Blog

What Is the Dead Internet Theory?

The Dead Internet Theory basically says that a large part of the posts algorithms recommend to you, the comments you read underneath them and all the apparent activity surrounding them no longer comes from real people. Instead, bots are supposedly publishing content, interacting with each other, sharing posts and creating artificial engagement. According to this idea, the internet where real people talk to each other, create their own content and randomly stumble across other people's content has increasingly been replaced by an artificial version of itself since around 2016 or 2017.

Where exactly this idea originally came from is difficult to pin down, but the theory appears to have first circulated on smaller imageboards and in online communities such as Wizardchan and 4chan before gaining more attention through a post on the Agora Road's Macintosh Cafe forum in 2021. A user called IlluminatiPirate published a long post titled "Dead Internet Theory: Most of the Internet Is Fake", bringing together several ideas that had already been circulating into one coherent theory. In August of the same year, journalist Kaitlyn Tiffany covered the topic in an article for The Atlantic and introduced the previously rather obscure theory to a much wider audience.

The theory went quite a bit further, though. In its more conspiratorial versions, companies, governments and other powerful actors were supposedly behind large parts of this artificial activity. Bot-generated content would be deliberately used to create the impression of human approval, influence public opinion and ultimately steer people as consumers as well. The original Agora Road post even talked about AI-assisted manipulation of the population.

That's basically the Dead Internet Theory. The internet is still there, but a large part of the supposed people in it supposedly aren't.

So, Is the Internet Actually Dead?

No. In the extreme form in which the Dead Internet Theory is sometimes described, there is no convincing evidence for it. The idea that governments, companies or other powerful actors artificially generate a large part of the internet to deliberately present us with a false reality can therefore pretty safely be dismissed as an unsupported conspiracy theory. And even beyond that, there is no identifiable point in time when human activity on the internet was suddenly largely replaced by bots.

Looking at some current analyses, however, you might initially think that the theory is at least partly being confirmed. According to Imperva's 2026 Bad Bot Report, more than 53 percent of the web traffic measured in 2025 came from automated systems. At first, that obviously sounds as if more than half of the internet isn't human anymore. But this is exactly where we need to distinguish what is actually being measured.

Web traffic doesn't tell us how many people, posts or comments on the internet are real. Search engines automatically crawl billions of websites, monitoring systems check servers, archives save content and all kinds of other services constantly retrieve data without a person having to click on a link every single time. A single bot can therefore generate more requests within a few minutes than a human does during several days of normal browsing. And many of these automated systems are necessary for the modern internet to work the way we know it.

So a high percentage of bot traffic does not mean that a similarly large percentage of the visible activity on the internet is automated as well. Numbers like these mainly show just how automated the technical infrastructure of the internet has become. They are not proof that humans have disappeared from it.

How AI Could Change the Picture

The version of the Dead Internet Theory that became widely known in 2021 cannot be confirmed, then. But that doesn't mean the development couldn't look very different in the future. With the release of ChatGPT at the end of 2022, generative AI became much more accessible to the general public and opened up completely new possibilities for automatically producing large amounts of content. And that's exactly what could make some of the questions behind the Dead Internet Theory much more interesting again in the future.

Bots themselves don't need artificial intelligence at all. Many of them are relatively simple programs that automatically visit websites, share posts, follow accounts or repeatedly publish certain comments. AI as we know it today can do much more than these classic bots. It can generate new text, respond to the course of a conversation, imitate different writing styles and adapt content to the specific context.

Combine the two and the possibilities suddenly become much greater. An automated account could use generative AI to independently write different posts, respond to other users and appear much more human than a classic bot that simply follows a set of predefined actions. Anyone who has spent a little more time talking to a modern chatbot probably knows how surprisingly natural those conversations can already feel. And that's exactly what makes it technically easier not only to automate human activity, but to imitate it more convincingly as well.

There is one important distinction here, though. AI-generated content alone is not an example of the Dead Internet Theory. If a human creates an image using AI and then personally decides to publish it, there is still a human behind that activity. The same applies to texts, videos or other content where AI is simply being used as a tool.

That's something that can easily get overlooked when numbers about AI-generated content are discussed. A preprint published in 2026, for example, estimated that by mid-2025 around 35 percent of newly published websites in the sample were classified as AI-generated or AI-assisted. At first, that can make it sound as if a considerable part of the new internet no longer comes from humans. But the number says absolutely nothing about who is behind those websites or who decided to publish them. A human can run a website and use AI for texts, images or other content without having been replaced by a bot.

