The Social Network Without Society

I keep seeing posts on X explaining how to make money from “AI influencers”.

One recent example argued that everyone trying to build followers had misunderstood the opportunity. Create an attractive AI character, choose a niche such as skincare or fitness, reproduce the UGC formats already working there and sell the adverts directly to brands. Human creators are slow and limited; your character can work twenty-four hours a day.

Higgsfield, whose AI Influencer tool inspired much of this latest gold rush, sells essentially that capability: create a consistent character and produce as much short-form content around them as you need.

As a way of producing cheap advertising, I completely understand the logic.

What I don’t understand is why we are calling the character an influencer.

If nobody follows them, who are they influencing?

When authenticity becomes a format

I had always assumed the value of influencer marketing was the relationship between a person and their audience.

Traditional advertising is a company telling you its product is good. Influencer marketing works because the recommendation comes from somebody you already chose to follow. You might value their expertise, admire their taste or simply feel familiar with them after watching them for years. That relationship is what the advertiser gets access to.

There have always been obvious compromises. Once somebody is being paid to recommend something, the recommendation becomes less authentic. Some influencers promote so many products that their enthusiasm stops meaning much. Others probably know less about the thing they are selling than the advertising team briefing them.

But there is at least a person with a reputation. If they spend years recommending rubbish, that should eventually make their recommendations less valuable.

The new AI influencer pitch removes all of that. There may be no audience, no reputation and no experience of using the product. The character cannot genuinely love the moisturiser because the character does not exist.

What remains is the appearance of influencer marketing: somebody looking into a phone camera from a bathroom or bedroom, holding a product and explaining why they love it.

This has already been happening with UGC. User-generated content once meant content that users actually generated. Brands discovered that the format worked, so they began commissioning people to reproduce it. The creator no longer had to be a customer who happened to make a video; they could be hired to make an advert that looked like one.

AI removes the next piece. If what matters is not who made the recommendation but what the recommendation looks like, why hire a person at all?

For years the advertising industry has talked endlessly about authenticity. AI influencers raise the uncomfortable possibility that some of what it actually learned was simply how authenticity looks.

When my social network stopped being social

I stopped using Facebook and Instagram towards the end of 2020.

The timing should have made that unlikely. We had spent much of the year physically separated from one another, so services built around seeing what friends were doing ought to have felt more valuable than ever.

For me, the opposite happened.

There seemed to be less and less of my friends in the feed. Every few posts was sponsored. Instagram was pushing harder towards video, much of it the same viral material I had already seen a week earlier on Reddit or Twitter. I would open an app supposedly built around people I knew and find myself watching content made by strangers because an algorithm thought it might keep me there.

At a moment when I could barely leave the house, the platforms seemed remarkably uninterested in helping me feel closer to people. They were extremely good at filling the hours.

That was when the bargain stopped making sense to me. The social network had been useful because the people I cared about were on it. Once those people became interruptions between adverts and recommended content, I could no longer see much reason to be there.

Lockdown also created an extreme version of a direction technology was already moving in. Work happened through a screen. Entertainment arrived through a screen. Shopping came to the door. Friends became messages, video calls and feeds. For long stretches, an extraordinary amount of life could happen without leaving one room.

I don’t think Meta or any other technology company wants people living under permanent lockdown. It doesn’t require that kind of conspiracy. The incentives are simpler. A company that earns more when you spend more time looking at its product benefits whenever another part of life can happen through that product.

The uncomfortable question is what happens when thousands of companies, each following perfectly ordinary commercial incentives, collectively make a physically smaller life increasingly convenient.

You don’t need to go shopping if the shop comes to you. You don’t need to find something to watch if an infinite feed chooses for you. You don’t even need your friends to have posted anything if the platform has millions of strangers ready to fill the space.

Increasingly, those strangers do not need to be real either.

The Narrative Factory

Last year I wrote about these platforms as a kind of Narrative Factory. My argument was that they do more than distribute stories. Their algorithms decide which stories spread, usually according to what generates attention and advertising revenue, and people gradually learn what the machinery rewards.

More recently, in Clip Farming the Real World, I was trying to understand what happens when those incentives leak back into physical life. Streamers go somewhere because it might create content. A stranger becomes a possible interaction. A football match becomes the background to somebody else’s broadcast. Reality starts adapting itself to the needs of the feed.

AI offers a shortcut.

