The dead internet theory is gaining ground as AI content increases online

Much of the internet still runs on human traffic, but it’s starting to feel less and less human.

As AI-generated messages, bots and automated agents spread across major platforms, researchers say the online world is beginning to resemble the concerns of the Dead Internet Theory, the idea that much of what people see online is no longer produced by humans, but by automated machines built to imitate them.

When the idea first circulated a few years ago on conspiracy forums like 4Chan and Agora Road’s Macintosh Café, it sounded unlikely – but the rise of generative AI has changed the way researchers view this claim.

Last year, bot activity even overtook human traffic for the first time. According to Imperva’s 2025 Bad Bot Report, a global study of automated traffic on the web, automated systems accounted for 51% of all web traffic in 2024. AI-generated articles also surpassed human-written work for the first time by the end of 2024, according to analytics firm Graphite.

“There is no direct way to measure it, but many signals indicate that the internet looks different than we think,” says Alex Turvy, a sociologist who studies how people interact on social media. Declutter.

When researchers say bots are reshaping the internet, they mean the increasing non-human traffic across the network and the growing presence of automated or AI-generated content on those platforms.

The broader concern, Turvy says, is not that there are fewer people online, but that automated activity is eroding the basic signals people use to tell who is real. When machines can mimic these signals, he says, users start to doubt everyone. Some withdraw. Others move conversations to semi-private spaces or enclosed spaces.

“Many people retreat to places like Discord or private group chats where they can be more confident about who they’re talking to,” he said. “When the usual clues no longer work, people look for other ways to know who they are talking to.”

That drift toward private channels makes the public Internet feel quieter, even though overall human activity hasn’t changed.

A February 2025 article in the Asian Journal of Research in Computer Science describes social platforms as “machine-driven ecosystems,” arguing that bots generate 40% to 60% of web traffic.

“These automated systems engage in scraping, spamming and manipulation, creating artificial interactions that mimic real human activity,” the researchers wrote. “Bots are also often used to increase metrics such as likes, shares and comments, furthering the illusion of vibrant online engagement.”

According to Turvy, the shift in momentum has become difficult to ignore.

“There are indications that this is more realistic than we thought,” Turvy said. “That’s because we see that technology is catching up, but we also see that the financial incentives are aligned.”

A September 2025 report from venture capital firm Galaxy Interactive found that automated operations now dominate major social platforms. Analysts say the increase in AI-generated material supports the trend, noting that Reddit, YouTube and X have seen an increasing amount of repetitive, low-quality or spammy content attributed to automation.

Even after Elon Musk said he would do something about the high number of bots on X, by one estimate, as many as 64% of X accounts could be bots responsible for 76% of peak traffic. Meanwhile, the same study estimated that as many as 95 million Instagram accounts – 9.5% of the total – could be fake or automated.

Meta did not immediately respond to a request for comment, and X is not responding to media inquiries.

While the number of synthetic messages continues to increase, researchers tracking the trend say the shift is already visible in the numbers.

“About half of the internet is written by AI,” Deedy Das, a partner at venture capital firm Menlo Ventures who studies the trend, told Declutter.

“Chatbots and AI tools summarize that material and give it back to you,” he added. “You end up reading machines that summarize other machines.”

While Turvy believes the growth of bots on social platforms will lead to an exodus to smaller, more intimate spaces, Das isn’t sure the ideal of the early web can return.

“There are very few people writing blogs anymore,” he says. “You can’t be discovered, and if you are, people assume it’s AI. The majority of conversations now happen on platforms built for performance, not fairness.”

Still, Das said a bigger problem is that software behaves like humans.

“The sewage system of the Internet assumed the person on the other end of the line was human,” he said. “CAPTCHAs, logins, two-factor codes, everything. Now software can imitate that perfectly, and there’s no shared rule for what counts as an agent.”

The rise of AI agents

If the Internet feels “dead” today, the proliferation of AI agents will only accelerate the trend. AI agents are autonomous programs that respond to prompts and perform tasks on the Internet on behalf of a user. They browse sites, perform searches, make purchases, trade cryptocurrencies, and interact with platforms in ways that resemble human activity.

Nirav Murthy, co-founder of intellectual property blockchain developer Camp Network, said the pattern is driven by both economics and technology.

“Agentic AI can remix material at machine speed and at virtually no cost,” he said. “Then that output comes back into circulation. Accounts look different, but behave the same. Engagement increases, variety decreases, and once you add human checks, the numbers fall apart.”

As users hand over more control to agents, more of their daily online activities will be handled by machines rather than people, deepening the automated environment they face online.

“Online ecosystems follow incentives,” he said. “When fake engagement is cheap and rewarded, you don’t just get more bots. You get production lines of automated content chasing clicks.”

That tension is already visible when used in the real world. Earlier this month, Amazon sent a cease and desist order to Perplexity after it discovered that its Comet browser was making purchases on Amazon’s site by disguising automated agents as human shoppers. Anthropic recently said it had blocked what it described as the first AI-driven cyberattack, after Chinese state-backed hackers used its Claude Code agent in attempts to penetrate 30 companies.

According to Das, the biggest risk for companies and platforms comes when AI agents are deployed in large numbers.

“When companies use a fleet of these systems, you get millions of requests acting like users,” he said. “That’s harder to see and harder to stop.”

AI-generated video is the next wave. Tools like OpenAI’s Sora 2 and Google’s Veo 3 can produce realistic clips and deepfakes of text prompts, adding to the amount of polished but synthetic content circulating on social platforms.

Both Murthy and Turvy agreed that financial incentives are driving the flood of AI bots online, and that proving personality could become a new challenge for AI. “Humanity has become just a signal to fake to make money,” Turvy said. “What’s missing now is the mess that used to prove someone was real.”

A growing number of blockchain projects, including World (formerly Worldcoin), Proof of Personhood, and Human (formerly Gitcoin) Passport, are introducing systems intended to prove personhood by linking online activity to a verified human.

“If you reward real creators and make fraud expensive, people will still have a place online,” Murthy said.