In 2021, an anonymous post circulated on an online forum, claiming that most of the content we see online is generated by AI bots, and that real human users have long since become a minority: this is “Internet death theory」（Dead Internet Theory). At the time, most people dismissed it as the absurd ramblings of some conspiracy-minded netizens. However, as of early 2026, all data points to the same unsettling conclusion: this theory is, essentially, on the verge of becoming (or has already become) fact.
「Dead Internet TheoryNo Longer a Conspiracy Theory: Bot Traffic Officially Surpasses 51%
Recently, the popular science account @Rainmaker1973 (Massimo) posted a video on X. The footage shows thousands of phone motherboards stripped out and neatly arranged on dedicated metal racks, automatically generating fake likes, comments, and shares on social media platforms. These so-called “phone farms” are industrial-scale factories that manufacture fake traffic.
Dead internet is no longer just a theorypic.twitter.com/PAGsPS1PZo
— Massimo (@Rainmaker1973) February 16, 2026
Massimo wrote in a tweet: “Dead internet is no longer just a theory.” The tweet quickly sparked widespread discussion, receiving over 9,157 likes, 934 retweets, and 148 replies. What made the video so shocking was that it transformed the vague sense people had that “the internet is becoming increasingly fake” from abstract statistics into concrete, chilling imagery.
In fact, “phone farms” have long been a mature industry in China. Many so-called “traffic companies” operate tens of thousands of devices that are little more than smartphone motherboards, paired with fake accounts to boost viewership and post comments on live-streaming platforms, or hired out as “internet armies” to clash over political stances. These fake accounts have even evolved to regularly post updates and photos, creating a sense of everyday life alongside other AI accounts to evade platform detection and cleanup.
These days, AI bots like this are flooding social platforms such as Facebook, Threads, and Twitter. Many accounts that post all day across various communities, churning out content that “sounds plausible” but is actually quite hollow, are using AI to maintain fake personas. Recently, a skilled user on Threads pulled a stunt that instantly made a KOL turn into a catgirl and a puppy. (The original link has been deleted.)
Data Confirms Robots Now Outnumber Humans
Phone farm videos are just the tip of the iceberg. According to the 2024 Bad Bot Report released by Imperva, bots accounted for 51% of global web traffic, crossing the 50% threshold for the first time—just three years ago, that figure was only 42%. In other words, when you browse the web, you are statistically likely already a “minority.”
Even more alarming is the accelerating trend in bot activity. According to TollBit’s 2025Q4 ReportFor every 31 human visits, there was one AI bot visit, while at the beginning of the same year, this ratio was still 1:200. During the same period, the number of human website visitors decreased by 5%.
The situation on social media platforms is equally grim. Research shows that about 64% of accounts on X (formerly Twitter) may be bots, and 54% of long-form posts on LinkedIn are AI-generated. SEO tool provider Ahrefs analyzed 900,000 newly published webpages and found that 74.2% of them contained AI-generated content. The Copenhagen Institute for Futures Studies even predicts that by 2030, 99% of online content will be AI-generated.
The Shift in Traffic Structure: From Training Crawlers to RAG Bots
The composition of bot traffic is also undergoing a qualitative change. According to The Register’s…report Between Q2 and Q4 2025, traditional AI training crawler traffic declined by 15%. This was not because AI companies stopped collecting data, but rather because traffic shifted to new forms: RAG (Retrieval-Augmented Generation) bot traffic surged 33% during the same period, while AI search indexer traffic increased even more, by 59%.
RAG bots are the tools that AI assistants like ChatGPT and Gemini use to search the web in real time and retrieve information to answer user questions. Among them, OpenAI’s ChatGPT-User RAG bot averages five times as many crawls per page as second-place Meta, making it the most aggressive “reader” on the web—except that after reading, it doesn’t click through to your website; it simply feeds the answer directly to users. This is changing how people search. Data shows that 37% of active AI users now start their searches from ChatGPT or Gemini rather than Google. Search engines are no longer the gateway to websites; they are being replaced by AI intermediaries. (That’s why Google immediately raised a “red alert” when ChatGPT first burst onto the scene.)