You no longer need to encourage a person to become better suited to the platform. You can design somebody for it from the beginning. If one face attracts more attention, use that face. If another personality performs better, change the personality. If a different room sells more moisturiser, put them there tomorrow.

The Narrative Factory can now manufacture the narrator.

That makes our obsession with engagement metrics look increasingly strange. Views, followers, clicks and likes became valuable because they were proxies for harder things to measure: attention, interest, trust and eventually behaviour.

Then we started optimising the proxies.

Who pays for the internet?

There is an economic reason this happened.

The internet I grew up with trained me to expect almost everything to be free at the point of use. I paid an internet provider for the connection, but I didn’t pay Google every time I searched, Facebook every time I looked at a photograph or YouTube every time I watched a video.

Advertising paid instead.

For all my criticism of that model, it enabled something remarkable. Huge amounts of information, entertainment and communication became available without asking us to make a financial decision every few minutes.

The bargain required advertisers to know what their money was buying, so the web became extraordinarily good at producing measurements: impressions, reach, clicks, views, engagement and conversions. Those measurements then became targets around which products and behaviour were redesigned.

The precision of the dashboards sometimes hid how opaque the machinery underneath them could be. In 2025 a US court found that Google had unlawfully monopolised important parts of the technology used to buy and sell advertising across the open web. The remedies ordered in September 2026 require greater interoperability and data access while restricting Google from favouring its own advertising tools.

Meta has faced a different case from advertisers over its “Potential Reach” metric. The advertisers allege that a figure presented as the number of people an advert might reach was based on accounts, including duplicates and fake accounts. The allegations have not been decided at trial, but years of litigation over what people meant on an advertising dashboard is a useful reminder that these numbers were never the same thing as the human behaviour they claimed to describe.

We built an enormous economic system around measuring human attention without ever becoming particularly good at proving what human attention was worth.

AI makes that problem harder. It can manufacture more content, more personalities and eventually more of the activity around them, making familiar engagement metrics less useful at exactly the moment AI makes online interactions more expensive to provide.

And something curious is happening as a result. The agentic internet is beginning to put prices back onto individual interactions.

Coinbase’s x402 and Stripe’s Machine Payments Protocol are designed to let software encounter a service, discover its price and pay for it while completing a task. An AI agent can, in principle, pay a few pence for a database query, another service or a piece of specialist information without asking you to approve each transaction.

That is almost the reverse of the web I grew up with. Advertising removed the need to put a price on every click; AI may bring the meter back.

Advertising isn’t disappearing either. ChatGPT now uses advertising on its ad-supported tiers as one way of subsidising access to expensive AI infrastructure. We may end up with both systems at once: subscriptions and tiny usage charges wherever someone can charge directly, advertising wherever the user still expects somebody else to pick up the bill.

The costs become more real just as the measurements used to pay for them become less real.

The social network without society

There is a version of this future that I find encouraging.

If anybody can generate an attractive influencer, perhaps an attractive influencer becomes less valuable. If anyone can manufacture UGC, perhaps pretending an advert is an ordinary person’s recommendation stops working. If likes, views and followers become increasingly easy to generate, perhaps we finally stop confusing them with trust.

Advertisers might have to return to less exciting questions. Did somebody remember the product? Did they buy it? Did they like what they bought? Would they buy it again?

I would quite like AI to accidentally kill our obsession with engagement.

Then I look at Jean Phil.

An apparently AI-generated Frenchman appeared across social media in September, silently shadowboxing in a brown suit. Within days he had millions of views, people were making their own versions of the joke, and the attention around the character had already been attached to a meme coin.

Nobody needed to believe Jean Phil was real for any of this to work.

That complicates my hope that synthetic content will push us back towards authenticity. Perhaps authenticity isn’t necessary. Perhaps being entertaining enough to keep us looking is sufficient.

Facebook was valuable to me because my friends were there. Over time, the friends became a smaller part of the product and professionally produced content became a larger one. Influencers sat somewhere in between: strangers, but at least strangers whose lives gave people a reason to follow them.

Now we are experimenting with removing them too.

That brings me back to the question I have been circling across these posts. The Narrative Factory was about platforms shaping the stories through which we understand the world. Clip Farming the Real World was about those incentives changing what happens in the world itself. AI influencers introduce another possibility: perhaps the system no longer needs as much of the world at all.

I stopped using Facebook and Instagram when the people I knew seemed to disappear behind the content.

Six years later, we are building tools that can generate the people as well.