firm without ever touching a public market. Aitana earns her agency around €10,000 a month, on the strength of an eleven-person human team standing behind a face that doesn’t exist.

And in January this year, the wedge went somewhere new: a Nasdaq-listed shell company called Rich Sparkle completed a $975 million all-stock deal to acquire the rights to Khaby Lame’s name, likeness, and content — including a planned AI “digital twin” built to scale his output using his own face, voice, and gestures. Lame is Senegalese-born, Italian by naturalisation, and famous for saying almost nothing — a persona built entirely on the appearance of not performing, silently reacting to other people’s overcomplicated videos with a raised eyebrow and open palms, which is precisely the “authenticity” the deal is now paying nearly a billion dollars to scale artificially. Not a synthetic influencer being financialised — a real, famous human being converted into a tradeable public-markets instrument, with the AI clone doing the actual scaling once the deal closes. Nobody’s studio needed to build a fake Aitana from scratch. They found a real Khaby Lame, decided his most marketable asset was the appearance of not performing, and are now selling shares in a manufactured version of him doing exactly that at scale.

Two Frenchmen would be ecstatic at this.

V.6 A Market Nobody Can Measure, Chasing a Market That May Not Survive Being Measured

Every number in this section has come with a range instead of a figure — 50 to 300 million creators, $32 to $48 billion in spend, 10 to 20% of it reaching the platform — because market sizing is always difficult and contentious.That difficulty is not unique to influencers. What’s unique here is the second failure sitting on top of the first: an entire category may be replaced by its synthetic version before the argument over how to measure the original is ever settled.

A market being superseded faster than it can be measured is a stranger thing than a market that’s merely hard to measure. Every market pricing exercise assumes the thing being priced holds still long enough to be counted accurately — a barrel of oil, an hour of labour, a follower’s trust. Here it might not. The commodity in question is trust extended to a face, and the count is coming apart in real time, precisely because nobody can agree whether the face needs to be real.

VI. Why You Believe Her Anyway

Trust extended to something you have never met and never will meet is not a glitch in this system. It is the oldest trick money itself ever ran. A banknote is a promise from a state you’ve never dealt with directly. A bank balance is a number representing money that mostly doesn’t exist anywhere as a physical thing. Every transaction beyond the most direct barter asks you to extend trust to an abstraction standing in for something real, and it has done so for as long as money has existed. Aitana is not a departure from that logic. She is the latest, pure expression of it — trust extended to something that was never a someone at all, one more layer of abstraction stacked on a habit humanity has been practising for centuries.

Parasocial relationships — the one-sided bond a person forms with a media figure who has no idea they exist — were first named by researchers in 1956, decades before social media. And the mechanism they described turns out to explain a great deal of what a brand is actually acquiring. A 2025 study found that followers who felt deeply invested in a creator were 71% more likely to act on that creator’s recommendation than those with a weaker bond. Nielsen puts Gen Z’s trust in influencer recommendations above their trust in traditional celebrity endorsement — and the reason usually given is our old friend, authenticity, the sense that an influencer is a real person sharing a real opinion rather than a paid face reading a script. That the whole arrangement runs on manufactured intimacy makes the reasoning almost perfectly circular. None of this requires the follower to believe, consciously, that a friendship exists. It only requires them to act as though one does — extending the kind of trust normally reserved for someone who knows your name, to someone who has never had a name to know.

So we can say it again: when Aitana’s creators confirmed she was AI, her engagement rose rather than collapsed. Believing the relationship is real was never the requirement. Acting on it was. That’s the uncomfortable width of the mechanism — it survives full disclosure, survives the follower knowing better, survives even the target of the trust turning out not to exist at all.

So: if knowing doesn’t disable the mechanism, what exactly does the reader think protects them from it?

VII. Naming the Collapse

Every economy runs on a simple circuit: production generates income, income gets spent, and that spending funds the next round of production. Break the circuit anywhere and it breaks everywhere else, because the worker who gets replaced by a cheaper alternative doesn’t just lose a job — they stop being a customer for whatever that cheaper alternative was supposed to be selling. Capital substituting for labour looks, from inside a single company’s accounts, like pure efficiency. Looked at from the level of the whole economy, it’s removing a consumer from the demand side at the exact moment it’s expanding supply on the other.

For roughly fifteen years, individual creators held real leverage over the businesses that needed their audience — the first time in two hundred and sixty years that the borrowed trust had run in that direction. What’s happened since is capital finding a route back to owning the audience-relationship itself, without needing a human being to hold onto any of the leverage that came with being the one people actually trusted. We might call this the Collapse — not a single crash, but the slow closing of a brief window in which labour, for once, held a genuine card, followed by capital quietly working out how to stop needing to deal it.

And this is not a forecast. Every piece of it has already happened. The agencies that used to broker trust for a living are merging or dismantling themselves in real time. A studio can build a face that outperforms a human one on engagement, at a fraction of the cost, with none of the risk. A real, famous person has already been converted into a tradeable security, with an AI clone doing the scaling that used to require paying him.

And the reader following along, whoever they are, is not outside this circuit. They are standing on the demand side of it — the attention being fought over is theirs, whether or not they’ve ever knowingly followed anyone at all.

VIII. The Invitation

You may not be the audience we have so far been describing. You didn’t grow up watching a stranger’s skincare routine, you’ve never felt anything for Aitana or Lil Miquela, and Khaby Lame’s silent raised eyebrow has likely never once changed what you bought.

But this mechanism was never limited to people who knowingly follow someone. It runs equally well through recommendation feeds you never asked for, through retargeted ads that learned what you paused on, through the ordinary drift of a browsing session you’d never describe as “being influenced” because nobody’s face was attached to it. We have already shown that the mechanism survives the follower knowing the relationship is manufactured. It does not require you to know a relationship exists at all.

So don’t ask whether or not you follow an influencer. Ask yourself whether you can actually locate the edge of the audience you’re being told you’re not part of. Somewhere on your phone, right now, something has already worked out what holds your attention better than you have — and it was trained on exactly the kind of data we have been describing. Platforms selling forward, twice over, to whoever was buying.

Methodology note: this piece was written collaboratively between a human and an AI — the human providing the instincts, provocations, editorial judgement and voice; the AI providing research synthesis, intellectual scaffolding and drafting. Full method at [athomehefeelslikeatourist.blog].