The Survival Crisis of the Media and Publishing Industries
This seismic shift in traffic structure is nothing short of a catastrophe for the publishing and media industries that depend on web traffic for their survival. Google referral traffic to publishers and news sites fell 33% globally between November 2024 and November 2025, and dropped even further, 38%, in the US. Even when users receive source links through AI apps, actual click-through rates have plummeted from 0.8% to 0.27%—after all, AI has already delivered the answer, so why would anyone still need to click through to the original article?
The actual figures are startling: independent music media outlet Stereogum lost 70% of its advertising revenue in 2025; Business Insider’s organic search traffic dropped 55%, forcing it to lay off 21% of staff; edtech company Chegg’s non-subscription traffic fell 49% year over year. (Editor’s note: The Computer King Ada website also saw its Google search traffic fall by more than 30%.)
“Zero-click search” lets users get answers directly on the search results page without clicking any links, and this share jumped from 56% to 69%. When the Google AI Overview feature rolled out, search result click-through rates were cut in half, plummeting from 15% to 8%. Gartner predicts that overall search engine usage will drop another 25% by the end of 2026.
AI Reaps What It Sows: Model Collapse
However, bots outnumbering humans is not the deepest crisis. The real concern lies in a vicious cycle taking shape: AI-generated content is now contaminating the very training data that AI depends on. It is currently estimated that 30% to 40% of active web content is synthetically generated. As AI models increasingly use this AI-generated content to train the next generation of models, researchers have identified a phenomenon known as “model collapse”: the statistical biases of each generation accumulate and amplify, much like making repeated photocopies of a document, each copy becoming blurrier and more distorted than the last.
Some researchers have dubbed 2024 to 2025 the “Slop Era”—a period during which a deluge of low-quality AI-generated content flooding the internet has contaminated the training data of all models currently under development. More worrying still, datasets considered “clean” before 2023 have already been irreversibly infiltrated by synthetic content. AI is eating its own tail.
Humanity in Retreat: The End of the Open Internet?
Faced with an open web inundated with bots and synthetic content, humans are doing something intriguing: retreating.
Paywalls and closed communities are on the rise. Substack is now valued at $1.1 billion, with more than 5 million paid subscribers. More and more people are turning to gated spaces like private Discord servers, paid newsletters, and corporate intranets—because that’s where they’re more likely to encounter real humans.
Linguist Adam Aleksic notes: “The Dead Internet Theory used to be a fringe conspiracy theory, but it’s starting to look more and more real.” Tech journalist Taylor Lorenz’s observation is even sharper: “Before ChatGPT was released, the internet was already terminally ill.” AI has only accelerated an existing trend: when platform algorithms prioritize content that maximizes engagement, authenticity has never been the main concern.
perspective
The most ironic aspect of the Dead Internet Theory is that humans and vested interests have actively turned themselves into the minority. Every automation tool, every efficiency-driven decision, and every choice to use AI-generated content to save time has quietly pushed authentic human voices to the margins. The “efficiency” people chase ultimately becomes the flood that drowns them.
But compared to bot traffic surpassing human traffic, the deeper crisis is the self-cannibalizing cycle in which AI training data is polluted by AI-generated content. When AI keeps learning from its own output, it isn’t getting smarter—it is accelerating its own degradation. This is not merely a technical issue; it is an existential battle over the quality of knowledge. For ordinary users, there may be limited options for now, but that does not mean they are powerless: cultivate the ability to recognize AI-generated content, remain skeptical of anything that looks too perfect or formulaic, and support trustworthy independent media (e.g.:電腦王阿達On an internet dominated by bots, choosing what to read and what to trust is itself an act of resistance. As for media professionals like us, frankly, there isn’t much we can do in the face of AI and Google except try to survive.
在 Threads 查看
Source: KOCPC Chinese