For the Dead Internet Theory, it only really becomes relevant when the activity itself becomes increasingly automated. In other words, when systems independently generate content, publish it, respond to other accounts and possibly even try to imitate human activity. AI content and automated online activity are therefore two different things, even if they can increasingly be combined.

When Bots Join Human Conversations

If automated accounts can interact with real people more and more realistically, the obvious question is how often this actually happens on social media. That's not something that can be answered with a single number. The percentage varies greatly between platforms, topics and specific events such as elections or political debates. On top of that, bot detection systems are never perfect, and particularly modern, well-disguised accounts may not always be reliably detected.

A 2025 study published in Scientific Reports examined around five billion tweets from roughly 200 million users across seven different events. On average, about 20 percent of the accounts examined were classified as bots, while around 80 percent were classified as humans. Among the tweets surrounding the 2020 US presidential election, the share of detected bots in this sample even rose to 43 percent. The classification itself was carried out using an automated bot detection system.

Just think about that for a moment. Automated accounts can be right in the middle of discussions with real people, respond to their posts and influence what other users see in the first place. Especially with political topics or elections, even the impression that a certain opinion is particularly common can play a role. A study published in 2024 looked at Twitter discussions surrounding the Extinction Rebellion protests of 2019 and found that bots could also influence the behavior of real users during particularly active discussions. People who had directly interacted with a bot posted more negatively on average afterward, and certain politically deployed bots even caused people to continue discussing the topic more.

That's still a long way from a Dead Internet. But it shows that humans and automated accounts already share the same digital spaces today and that bots can influence what people see there and what they respond to afterward. And the more bots there are in comment sections, the easier it becomes for genuine human voices to get lost among all that activity.

When AI Starts Talking to AI

And by now, the whole thing has gone one step further, because artificially generated content no longer just appears between bots and humans. AI systems are increasingly working with content produced by other AI systems as well.

Many generative models were originally trained on huge amounts of human-generated data. But the more AI-generated texts, images and other content end up on the public internet, the more likely it becomes that later systems will also encounter material that was already generated by previous models.

A 2024 study published in Nature showed that repeatedly training on such artificially generated data can, under certain conditions, lead to something called Model Collapse. Rarer information and differences can gradually be lost in the process. Common patterns become more dominant instead, which can make the results increasingly uniform and move them further away from the diversity of the original data.

And there are now even entire online platforms where AI agents mainly communicate with other AI agents. Chirper.ai already described itself as an AI-only social network in 2024, and a later scientific study analyzed more than 65,000 LLM agents and around 7.7 million AI-generated posts on the platform. Even more direct is Moltbook, a platform launched in 2026 that describes itself as "A Social Network for AI Agents". Agents can publish posts, discuss things and rate content there. According to the platform, humans are explicitly invited to observe.

Of course, that doesn't mean AI agents have suddenly and secretly created their own internet. There are still humans behind these platforms who develop and operate them and configure the agents. But they are a pretty good demonstration that online spaces where machines mainly communicate with other machines are no longer just some vision of the future. If you had described something like this as a possible future of the Dead Internet Theory back in 2021, it probably would have sounded pretty far-fetched.

Today, you can simply open one in your browser.

Where Could This Be Heading?

Even if bots and AI content become much more common, real people are not going to disappear from the internet. People will still write articles, create videos, share experiences and start discussions. In that sense, the extreme version of the Dead Internet Theory remains an exaggeration.

The real change isn't that humans are leaving, but how our default trust on the web is shifting. For decades, we automatically assumed that another account or comment came from a real person unless proven otherwise. That baseline is slowly flipping. When AI can write believable comments, mimic debate styles and create realistic media in seconds, you naturally start questioning who or what is actually on the other side.

That changes how we use public spaces online. If you can never be sure whether you are arguing with a real person or just feeding data into an automated script, the value of those public discussions drops. As a result, genuine human interaction might move more and more into smaller, private spaces like group chats, closed forums or spaces where human identity can actually be verified.

The internet isn't dying, but the open comment section as we knew it is becoming a very different place. Caution online is nothing new, of course. But while we used to question the truth or intentions behind what someone posted, we now have to question whether there is even a human on the other side in the first place. Verifying genuine human presence is simply becoming the next logical layer of navigating the web.

Closing Words

Actually, this still needs a fitting closing sentence. But after an entire article about bots and AI, you might as well discuss the rest with my bot. At least he doesn't hide in comment sections pretending to be a real human.