A few years ago, the "dead internet theory" was the kind of thing you brought up to sound interesting at a party and then disowned before anyone took you seriously. The claim, in its strong form, was that most of the internet was no longer made by people at all — that the posts, the comments, the accounts, the traffic were overwhelmingly bots and automated content, with real humans reduced to a shrinking minority talking past one another in a sea of machines. It was paranoid, unfalsifiable, and faintly absurd. It was also, everyone agreed, not literally true. That last reassurance is the part that has quietly stopped holding. Generative AI did not prove the dead internet theory correct so much as set about making it true, one feed at a time, and the result is the strangest and most consequential shift the social internet has undergone in a decade.
The arrival of slop
The word that emerged for it is "slop," and it is exactly the right word — content that is cheap, generic, and produced in industrial quantity with no human care behind it. Generative AI made it possible, for the first time, to manufacture text, images, and video at essentially zero marginal cost and infinite scale. What had been a bottleneck for the entire history of media — that making content required a person spending time — simply dissolved. A single operator could now generate thousands of articles, images, or videos a day, none of them touched by a human mind beyond the prompt.
The early signs were almost comic. Bizarre, uncanny AI-generated images, engineered to trip the engagement reflexes of older audiences, began racking up enormous reach on the big platforms — surreal pictures that no human would have made and that existed for no reason except to harvest clicks and reactions. Whole pages and accounts sprang up posting nothing but this material, farming engagement to be converted into advertising revenue or follower counts to be sold. It was easy to laugh at. It was also the leading edge of something that has since stopped being funny, because the slop got better, and then it got everywhere.
Why the machine loves it
To understand why slop conquered the feeds so quickly, you have to understand what the feeds are actually optimised for, and it is not truth, quality, or human connection. The dominant logic of the modern social platform is engagement: the recommendation system surfaces whatever keeps people watching, reacting, and scrolling, and it is profoundly indifferent to where that content came from. A post engineered by an algorithm to provoke a reaction is, from the platform's point of view, indistinguishable from a post written by a person who had something to say — and often it performs better, because it was built for nothing else.
This is the crucial shift that made slop inevitable. As platforms moved from showing you what the people you followed posted to showing you whatever their recommendation engine predicted would hold your attention, the social graph stopped being the filter. You no longer see mostly your friends; you see mostly whatever the machine surfaces, drawn from the entire ocean of content, the overwhelming majority produced by strangers and, increasingly, by no one at all. The feed became a slot machine of optimised stimulus, and slop is the perfect ammunition for it: cheap, endless, and tuned for exactly the reactions the system rewards. The platform earns its money whether a human or a model made the thing you stopped to look at. The incentive points one direction, and the content followed.
The accounts aren't people either
The content is only half of it. The other half is the accounts. Inauthentic and automated accounts have plagued social media for as long as it has existed, but generative AI has supercharged them past the point where the old defences work. A bot used to be detectable because it was crude — repetitive, off-topic, obviously mechanical. A bot powered by a modern language model is not. It can hold a plausible conversation, adopt a consistent persona, post varied and topical content, and respond to replies in ways that feel human enough to pass. The result is a layer of synthetic personalities woven through the real conversation, indistinguishable at a glance, operating at a scale no human troll farm could match.
This is where the dead internet theory stops sounding paranoid and starts sounding like a sober description of a measurable trend. When a meaningful and growing share of the accounts, the posts, the likes, and the replies are generated rather than authored, the basic premise of social media — that you are in a space full of other people — quietly becomes false. You are in a space full of things shaped like people, and telling the difference is getting harder by the month.
The collapse of the default assumption
The deepest damage is not any single piece of slop or any individual bot. It is the erosion of a default assumption so fundamental that we never noticed we were relying on it: that there is a person on the other end. For the entire history of the internet, you could assume, unless given reason not to, that the review was written by a customer, the photo was taken by someone, the comment came from a human who held the opinion, the account belonged to a person. That assumption was the invisible foundation of online trust, and generative AI has knocked it out.
Now the default has to flip. The reasonable starting position, increasingly, is suspicion: that the glowing review might be generated, the viral image might be fake, the passionate commenter might be a persona, the trending account might be a machine. This is corrosive in a way that is hard to overstate, because trust does not degrade gracefully. Once you cannot assume authenticity anywhere, you begin to doubt it everywhere, including the genuine human voices that are still, in fact, the majority. The slop poisons the well not only with its own falseness but by making everything around it suspect.
What it does to the people still making things
For the humans actually creating — the writers, photographers, artists, and ordinary people sharing real things — the consequences are brutal and paradoxical. On one hand, their work is being drowned. A feed that rewards volume and optimised stimulus buries the careful, the slow, and the genuine under an avalanche of generated material that can always be produced faster and cheaper. The human creator is competing, on the platform's own terms, against an opponent that never sleeps and costs nothing.
On the other hand, and this is the thread of hope, authentic human work is quietly becoming more valuable precisely because it is becoming rare and hard to verify. In an environment saturated with the synthetic, the provably human — the real person with a real history, the creator whose authenticity can be trusted — acquires a scarcity value it never had when humanity was simply assumed. The problem is that the platforms, as currently built, do almost nothing to surface that value, because their machinery cannot tell the difference and their incentives do not push them to try.
The weakness of the response
Platforms are not entirely ignoring this. There have been moves toward labelling AI-generated content, gestures toward provenance standards that would let a piece of media carry a verifiable record of how it was made, periodic purges of bot networks. But these responses share a fatal weakness: they run against the platform's own economic interest. A company that profits from engagement has little reason to aggressively remove the engaging slop that pads its numbers, and every labelling scheme is voluntary, gameable, or trivially ignored by the operators flooding the system. Detecting AI content reliably is, in any case, a losing technical race against the generators, which improve faster than the detectors.
The honest assessment is that the platforms cannot solve this within their current model, because the thing destroying the feed is the same thing that feeds their business. As long as the system rewards engagement without regard to authorship, slop will keep winning, because slop is simply engagement bait stripped of the inconvenient cost of a human.
Rebuilding an internet for people
The dead internet theory was always a metaphor — a way of naming the creeping sense that the web no longer felt human. What has changed is that the metaphor is acquiring a literal core. We are not yet living in the fully automated internet the theory imagined, but the trajectory is unmistakable, and the comfortable certainty that "of course there are real people here" no longer survives contact with the average feed.
The response taking shape is not a single fix but an instinct: a flight toward spaces where humanity can be assumed again. Smaller communities, private channels, curated and verified spaces, platforms organised around proof of personhood rather than raw engagement — the common thread is a hunger to be somewhere the other accounts are people. This is the quiet project the next era of social technology will be judged by: not how cleverly it can generate content, but whether it can rebuild, on purpose and against the economic grain, an internet where the basic promise of social media is true again — that on the other end of the conversation, there is someone actually there. The dead internet theory will be answered not by proving it wrong, but by deciding, deliberately, to keep the internet alive.
