The Dead Internet Theory Is Becoming Less Theory and More Reality

="">

The internet feels off. Not broken exactly, but hollow — like walking through a shopping mall where half the storefronts are staffed by mannequins. Adrian Krebs, a Swiss software engineer and founder, recently published a detailed examination of what’s been called the Dead Internet Theory, and his argument is uncomfortably persuasive: the web as we knew it is being drowned in AI-generated content, and we’re already past the tipping point.

The Dead Internet Theory originated on fringe forums around 2021, suggesting that most online content and interactions are generated by bots and algorithms rather than real humans. It was easy to dismiss then. Conspiracy thinking. Paranoid speculation. But Krebs marshals enough evidence to suggest the core thesis — even if not the conspiratorial framing — has essentially come true.

Here’s the math. Krebs cites research from Similarweb and other analytics sources indicating that bot traffic now accounts for roughly half of all web activity. Some estimates put it higher. Imperva’s 2024 Bad Bot Report found that automated traffic surpassed human traffic for the first time in a decade, with bad bots alone representing 32% of all internet traffic. That’s not including the “good” bots from search engines and monitoring tools. Add AI-generated content farms, synthetic social media profiles, and automated comment systems, and the picture gets bleak fast.

Krebs traces how the economics of content creation shifted. Producing thousands of AI-written articles costs almost nothing. A single operator with access to GPT-4 or Claude can generate more text in a day than a newsroom produces in a month. The incentive structure is obvious: ad revenue scales with pageviews, and AI slop is cheap to produce. So the web fills up with it. Google’s search results, once the internet’s reliable front door, now surface AI-generated summaries that themselves pull from AI-generated sources. A feedback loop with no human in it.

Social media is worse. Krebs points to investigations showing that platforms like X (formerly Twitter) are saturated with bot accounts pushing AI-generated content. A Washington Post analysis in early 2024 found bot networks on X generating millions of posts per day, many indistinguishable from human output. Meta’s platforms face similar problems — Facebook groups flooded with AI-generated images of Jesus holding shrimp sculptures, designed to farm engagement from unsuspecting users. Absurd. But profitable.

The implications for trust are severe.

Krebs argues that we’re entering what he calls a “post-authentic” internet, where the default assumption should be that any given piece of content was machine-generated unless proven otherwise. This isn’t hyperbole. Amazon has been battling AI-generated book spam flooding Kindle Unlimited. Academic publishers are catching papers with telltale ChatGPT phrases like “as of my last training data.” Product reviews, restaurant recommendations, LinkedIn posts — all increasingly synthetic.

And the detection tools aren’t keeping up. Krebs notes that AI text detectors remain unreliable, with high false-positive rates that punish human writers while letting sophisticated AI output through. OpenAI itself shelved its AI text classifier in 2023 due to poor accuracy. The arms race between generation and detection favors the generators, and it’s not particularly close.

What does this mean for professionals? Several things. SEO strategies built on content volume are now competing against infinite AI output — a race to the bottom that rewards nobody except platform intermediaries. Brand trust becomes harder to establish when your authentic content sits alongside thousands of AI knockoffs. And if you’re relying on web data for market research, sentiment analysis, or competitive intelligence, your inputs are increasingly contaminated.

Krebs makes a subtler point too. The cultural texture of the internet is changing. The weird, personal, idiosyncratic web — people blogging about niche hobbies, writing bad poetry, sharing genuine experiences — gets buried under algorithmic optimization. Communities that once formed organically around shared interests now compete for attention with synthetic engagement. The human web doesn’t disappear. It just becomes invisible.

Some companies are responding. Reddit has become more valuable precisely because its content is (mostly) human-generated, which is why Google signed a $60 million deal to access its data for AI training, as reported by Reuters. Stack Overflow, Wikipedia, and other human-curated platforms are renegotiating their relationship with AI companies that scraped their content to build the very tools now flooding the web with synthetic replacements. There’s an irony there that borders on tragic.

Krebs also flags the political dimension. Bot-driven influence operations aren’t new, but generative AI makes them dramatically more effective and cheaper to run. The barrier to manufacturing consensus — or at least the appearance of it — has collapsed. Governments and bad actors can simulate grassroots movements at scale. During election cycles, this isn’t theoretical. It’s operational.

So where does this leave us? Krebs doesn’t offer easy solutions, and that honesty is refreshing. He suggests that verification systems, cryptographic content provenance (like the C2PA standard being developed by Adobe, Microsoft, and others), and a cultural shift toward smaller, authenticated communities might help. But he’s realistic about the scale of the problem. The economic incentives pushing toward synthetic content are enormous, and they’re accelerating.

The dead internet isn’t a future scenario. It’s a present condition that most people haven’t fully registered yet. The web still works — you can still find real information, connect with real people, read real writing. But the ratio is shifting, steadily and measurably, toward the artificial. For anyone building products, publishing content, or making decisions based on online data, ignoring this shift is no longer an option.

Not because the internet is literally dead. But because the living parts are getting harder to find.